xproc 2010-03-09 XML Revision markup Norman Walsh Mark Logic Corporation norman.walsh@marklogic.com Alex Milowski Invited expert alex@milowski.org Henry S. Thompson University of Edinburgh ht@inf.ed.ac.uk This specification describes the syntax and semantics of XProc: An XML Pipeline Language, a language for describing operations to be performed on XML documents. An XML Pipeline specifies a sequence of operations to be performed on zero or more XML documents. Pipelines generally accept zero or more XML documents as input and produce zero or more XML documents as output. Pipelines are made up of simple steps which perform atomic operations on XML documents and constructs similar to conditionals, iteration, and exception handlers which control which steps are executed. This section describes the status of this document at the time of its publication. Other documents may supersede this document. A list of current W3C publications and the latest revision of this technical report can be found in the W3C technical reports index at http://www.w3.org/TR/. This is a Proposed Recommendation for review by W3C members and other interested parties. This document is a product of the XML Processing Model Working Group which is part of the W3C XML Activity. The English version of this specification is the only normative version. However, for translations of this document, see . Though the previous version of this document was a Last Call Working Draft, there was an earlier Candidate Recommendation version of this document that has already resulting in successful implementation feedback. Given that the changes to this draft do not affect the validity of that earlier implementation feedback, except in specific areas also now covered by more recent implementation feedback, the Working Group is now publishing this version as a Proposed Recommendation. There is an Implementation Report for XProc. W3C Advisory Committee Representatives are invited to submit their formal review per the instructions in the Call for Review (see Advisory Committee questionnaires). The review period ends on 15 April 2010. Members of the public are also invited to send comments on this Proposed Recommendation to the public mailing list public-xml-processing-model-comments@w3.org (public archives are available). This specification implements all of the requirements documented in . Publication as a Proposed Recommendation does not imply endorsement by the W3C Membership. This is a draft document and may be updated, replaced or obsoleted by other documents at any time. It is inappropriate to cite this document as other than work in progress. This document was produced by a group operating under the 5 February 2004 W3C Patent Policy. W3C maintains a public list of any patent disclosures made in connection with the deliverables of the group; that page also includes instructions for disclosing a patent. An individual who has actual knowledge of a patent which the individual believes contains Essential Claim(s) must disclose the information in accordance with section 6 of the W3C Patent Policy.

An XML Pipeline specifies a sequence of operations to be performed on a collection of XML input documents. Pipelines take zero or more XML documents as their input and produce zero or more XML documents as their output. A pipeline consists of steps. Like pipelines, steps take zero or more XML documents as their inputs and produce zero or more XML documents as their outputs. The inputs of a step come from the web, from the pipeline document, from the inputs to the pipeline itself, or from the outputs of other steps in the pipeline. The outputs from a step are consumed by other steps, are outputs of the pipeline as a whole, or are discarded. There are three kinds of steps: atomic steps, compound steps, and multi-container steps. Atomic steps carry out single operations and have no substructure as far as the pipeline is concerned. Compound steps and multi-container steps control the execution of other steps, which they include in the form of one or more subpipelines. This specification defines a standard library, , of steps. Pipeline implementations may support additional types of steps as well. is a graphical representation of a simple pipeline that performs XInclude processing and validation on a document.

A simple, linear XInclude/Validate pipeline

This is a pipeline that consists of two atomic steps, XInclude and Validate with XML Schema. The pipeline itself has two inputs, “source” (a source document) and “schemas” (a sequence of W3C XML Schemas). The XInclude step reads the pipeline input “source” and produces a result document. The Validate with XML Schema step reads the pipeline input “schemas” and the result of the XInclude step and produces its own result document. The result of the validation, “result”, is the result of the pipeline. (For consistency across the step vocabulary, the standard input is usually named “source” and and the standard output is usually named “result”.) The pipeline document determines how the steps are connected together inside the pipeline, that is, how the output of one step becomes the input of another. The pipeline document for this pipeline is shown in . The example in is very verbose. It makes all of the connections seen in the figure explicit. In practice, pipelines do not have to be this verbose. XProc supports defaults for many common cases: If you use p:pipeline instead of p:declare-step, the “source” input port and “result” output port are implicitly declared for you. Where inputs and outputs are connected between sequential sibling steps, they do not have to be made explicit. The same pipeline, using XProc defaults, is shown in . is a more complex example: it performs schema validation with an appropriate schema and then styles the validated document.

A validate and transform pipeline

The heart of this example is the conditional. The “choose” step evaluates an XPath expression over a test document. Based on the result of that expression, one or another branch is run. In this example, each branch consists of a single validate step. This example, like the preceding, relies on XProc defaults for simplicity. It is always valid to write the fully explicit form if you prefer. The media type for pipeline documents is application/xml. Often, pipeline documents are identified by the extension .xpl. In this specification the words must, must not, should, should not, may and recommended are to be interpreted as described in .

A pipeline is a set of connected steps, with outputs of one step flowing into inputs of another. A pipeline is itself a step and must satisfy the constraints on steps. Connections between steps occur where the input of one step is connected to the output of another. The result of evaluating a pipeline (or subpipeline) is the result of evaluating the steps that it contains, in an order consistent with the connections between them. A pipeline must behave as if it evaluated each step each time it is encountered. Unless otherwise indicated, implementations must not assume that steps are functional (that is, that their outputs depend only on their inputs, options, and parameters) or side-effect free. The pattern of connections between steps will not always completely determine their order of evaluation. The evaluation order of steps not connected to one another is implementation-dependent.

A step is the basic computational unit of a pipeline. A typical step has zero or more inputs, from which it receives XML documents to process, zero or more outputs, to which it sends XML document results, and can have options and/or parameters. There are three kinds of steps: atomic, compound, and multi-container. An atomic step is a step that performs a unit of XML processing, such as XInclude or transformation, and has no internal subpipeline. Atomic steps carry out fundamental XML operations and can perform arbitrary amounts of computation, but they are indivisible. An XSLT step, for example, performs XSLT processing; a Validate with XML Schema step validates one input with respect to some set of XML Schemas, etc. There are many types of atomic steps. The standard library of atomic steps is described in , but implementations may provide others as well. It is implementation-defined what additional step types, if any, are provided. Each use, or instance, of an atomic step invokes the processing defined by that type of step. A pipeline may contain instances of many types of steps and many instances of the same type of step. Compound steps, on the other hand, control and organize the flow of documents through a pipeline, reconstructing familiar programming language functionality such as conditionals, iterators and exception handling. They contain other steps, whose evaluation they control. A compound step is a step that contains a subpipeline. That is, a compound step differs from an atomic step in that its semantics are at least partially determined by the steps that it contains. Finally, there are two “multi-container steps”: p:choose and p:try. A multi-container step is a step that contains several alternate subpipelines. Each subpipeline is identified by a non-step wrapper element: p:when and p:otherwise in the case of p:choose, p:group and p:catch in the case of p:try. The output of a multi-container step is the output of exactly one of its subpipelines. In this sense, a multi-container step functions like a compound step. However, evaluating a multi-container step may involve evaluating, or partially evaluating, more than one of its subpipelines. It's possible for steps in a partially evaluated pipeline to have side effects that are visible outside the processor, even if the final output of the multi-container step is the result of some other subpipeline. For example, a web server might record that some interaction was performed, or a file on the local file system might have been modified. A compound step or multi-container step is a container for the steps directly within it or within non-step wrappers directly within it. The steps that occur directly within, or within non-step wrappers directly within, a step are called that step's contained steps. In other words, “container” and “contained steps” are inverse relationships. The ancestors of a step, if it has any, are its container and the ancestors of its container. Sibling steps (and the connections between them) form a subpipeline. The last step in a subpipeline is its last step in document order. subpipeline = p:variable*, (p:for-each|p:viewport|p:choose|p:group|p:try|p:standard-step|pfx:user-pipeline)+ User-defined pipelines (identified with pfx:user-pipeline in the preceding syntax summary) are atomic. A pipeline declaration may contain a subpipeline, but the invocation of that pipeline is atomic and does not contain a subpipeline. Steps have “ports” into which inputs and outputs are connected. Each step has a number of input ports and a number of output ports; a step can have zero input ports and/or zero output ports. (All steps have an implicit output port for reporting errors that must not be declared.) The names of all ports on each step must be unique on that step (you can't have two input ports named “source”, nor can you have an input port named “schema” and an output port named “schema”). A Step may have zero or more options, all with unique names. Steps may have parameter input ports, on which parameters can be passed. A step can have zero, one, or many parameter input ports, and each parameter input port can have zero or more parameters passed on it. If more than one parameter with the same name is passed to any given parameter input port, only the last value specified will be available to the step; the names of the parameters passed to the step are unique on each port. Parameters with the same name can be passed to different ports, the uniqueness constraint on names only applies to the parameters passed on each individual port. All of the different instances of steps (atomic or compound) in a pipeline can be distinguished from one another by name. If the pipeline author does not provide a name for a step, a default name is manufactured automatically.

The name attribute on any step can be used to give it a name. The name must be unique within its scope, see . If the pipeline author does not provide an explicit name, the processor manufactures a default name. All default names are of the form “!1.m.n…” where “m” is the position (in the sense of counting sibling elements) of the step's highest ancestor element within the pipeline document or library which contains it, “n” is the position of the next-highest ancestor, and so on, including both steps and non-step wrappers. For example, consider the pipeline in . The p:pipeline step has no name, so it gets the default name “!1”; the p:choose gets the name “!1.1”; the first p:when gets the name “!1.1.1”; the p:otherwise gets the name “!1.1.2”, etc. If the p:choose had had a name, it would not have received a default name, but it would still have been counted and its first p:when would still have been “!1.1.1”. Providing every step in the pipeline with an interoperable name has several benefits: It allows implementors to refer to all steps in an interoperable fashion, for example, in error messages. Pragmatically, we say that readable ports are identified by a step name/port name pair. By manufacturing names for otherwise anonymous steps, we include implicit connections without changing our model. In a valid pipeline that runs successfully to completion, the manufactured names aren't visible (except perhaps in debugging or logging output). The format for defaulted names does not conform to the requirements of an NCName. This is an explicit design decision; it prevents pipelines from using the defaulted names on p:pipe elements. If an explicit connection is required, the pipeline author must provide an explicit name for the step.

Although some steps can read and write non-XML resources, what flows between steps through input ports and output ports are exclusively XML documents or sequences of XML documents. For the purposes of this specification, an XML document is an . Implementations are free to transmit Infosets as sequences of characters, sequences of events, object models, or any other representation that preserves the necessary Infoset properties (see ). Most steps in this specification manipulate XML documents, or portions of XML documents. In these cases, we speak of changing elements, attributes, or nodes without prejudice to the actual representation used by an implementation. An implementation may make it possible for a step to produce non-XML output (through channels other than a named output port)—for example, writing a PDF document to a URI—but that output cannot flow through the pipeline. Similarly, one can imagine a step that takes no pipeline inputs, reads a non-XML file from a URI, and produces an XML output. But the non-XML data cannot arrive on an input port to a step. It is a dynamic error if a non-XML resource is produced on a step output or arrives on a step input. The common case is that each step has one or more inputs and one or more outputs. illustrates symbolically an atomic step with two inputs and one output.

An atomic step with two inputs and one output

All atomic steps are defined by a p:declare-step. The declaration of an atomic step type defines the input ports, output ports, and options of all steps of that type. For example, every p:validate-with-xml-schema step has two inputs, named “source” and “schema”, one output named “result”, and the same set of options. Like atomic steps, top level, user-defined pipelines also have declarations. The situation is slightly more complicated for the other compound steps because they don't have separate declarations; each instance of the compound step serves as its own declaration. On these compound steps, the number and names of the outputs can be different on each instance of the step. illustrates symbolically a compound step with one subpipeline and one output. As you can see from the diagram, the output from the compound step comes from one of the outputs of the subpipeline within the step.

A compound step with two inputs and one output

The input ports declared on a step are its declared inputs. The output ports declared on a step are its declared outputs. When a step is used in a pipeline, it is connected to other steps through its inputs and outputs. When a step is used, all of the declared inputs of the step must be connected. Each input can be connected to: The output port of some other step. A fixed, inline document or sequence of documents. A document read from a URI. One of the inputs declared on one of its ancestors. A special port provided by an ancestor compound step, for example, “current” in a p:for-each or p:viewport. When an input accepts a sequence of documents, the documents can come from any combination of these locations. It is a static error if any declared input is not connected. The declared outputs of a step may be connected to: The input port of some other step (including p:iteration-source or p:viewport-source). The p:xpath-context of a p:choose or p:when. An option assigned with p:with-option or a parameter assigned with p:with-param. One of the outputs declared on its container. The primary output port of a step must be connected. It is a static error if the primary output port of any step is not connected. Other outputs can remain unconnected. Any documents produced on an unconnected output port are discarded. Primary input and primary output ports may be implicitly connected if no explicit connection is given, see . Output ports on compound steps have a dual nature: from the perspective of the compound step's siblings, its outputs are just ordinary outputs and must be connected as described above. From the perspective of the subpipeline inside the compound step, they are inputs into which something may be connected. Within a compound step, the declared outputs of the step can be connected to: The output port of some contained step. A fixed, inline document or sequence of documents. A document read from a URI. If a (non-primary) output port of a compound step is left unconnected, it produces an empty sequence of documents from the perspective of its siblings. Each input and output on a step is declared to accept or produce either a single document or a sequence of documents. It is not an error to connect a port that is declared to produce a sequence of documents to a port that is declared to accept only a single document. It is, however, an error if the former step actually produces more than one document at run time. It is also not an error to connect a port that is declared to produce a single document to a port that is declared to accept a sequence. A single document is the same as a sequence of one document. An output port may have more than one connection: it may be connected to more than one input port, more than one of its container's output ports, or both. At runtime this will result in distinct copies of the output. The signature of a step is the set of inputs, outputs, and options that it is declared to accept. The declaration for a step provides a fixed signature which all its instances share. A step matches its signature if and only if it specifies an input for each declared input, it specifies no inputs that are not declared, it specifies an option for each option that is declared to be required, and it specifies no options that are not declared. In other words, every input and required option must be specified and only inputs and options that are declared may be specified. Options that aren't required do not have to be specified. Steps may also produce error, warning, and informative messages. These messages are captured and provided on the error port inside of a p:catch. Outside of a try/catch, the disposition of error messages is implementation-dependent. How inputs are connected to XML documents outside the pipeline is implementation-defined. In order to be consistent with the XPath data model, all general and external parsed entities in such documents must be fully expanded; they must not contain any representation of unexpanded entity reference information items. How pipeline outputs are connected to XML documents outside the pipeline is implementation-defined.

It's common for some of the documents used in processing a pipeline to be read from URIs. Sometimes this occurs directly, for example with a p:document element. Sometimes it occurs indirectly, for example if an implementation allows the URI of a pipeline input to be specified on the command line or if an p:xslt step encounters an xsl:import in the stylesheet that it is processing. It's also common for some of the documents produced in processing a pipeline to be written to locations which have, or at least could have, a URI. The process of dereferencing a URI to retrieve a document is often more interesting than it seems at first. On the web, it may involve caches, proxies, and various forms of indirection. Resolving a URI locally may involve resolvers of various sorts and possibly appeal to implementation-dependent mechanisms such as catalog files. In XProc, the situation is made even more interesting by the fact that many intermediate results produced by steps in the pipeline have base URIs. Whether (and when and how) or not the intermediate results that pass between steps are ever written to a filesystem is implementation-dependent. In Version 1.0 of XProc, how (or if) implementers provide local resolution mechanisms and how (or if) they provide access to intermediate results by URI is implementation-defined. Version 1.0 of XProc does not require implementations to guarantee that multiple attempts to dereference the same URI always produce consistent results. On the one hand, this is a somewhat unsatisfying state of affairs because it leaves room for interoperability problems. On the other, it is not expected to cause such problems very often in practice. If these problems arise in practice, implementers are encouraged to use the existing extension mechanisms to give users the control needed to circumvent them. Should such mechanisms become widespread, a standard mechanism could be added in some future version of the language.

XProc is designed to allow pipeline authors to specify how an XML document, or sequence of XML documents, flows through a series of steps. For the most part, non-XML documents are considered out-of-scope. However, to be useful, XProc pipelines must interact with the real world where non-XML documents (HTML documents, raster images, non-XML encodings of data, etc.) are a fact of life. Accordingly, some pipelines may need to access non-XML documents and some non-XML documents may “leak” into pipelines. XProc provides a limited set of tools for processing these documents. In particular, XProc offers the ability to turn some “almost-XML” documents into XML and to allow some non-XML documents to flow quietly through the pipeline. It is not a goal of XProc that it should be a general-purpose pipeline language for manipulating arbitrary, non-XML resources. There are two standard ways that a non-XML document may enter a pipeline: directly through p:data or as the result of performing an p:http-request step. Loading non-XML data with a computed URI requires the p:http-request step. Implementors are encouraged to support the file: URI scheme so that users can load local data from computed URIs. In either case, non-XML documents are converted into text or are base64-encoded, depending on their content type and character encoding. The result is an XML document that consists of a document element containing either escaped text or base64-encoded text. This document can be processed like any other XML document. The p:unescape-markup step can be used to (attempt to) convert a non-XML document into XML. Well-formed XML that just happens to be represented with escaped markup can always be recovered. For other media types, the ability to construct XML and the precise mechanisms used to make the markup well-formed are implementation-defined. XProc provides no standard means to save encoded data in its unencoded binary form. Implementors may provide extension methods to allow the p:store step to save the binary data. For example, an implementation might provide a ext:binary serialization method that decoded base64 encoded data before saving it: BASE64ENCODEDDATA ]]>

As a convenience for pipeline authors, each step may have one input port designated as the primary input port and one output port designated as the primary output port. If a step has a document input port which is explicitly marked “primary='true'”, or if it has exactly one document input port and that port is not explicitly marked “primary='false'”, then that input port is the primary input port of the step. If a step has a single input port and that port is explicitly marked “primary='false'”, or if a step has more than one input port and none is explicitly marked as the primary, then the primary input port of that step is undefined. A step can have at most one primary input port. If a step has a document output port which is explicitly marked “primary='true'”, or if it has exactly one document output port and that port is not explicitly marked “primary='false'”, then that output port is the primary output port of the step. If a step has a single output port and that port is explicitly marked “primary='false'”, or if a step has more than one output port and none is explicitly marked as the primary, then the primary output port of that step is undefined. A step can have at most one primary output port. The special significance of primary input and output ports is that they are connected automatically by the processor if no explicit connection is given. Generally speaking, if two steps appear sequentially in a subpipeline, then the primary output of the first step will automatically be connected to the primary input of the second. Additionally, if a compound step has no declared outputs and the last step in its subpipeline has an unconnected primary output, then an implicit primary output port will be added to the compound step (and consequently the last step's primary output will be connected to it). This implicit output port has no name. It inherits the sequence property of the port connected to it. This rule does not apply to p:declare-step; step declarations must provide explicit names for all of their outputs.

Steps are connected together by their input ports and output ports. It is a static error if there are any loops in the connections between steps: no step can be connected to itself nor can there be any sequence of connections through other steps that leads back to itself.

XProc processors are expected, and sometimes required, to perform namespace fixup. Unless the semantics of a step explicitly says otherwise: The in-scope namespaces associated with a node (even those that are inherited from namespace bindings that appear among its ancestors in the document in which it appears initially) are assumed to travel with that node. Changes to one part of a tree (wrapping or unwrapping a node or renaming an element, for example) do not change the in-scope namespaces associated with the descendants of the node so changed. As a result, some steps can produce XML documents which have no direct serialization (because they include nodes with conflicting or missing namespace declarations, for example). To produce a serializable XML document, the XProc processor must sometimes add additional namespace nodes, perhaps even renaming prefixes, to satisfy the constraints of Namespaces in XML. This process is referred to as namespace fixup. Implementors are encouraged to perform namespace fixup before passing documents between steps, but they are not required to do so. Conversely, an implementation which does serialize between steps and therefore must perform such fixups, or reject documents that cannot be serialized, is also conformant. Except where the semantics of a step explicitly require changes, processors are required to preserve the information in the documents and fragments they manipulate. In particular, the information corresponding to the properties attributes, base URI, children, local name, namespace name, normalized value, owner, and parent must be preserved. The information corresponding to prefix, in-scope namespaces, namespace attributes, and attribute type should be preserved, with changes to the first three only as required for namespace fixup. In particular, processors are encouraged to take account of prefix information in creating new namespace bindings, to minimize negative impact on prefixed names in content. Except for cases which are specifically called out in , the extent to which namespace fixup, and other checks for outputs which cannot be serialized, are performed on intermediate outputs is implementation-defined. Whenever an implementation serializes pipeline contents, for example for pipeline outputs, logging, or as part of steps such as p:store or p:http-request, it is a dynamic error if that serialization could not be done so as to produce a document which is both well-formed and namespace-well-formed, as specified in XML and Namespaces in XML, regardless of what serialization method, if any, is called for.

The environment is a context-dependent collection of information available within subpipelines. Most of the information in the environment is static and can be computed for each subpipeline before evaluation of the pipeline as a whole begins. The in-scope bindings have to be calculated as the pipeline is being evaluated. The environment consists of: A set of readable ports. The readable ports are a set of step name/port name pairs. Inputs and outputs can only be connected to readable ports. A default readable port. The default readable port, which may be undefined, is a specific step name/port name pair from the set of readable ports. A set of in-scope bindings. The in-scope bindings are a set of name-value pairs, based on option and variable bindings. The empty environment contains no readable ports, an undefined default readable port and no in-scope bindings. Unless otherwise specified, the environment of a contained step is its inherited environment. The inherited environment of a contained step is an environment that is the same as the environment of its container with the standard modifications. The standard modifications made to an inherited environment are: The declared inputs of the container are added to the readable ports. In other words, contained steps can see the inputs to their container. The union of all the declared outputs of all of the step's sibling steps are added to the readable ports. In other words, sibling steps can see each other's outputs in addition to the outputs visible to their container. If there is a preceding sibling step element: If that preceding sibling has a primary output port, then that output port becomes the default readable port. Otherwise, the default readable port is undefined. If there is not a preceding sibling step element: If the container has a primary input port, the default readable port is that primary input port. Otherwise, the default readable port is unchanged. The names and values from each p:variable present at the beginning of the container are added, in document order, to the in-scope bindings. A new binding replaces an old binding with the same name. See for the specification of variable evaluation. A step with no parent inherits the empty environment.

XProc uses XPath as an expression language. XPath expressions are evaluated by the XProc processor in several places: on compound steps, to compute the default values of options and the values of variables; on atomic steps, to compute the actual values of options and the values of parameters. XPath expressions are also passed to some steps. These expressions are evaluated by the implementations of the individual steps. This distinction can be seen in the following example: The select expression on the variable “home” is evaluated by the XProc processor. The value of the variable is “http://example.com/docs”. The href option of the p:load step is evaluated by the XProc processor. The actual href option received by the step is simply the string literal “http://example.com/docs/document.xml”. (The select expression on the source input of the p:split-sequence step is also evaluated by the XProc processor.) The XPath expression “@role='chapter'” is passed literally to the test option on the p:split-sequence step. That's because the nature of the p:split-sequence is that it evaluates the expression. Only some options on some steps expect XPath expressions. The XProc processor evaluates all of the XPath expressions in select attributes on variables, options, parameters, and inputs, in match attributes on p:viewport, and in test attributes on p:when steps. An XProc implementation can use either or to evaluate these expressions. Allowing either XPath 1.0 or XPath 2.0 is a compromise driven entirely by the timing of XProc development. During the development of this specification, the community indicated that it was too early to mandate that all implementations use XPath 2.0 and too late to mandate that all implementations use XPath 1.0. Many, many expressions that are likely to be used in XProc pipelines are the same in both versions (simple element tests, ancestor and descendant tests, string-based attribute tests, etc.). As an aid to interoperability, pipeline authors may indicate the version of XPath that they require. The attribute xpath-version may be used on p:pipeline, p:declare-step, or p:library to identify the XPath version that must be used to evaluate XPath expressions on the pipeline(s). The attribute is lexically scoped, but see below. In Version 1.0 of XProc, no similar level of control is provided for specifying (or testing) the version of XSLT used when evaluating XSLTMatchPatterns. The expectation is that XPath 1.0 processors will be using XSLT 1.0 match patterns and XPath 2.0 processors will be using XSLT 2.0 match patterns, but that is not necessarily the case. As XPath, XSLT, and XProc continue to evolve, additional facilities for specifying and testing the version of XSLT used to evaluate match patterns may be added to XProc. If an xpath-version is specified on a p:pipeline or p:declare-step, then that is the version of XPath that the step requires. If it does not specify a version, but a version is specified on one of its ancestors, the nearest ancestor version specified is the version that it requires. An xpath-version attribute on a p:library specifies a default version for all steps defined in that library. If no version is specified on the step or among its ancestors, then its XPath version is implementation-defined. The decision about which XPath version applies can be made dynamically. For example, if a pipeline explicitly labeled with xpath-version “1.0” imports a library that does not specify a version, the implementation may elect to make the implementation-defined XPath version of the steps in the library also “1.0”. If the same implementation imports that library into a pipeline explicitly labeled with xpath-version “2.0”, it can make the implementation-defined version of those steps “2.0”. The following rules determine how the indicated version and the implementation's actual version interact: If the indicated version and the implementation version are the same, then that version is used. If the indicated version is 1.0 and the implementation uses XPath 2.0 (or later), the expression must be evaluated in XPath 1.0 compatibility mode. It is a dynamic error if a 2.0 processor encounters an XPath 1.0 expression and it does not support XPath 1.0 compatibility mode. Otherwise: It is a dynamic error if the processor encounters an xpath-version that it does not support. XProc processors divide naturally into two classes: XPath 1.0 processors and XPath 2.0 processors. Irrespective of which version of XPath is used, all expressions evaluated by XProc or passed to steps for evaluation must be valid XPath expressions. It is a dynamic error if an XPath expression is encountered which cannot be evaluated (because it is syntactically incorrect, contains references to unbound variables or unknown functions, or for any other reason).

XProc processors that support only XPath 1.0 do not support any types (or features) beyond those described in . They use the XPath 1.0 data model. Processors must implement all of the XPath 1.0 functions, but are not expected to implement the functions described in .

When the XProc processor evaluates an XPath expression using XPath 1.0, unless otherwise indicated by a particular step, it does so with the following initial context: context node The document node of a document. The document is either specified with a connection or is taken from the default readable port. It is a dynamic error if a document sequence appears where a document to be used as the context node is expected. If there is no explicit connection and there is no default readable port then the context node is an empty document node. context position and context size The context position and context size are both “1”. variable bindings The union of the in-scope specified options and variables are available as variable bindings to the XPath processor. An option that has neither a specified value nor a default value will not appear as an in-scope variable. Consequently, an attempt to refer to that variable will raise an error. function library The core function library and the . Function names that do not contain a colon always refer to the XPath 1.0 functions, any in-scope binding for the default namespace does not apply. in-scope namespaces The namespace bindings in-scope on the element where the expression occurred.

When a step evaluates an XPath expression using XPath 1.0, unless otherwise indicated by a particular step, it does so with the following initial context: context node The document node that appears on the primary input port of the step, unless otherwise specified by the step. context position and context size The position and size are both “1”, unless otherwise specified by the step. variable bindings None, unless otherwise specified by the step. function library The core function library, unless otherwise specified by the step. Function names that do not contain a colon always refer to the XPath 1.0 functions, any in-scope binding for the default namespace does not apply. in-scope namespaces The set of namespace bindings provided by the XProc processor. The processor computes this set of bindings by taking a union of the bindings on the step element itself as well as the bindings on any of the options and parameters used in computing values for the step (see ). The results of computing the union of namespaces in the presence of conflicting declarations for a particular prefix are implementation-dependent. Some steps may also provide for implementation-defined or implementation-dependent amendments to the contexts. Those amendments are in addition to any specified by XProc.

XProc processors that support XPath 2.0 are XPath 2.0 processors. Such processors can refer to the primitive atomic schema types, but cannot import additional types. XPath 2.0 processors must implement all of the XPath 2.0 functions, but are not expected to implement the functions described in .

When the XProc processor evaluates an XPath expression using XPath 2.0, unless otherwise indicated by a particular step, it does so with the following static context: XPath 1.0 compatibility mode Is true if the indicated XPath version is 1.0, false otherwise. Statically known namespaces The namespace declarations in-scope for the containing element. Default element/type namespace The null namespace. Default function namespace The function namespace. Function names that do not contain a colon always refer to the default function namespace, any in-scope binding for the default namespace does not apply. This specification does not provide a mechanism to override the default function namespace. In-scope schema definitions A basic XPath 2.0 XProc processor includes the following named type definitions in its in-scope schema definitions: All the primitive atomic types defined in , with the exception of xs:NOTATION. That is: xs:string, xs:boolean, xs:decimal, xs:double, xs:float, xs:date, xs:time, xs:dateTime, xs:duration, xs:QName, xs:anyURI, xs:gDay, xs:gMonthDay, xs:gMonth, xs:gYearMonth, xs:gYear, xs:base64Binary, and xs:hexBinary. The derived atomic type xs:integer defined in . The types xs:anyType, xs:anySimpleType, xs:yearMonthDuration, xs:dayTimeDuration, xs:anyAtomicType, xs:untyped, and xs:untypedAtomic defined in . In-scope variables The union of the in-scope specified options and variables are available as variable bindings to the XPath processor. An option that has neither a specified value nor a default value will not appear as an in-scope variable. Consequently, an attempt to refer to that variable will raise an error. Context item static type Document. Function signatures The signatures of the and the . Statically known collations Implementation-defined but must include the Unicode code point collation. The version of Unicode supported is implementation-defined, but it is recommended that the most recent version of Unicode be used. Default collation Unicode code point collation. Base URI The base URI of the element on which the expression occurs. Statically known documents None. Statically known collections None. And the following dynamic context: context item The document node of a document. The document is either specified with a connection or is taken from the default readable port. It is a dynamic error if a document sequence appears where a document to be used as the context node is expected. If there is no explicit connection and there is no default readable port then the context node is undefined. context position and context size The context position and context size are both “1”. Variable values The union of the in-scope options and variables are available as variable bindings to the XPath processor. Function implementations The and the . Current dateTime The point in time returned as the current dateTime is implementation-defined. Implicit timezone The implicit timezone is implementation-defined. Available documents The set of available documents (those that may be retrieved with a URI) is implementation-dependent. Available collections The set of available collections is implementation-dependent. Default collection None.

When a step evaluates an XPath expression using XPath 2.0, unless otherwise indicated by a particular step, it does so with the following static context: XPath 1.0 compatibility mode Is true if the indicated XPath version is 1.0, false otherwise. Statically known namespaces The namespace declarations in-scope for the containing element or made available through p:namespaces. Default element/type namespace The null namespace. Default function namespace The function namespace. Function names that do not contain a colon always refer to the default function namespace, any in-scope binding for the default namespace does not apply. This specification does not provide a mechanism to override the default function namespace. In-scope schema definitions The same as the . In-scope variables None, unless otherwise specified by the step. Context item static type Document. Function signatures The signatures of the . Statically known collations Implementation-defined but must include the Unicode code point collation. Default collation Unicode code point collation. Base URI The base URI of the element on which the expression occurs. Statically known documents None. Statically known collections None. And the following initial dynamic context: context item The document node of the document that appears on the primary input of the step, unless otherwise specified by the step. context position and context size The context position and context size are both “1”, unless otherwise specified by the step. Variable values None, unless otherwise specified by the step. Function implementations The . Current dateTime An implementation-defined point in time. Implicit timezone The implicit timezone is implementation-defined. Available documents The set of available documents (those that may be retrieved with a URI) is implementation-dependent. Available collections None. Default collection None. Some steps may also provide for implementation-defined or implementation-dependent amendments to the contexts. Those amendments are in addition to any specified by XProc.

The XProc processor must support the additional functions described in this section in XPath expressions evaluated by the processor. In the following descriptions, the names of types (string, boolean, etc.) should be taken to mean the corresponding data types for an implementation that uses XPath 2.0 and as the most appropriate XPath 1.0 types for an XPath 1.0 implementation.

XPath expressions within a pipeline document can interrogate the processor for information about the current state of the pipeline. Various aspects of the processor are exposed through the p:system-property function in the pipeline namespace: xs:string p:system-property xs:stringproperty The $property string must have the form of a QName; the QName is expanded into a name using the namespace declarations in scope for the expression. It is a dynamic error if the specified QName cannot be resolved with the in-scope namespace declarations. The p:system-property function returns the string representing the value of the system property identified by the QName. If there is no such property, the empty string must be returned. Implementations must provide the following system properties, which are all in the XProc namespace: p:episode Returns a string which should be unique for each invocation of the pipeline processor. In other words, if a processor is run several times in succession, or if several processors are running simultaneously, each invocation of each processor should get a distinct value from p:episode. The unique identifier must be a valid XML name. p:language Returns a string which identifies the current language, for example, for message localization purposes. The exact format of the language string is implementation-defined but should be consistent with the xml:lang attribute. p:product-name Returns a string containing the name of the implementation, as defined by the implementer. This should normally remain constant from one release of the product to the next. It should also be constant across platforms in cases where the same source code is used to produce compatible products for multiple execution platforms. p:product-version Returns a string identifying the version of the implementation, as defined by the implementer. This should normally vary from one release of the product to the next, and at the discretion of the implementer it may also vary across different execution platforms. p:vendor Returns a string which identifies the vendor of the processor. p:vendor-uri Returns a URI which identifies the vendor of the processor. Often, this is the URI of the vendor's web site. p:version Returns the version(s) of XProc implemented by the processor as a space-separated list. For example, a processor that supports XProc 1.0 would return “1.0”; a processor that supports XProc 1.0 and 2.0 would return “1.0 2.0”; a processor that supports only XProc 2.0 would return “2.0”. p:xpath-version Returns the version(s) of XPath implemented by the processor for evaluating XPath expressions on XProc elements. The result is a space-separated list of versions supported. For example, a processor that only supports XPath 1.0 would return “1.0”; a processor that supports XPath 2.0 and XPath 1.0 backwards compatibility mode could return “1.0 2.0”; a processor that supports only XPath 2.0 would return “2.0”. p:psvi-supported Returns true if the implementation supports passing PSVI annotations between steps, false otherwise. Implementations may support additional system properties but such properties must be in a namespace and must not be in the XProc namespace.

The p:step-available function reports whether or not a particular type of step is understood by the processor. xs:boolean p:step-available xs:stringstep-name The $step-type string must have the form of a QName; the QName is expanded into a name using the namespace declarations in-scope for the expression. The p:step-available function returns true if and only if the processor knows how to evaluate steps of the specified type.

The p:value-available function reports whether or not a particular in-scope option has a value. xs:boolean p:value-available xs:stringoption-name xs:boolean p:value-available xs:stringoption-name xs:booleanfail-if-unknown The $option-name string must have the form of a QName; the QName is expanded into a name using the namespace declarations in-scope for the expression. The p:value-available function returns true if and only if the name specified is the name of an in-scope binding and the binding has a value. It is a dynamic error if the name specified is not the name of an in-scope option or variable. In the two-argument form, it is not an error to specify a name that is not the name of an in-scope option or variable if $fail-if-unknown is false; the function simply returns false. The semantics of the two-argument form when $fail-if-unknown is true are precisely the same as the single argument form. Consider the following example: If the path option is specified in the call to ex:dir-list, then the first p:when clause will be evaluated and the specified value will be used. If the option is not specified, then the p:otherwise clause will be evaluated and "." will be used instead.

Both p:for-each and p:viewport process a sequence of documents. The iteration position is the position of the current document in that sequence: the first document has position 1, the second 2, etc. The p:iteration-position function returns the iteration position of the nearest ancestor p:for-each or p:viewport. xs:integer p:iteration-position If there is no p:for-each or p:viewport among the ancestors of the element on which the expression involving p:iteration-position occurs, it returns 1.

Both p:for-each and p:viewport process a sequence of documents. The iteration size is the total number of documents in that sequence. The p:iteration-size function returns the iteration size of the nearest ancestor p:for-each or p:viewport. xs:integer p:iteration-size If there is no p:for-each or p:viewport among the ancestors of the element on which the expression involving p:iteration-size occurs, it returns 1.

Returns the base URI of the specified node, if it has one. This function provides an interoperable way for XPath 1.0 based processors to access the base URI of a node. It is conceptually the same as the XPath 2.0 fn:base-uri() function. xs:string p:base-uri xs:string p:base-uri node()node If no argument is specified, the context node is taken to be the argument. This function returns the [base-uri] property of its argument, or the empty string if no base URI is defined for that argument or argument type. This function is defined in our namespace because it would be inappropriate to require XPath 1.0 based processors to support the fn:base-uri function; its semantics are deeply rooted in the XPath 2.0 data model which differs from the XPath 1.0 data model.

Resolves a relative URI with respect to a particular base URI. This function provides an interoperable way for XPath 1.0 based processors to compose URI references. It is conceptually the same as the XPath 2.0 fn:resolve-uri() function. xs:string p:resolve-uri xs:stringrelative xs:string p:resolve-uri xs:stringrelative xs:stringbase If no base is specified, the base URI of the context node is used. This function is defined in our namespace because it would be inappropriate to require XPath 1.0 based processors to support the fn:resolve-uri function; its semantics are rooted in the XPath 2.0 data model which differs from the XPath 1.0 data model.

Returns true if and only if the processor supports the version specified. xs:boolean p:version-available xs:decimalversion A version 1.0 processor will return true() when p:version-available(1.0) is evaluated.

Returns true if and only if the processor supports the XPath version specified. xs:boolean p:xpath-version-available xs:decimalversion A processor that supports XPath 2.0 will return true() when p:xpath-version-available(2.0) is evaluated.

It is implementation-defined if the processor supports any other XPath extension functions. Additional extension functions, if any, must not use any of the XProc namespaces.

XML documents flow between steps in an XProc pipeline. identifies the properties of those documents that must be available. Implementations may also have the ability to pass PSVI annotations between steps. Whether or not the pipeline processor supports passing PSVI annotations between steps is implementation-defined. The exact PSVI properties that are preserved when documents are passed between steps is implementation-defined. A pipeline can use the p:psvi-supported system property to determine whether or not PSVI properties can be passed between steps. A pipeline can assert that PSVI support is required with the psvi-required attribute: On a p:pipeline or p:declare-step, psvi-required indicates whether or not the declared step requires PSVI support. It is a dynamic error if a processor that does not support PSVI annotations attempts to invoke a step which asserts that they are required. On a p:library, the psvi-required attribute provides a default value for all of its p:pipeline and p:declare-step children that do not specify a value themselves. Many of the steps that an XProc pipeline can use are transformative in nature. The p:delete step, for example, can remove elements and attributes; the p:label-elements step can add attributes; etc. If PSVI annotations were always preserved, the use of such steps could result in documents that were inconsistent with their schema annotations. In order to avoid these inconsistencies, most steps must not produce PSVI annotated results even when PSVI passing is supported. If PSVI passing is supported, the following constraints apply: Implementations must faithfully transmit any PSVI properties produced on step outputs to the steps to which they are connected. When only a subset of the input is processed by a step (because a select expression appears on an input port or a match expression is used to process only part of the input), any PSVI annotations that appear on the selected input must be preserved in the resulting documents passed to the step. Note that ID/IDREF constraints, and any other whole-document constraints, may not be satisfied within the selected portion, irrespective of what its PSVI properties claim. If an output of a compound step is connected to an output which includes PSVI properties, those properties must be preserved on the output of the compound step, except for the output of p:viewport which must not contain any PSVI properties. If an implementation supports XPath 2.0, the data model constructed with which to evaluate XPath expressions and match patterns should take advantage of as much PSVI information as possible. Except as specified above, or in the descriptions of individual steps, implementations must not include PSVI properties in the outputs of steps defined by this specification. It is implementation-defined what PSVI properties, if any, are produced by extension steps. The exceptions in the standard XProc steps are the p:validate-with-xml-schema, p:validate-with-relax-ng, and p:validate-with-schematron steps, p:xslt (when XSLT 2.0 is used), p:xquery, p:identity, and p:split-sequence. A processor that supports passing PSVI properties between steps is always free to do so. Even if psvi-required="false" is explicitly specified, it is not an error for a step to produce a result that includes additional PSVI properties, provide it does not violate the constraints above.

Variables are name/value pairs. Pipeline authors can create variables to hold computed values. A variable is a name/value pair where the name is an expanded name and the value must be a string or xs:untypedAtomic. Variables and options share the same scope and may shadow each other.

Some steps accept options. Options are name/value pairs, like variables. Unlike variables, the value of an option can be changed by the caller. An option is a name/value pair where the name is an expanded name and the value must be a string or xs:untypedAtomic. The options declared on a step are its declared options. Option names are always expressed as literal values, pipelines cannot construct option names dynamically. The options on a step which have specified values, either because a p:with-option element specifies a value or because the declaration included a default value, are its specified options. How outside values are specified for pipeline options on the pipeline initially invoked by the processor is implementation-defined. In other words, the command line options, APIs, or other mechanisms available to specify such options values are outside the scope of this specification.

Some steps accept parameters. Parameters are name/value pairs, like variables and options. Unlike variables and options, which have names known in advance to the pipeline, parameters are not declared and their names may be unknown to the pipeline author. Pipelines can dynamically construct sets of parameters. Steps can read dynamically constructed sets on parameter input ports. A parameter is a name/value pair where the name is an expanded name and the value must be a string or xs:untypedAtomic. A parameter input port is a distinguished kind of input port which accepts (only) dynamically constructed parameter name/value pairs. See . Analogous to primary input ports, steps that have parameter inputs may designate at most one parameter input port as a primary parameter input port. If a step has a parameter input port which is explicitly marked “primary='true'”, or if it has exactly one parameter input port and that port is not explicitly marked “primary='false'”, then that parameter input port is the primary parameter input port of the step. If a step has a single parameter input port and that port is explicitly marked “primary='false'”, or if a step has more than one parameter input port and none is explicitly marked as the primary, then the primary parameter input port of that step is undefined. How outside values are specified for pipeline parameters on the pipeline initially invoked by the processor is implementation-defined. In other words, the command line options, APIs, or other mechanisms available to specify such parameter values are outside the scope of this specification.

An XProc pipeline may attempt to access arbitrary network resources: steps such as p:load and p:http-request can attempt to read from an arbitrary URI; steps such as p:store can attempt to write to an arbitrary location; p:exec can attempt to execute an arbitrary program. Note, also, that some steps, such as p:xslt and p:xquery, include extension mechanisms which may attempt to execute arbitrary code. In some environments, it may be inappropriate to provide the XProc pipeline with access to these resources. In a server environment, for example, it may be impractical to allow pipelines to store data. In environments where the pipeline cannot be trusted, allowing the pipeline to access arbitrary resources or execute arbitrary code may be a security risk. It is a dynamic error for a pipeline to attempt to access a resource for which it has insufficient privileges or perform a step which is forbidden. Which steps are forbidden, what privileges are needed to access resources, and under what circumstances these security constraints apply is implementation-dependent. Steps in a pipeline may call themselves recursively which could result in pipelines which will never terminate. A conformant XProc processor may limit the resources available to any or all steps in a pipeline. A conformant implementation may raise dynamic errors, or take any other corrective action, for any security problems that it detects.

A pipeline author may identify the version of XProc for which a particular pipeline was authored by setting the version attribute. The version attribute can be specified on p:declare-step, p:pipeline, or p:library. If specified, the value of the version attribute must be a xs:decimal. It is a static error if the value of the version attribute is not a xs:decimal. The version of XProc defined by this specification is “1.0”. A pipeline author must identify the version of XProc on the document element of a pipeline document. It is a static error if a required version attribute is not present. The version identified applies to the element on which the version attribute appears and all of its descendants, unless or until another version is explicitly identified. When a processor encounters an explicit version (other than a version which it implements), it proceeds in backwards- or forwards-compatible mode.

If the processor encounters a request for a previous version of XProc (e.g, if a "2.0" processor encounters an explicit request for the "1.0" language), it must process the pipeline as if it was a processor for the requested version: it must enforce the semantics of the requested version, it must report steps not known in that version as errors, etc. It is a static error if the processor encounters an explicit request for a previous version of the language and it is unable to process the pipeline using those semantics.

If the processor encounters an explicit version which it does not recognize, it processes the pipeline in forwards-compatible mode. Forwards-compatible mode relaxes several static errors, turning them into dynamic errors so that a pipeline author can write a pipeline which conditionally uses new language features. In forwards-compatible mode: On any element in the XProc namespace, unrecognized attributes (other than extension attributes) are ignored. On any step in the XProc namespace, unknown options are ignored. If a step in the XProc namespace includes an unknown input port with an explicit connection, the connection is treated normally for the purpose of computing the dependencies in the pipeline but it is otherwise ignored. Unknown input ports must not be treated as primary input ports; it will always be an error if they are used but not explicitly connected. If a step in the pipeline includes an explicit connection to an unknown output port on a step in the XProc namespace, the connection is treated normally for the purpose of computing the dependencies in the pipeline. An empty sequence of documents must appear on that connection. As a consequence of the rules above, future specifications must not change the semantics of existing step types without changing their names. Although they may add new input and output ports, such changes should be done with care; they should in some sense be limited to ancillary inputs and outputs and they must not be primary input ports.

In forwards-compatible mode, it is not a static error to encounter the following step: ]]> The processor will simply ignore the “ancillary” port. Suppose that XProc version 2.0 changes the definition of the p:xslt step so that it has an additional output port, messages. Then consider the following pipeline: ]]> When run by a "2.0" or later processor, it will count the documents that appear on the messages port. When run by a “1.0” processor in forwards-compatible mode, the binding to the “messages” port is not a static error. Dynamically, the "1.0" processor will always produce a count of zero, because an empty sequence of documents will always appear on the messages port.

This section describes the normative XML syntax of XProc. This syntax is sufficient to represent all the aspects of a pipeline, as set out in the preceding sections. XProc is intended to work equally well with and . Unless otherwise noted, the term “XML” refers equally to both versions. Unless otherwise noted, the term Namespaces in XML refers equally to and . Support for pipeline documents written in XML 1.1 and pipeline inputs and outputs that use XML 1.1 is implementation-defined. Elements in a pipeline document represent the pipeline, the steps it contains, the connections between those steps, the steps and connections contained within them, and so on. Each step is represented by an element; a combination of elements and attributes specify how the inputs and outputs of each step are connected and how options and parameters are passed. Conceptually, we can speak of steps as objects that have inputs and outputs, that are connected together and which may contain additional steps. Syntactically, we need a mechanism for specifying these relationships. Containment is represented naturally using nesting of XML elements. If a particular element identifies a compound step then the step elements that are its immediate children form its subpipeline. The connections between steps are expressed using names and references to those names. Six kinds of things are named in XProc: Step types, Steps, Input ports (both parameter and document), Output ports, Options and variables, and Parameters

There are three namespaces associated with XProc: http://www.w3.org/ns/xproc The namespace of the XProc XML vocabulary described by this specification; by convention, the namespace prefix “p:” is used for this namespace. http://www.w3.org/ns/xproc-step The namespace used for documents that are inputs to and outputs from several standard and optional steps described in this specification. Some steps, such as p:http-request and p:store, have defined input or output vocabularies. We use this namespace for all of those documents. The conventional prefix “c:” is used for this namespace. http://www.w3.org/ns/xproc-error The namespace used for errors. The conventional prefix “err:” is used for this namespace. This specification also makes use of the prefix “xs:” to refer to the namespace http://www.w3.org/2001/XMLSchema.

Names are used to identify step types, steps, ports, options and variables, and parameters. Step types, options, variables, and parameters are named with QNames. Steps and ports are named with NCNames. The scope of a name is a measure of where it is available in a pipeline. If two names are in the same scope, we say that they are visible to each other. The scope of the names of the step types is the pipeline in which they are declared, including any declarations imported from libraries via p:import. Nested pipelines inherit the step types in scope for their parent. In other words, the step types that are in scope in a p:pipeline or p:declare-step are: The standard, built-in types (p:pipeline, p:choose, etc.). Any implementation-provided types. Any step types declared in the pipeline (the p:pipeline and p:declare-step children of the pipeline element). The types of any p:pipelines or p:declare-steps that are imported. Any types that are in the scope of any p:library that is imported. Any step types that are in scope for the pipeline's parent p:pipeline or p:declare-step, if it has one. The type of the pipeline itself, if it has one. The step types that are in scope in a p:library are: The standard, built-in types (p:pipeline, p:choose, etc.). Any implementation-provided types. Any step types declared in the library (the p:pipeline and p:declare-step children of the p:library element). The types of p:pipelines or p:declare-steps that are imported into the library. Any types that are in the scope of any p:library that is imported. All the step types in a pipeline or library must have unique names: it is a static error if any step type name is built-in and/or declared or defined more than once in the same scope. The scope of the names of the steps themselves is determined by the environment of each step. In general, the name of a step, the names of its sibling steps, the names of any steps that it contains directly, the names of its ancestors, and the names of the siblings of its ancestors are all in a common scope. All steps in the same scope must have unique names: it is a static error if two steps with the same name appear in the same scope. The scope of an input or output port name is the step on which it is defined. The names of all the ports on any step must be unique. Taken together, these uniqueness constraints guarantee that the combination of a step name and a port name uniquely identifies exactly one port on exactly one in-scope step. The scope of option and variable names is determined by where they are declared. When an option is declared with p:option (or a variable with p:variable), unless otherwise specified, its scope consists of the sibling elements that follow its declaration and the descendants of those siblings. It is a static error if an option or variable declaration duplicates the name of any other option or variable in the same environment. That is, no option or variable may lexically shadow another option or variable with the same name. Parameter names are not scoped; they are distinct on each step.

When a relative URI appears in an option value, the base URI against which it must be made absolute is the base URI of the p:option element. If an option value is specified using a syntactic shortcut, the base URI of the step on which the shortcut attribute appears must be used. In general, whenever a relative URI appears, its base URI is the base URI of the nearest ancestor element. The pipeline author can control the base URIs of elements within the pipeline document with the xml:base attribute. The xml:base attribute may appear on any element in a pipeline and has the semantics outlined in .

A pipeline author can provide a globally unique identifier for any element in a pipeline with the xml:id attribute. The xml:id attribute may appear on any element in a pipeline and has the semantics outlined in .

A connection associates an input or output port with some data source. A document or a sequence of documents can be connected to a port in four ways: by source, by URI, by providing an inline document, or by making it explicitly empty. Each of these mechanisms is allowed on the p:input, p:output, p:xpath-context, p:iteration-source, and p:viewport-source elements. Specified by URI A document is specified by URI if it is referenced with a URI. The href attribute on the p:document or p:data element is used to refer to documents by URI. In this example, the input to the p:identity step named “otherstep” comes from “http://example.com/input.xml”. Specified by source A document is specified by source if it references a specific port on another step. The step and port attributes on the p:pipe element are used for this purpose. In this example, the “source” input to the p:xinclude step named “expand” comes from the “result” port of the step named “otherstep”. See the description of p:pipe for a complete description of the ports that can be connected. Specified inline An inline document is specified directly in the body of the element to which it connects. The content of the p:inline element is used for this purpose. In this example, the “stylesheet” input to the XSLT step named “xform” comes from the content of the p:input element itself. Inline documents are considered “quoted”. The pipeline processor passes them literally to the port, even if they contain elements from the XProc namespace or other namespaces that would have other semantics outside of the p:inline. Specified explicitly empty An empty sequence of documents is specified with the p:empty element. In this example, the “source” input to the XSLT 2.0 step named “generate” is explicitly empty: If you omit the connection on a primary input port, a connection to the default readable port will be assumed. Making the connection explicitly empty guarantees that the connection will be to an empty sequence of documents. It is inconsistent with the specification to specify an empty connection as the context for evaluating an XPath expression. When an empty connection is specified for an XPath 1.0 expression, an empty document node must be used instead as the context node. Note that a p:input or p:output element may contain more than one p:pipe, p:document, p:data, or p:inline element. If more than one connection is provided, then the specified sequence of documents is made available on that port in the same order as the connections.

Pipeline authors may add documentation to their pipeline documents with the p:documentation element. Except when it appears as a descendant of p:inline, the p:documentation element is completely ignored by pipeline processors, it exists simply for documentation purposes. If a p:documentation is provided as a descendant of p:inline, it has no special semantics, it is treated literally as part of the document to be provided on that port. The p:documentation element has no special semantics when it appears in documents that flow through the pipeline. Pipeline processors that inspect the contents of p:documentation elements and behave differently on the basis of what they find are not conformant. Processor extensions must be specified with p:pipeinfo.

Pipeline authors may add annotations to their pipeline documents with the p:pipeinfo element. The semantics of p:pipeinfo elements are implementation-defined. Processors should specify a way for their annotations to be identified, perhaps with extension attributes. Where p:documentation is intended for human consumption, p:pipeinfo elements are intended for processor consumption. A processor might, for example, use annotations to identify some particular aspect of an implementation, to request additional, perhaps non-standard features, to describe parallelism constraints, etc. When a p:pipeinfo appears as a descendant of p:inline, it has no special semantics; in that context it must be treated literally as part of the document to be provided on that port. The p:pipeinfo element has no special semantics when it appears in documents that flow through the pipeline.

An element from the XProc namespace may have any attribute not from the XProc namespace, provided that the expanded-QName of the attribute has a non-null namespace URI. Such an attribute is called an extension attribute. The presence of an extension attribute must not cause the connections between steps to differ from the connections that would arise in the absence of the attribute. They must not cause the processor to fail to signal an error that would be signaled in the absence of the attribute. A processor which encounters an extension attribute that it does not implement must behave as if the attribute was not present.

Any element in the XProc namespace may have a use-when attribute which must contain an XPath expression that can be evaluated statically. If the attribute is present and the effective boolean value of the expression is false, then the element and all of its descendants are effectively excluded from the pipeline document. If a node is effectively excluded, the processor must behave as if the element was not present in the document. Elements that are not in the XProc namespace may also have a use-when attribute, but the attribute must be in the XProc namespace. The semantics of a p:use-when attribute on an element not in the XProc namespace are the same as the semantics of a use-when attribute on an element in the XProc namespace. Conditional element exclusion occurs before any static analysis of the pipeline. The effective exclusion of use-when processing occurs after XML parsing and has no effect on well-formedness or validation errors which will be reported in the usual way. Note also that use-when is not performed when it occurs on the descendant of a p:inline element. For the purposes of evaluating a use-when expression, the context node, position, and size are all undefined. No in-scope bindings are available. There are no readable ports. There are no available documents or available collections. There are some additional restrictions on the XPath extension functions that are available in a use-when expression: The p:episode system property should not be used. The value of the p:episode system property in a use-when expression is implementation-dependent. The p:step-available function cannot be used to test for the availability of extension steps (because the libraries that declare them may not have been imported). The results of testing for steps not in the XProc namespace in a use-when expression are implementation-dependent. The steps available and possibly other aspects of the expression may depend on the version specified for a pipeline, see . For example, in a “1.0” pipeline, the processor should not report that “2.0” steps are available. It is a static error if a use-when expression refers to the context or attempts to refer to any documents or collections.

The description of each element in the pipeline namespace is accompanied by a syntactic summary that provides a quick overview of the element's syntax: The content model fragments in these tableaux are presented in a simple, compact notation. In brief: A name represent exactly one occurrence of an element with that name. Parentheses are used for grouping. Elements or groups separated by a comma (“,”) represent an ordered sequence: a followed by b followed by c: (a,b,c). Elements or groups separated by a vertical bar (“|”) represent a choice: a or b or c: (a | b | c). Elements or groups separated by an ampersand (“&”) represent an unordered sequence: a and b and c, in any order: (a & b & c). An element or group followed by a question mark (“?”) is optional; it may or may not occur but if it occurs it can occur only once. An element or group followed by an asterisk (“*”) is optional and may be repeated; it may or may not occur and if it occurs it can occur any number of times. An element or group followed by a plus (“+”) is required and may be repeated; it must occur at least once, and it can occur any number of times. For clarity of exposition, some attributes and elements are elided from the summaries: An xml:id attribute is allowed on any element. It has the semantics of . An xml:base attribute is allowed on any element. It has the semantics of . A use-when attribute is allowed on any element, see . The p:documentation and p:pipeinfo elements are not shown, but they are allowed anywhere. The p:log element is allowed on any step that has a p:output. Attributes that are syntactic shortcuts for option values are not shown. The types given for attributes should be understood as follows: ID, NCName, NMTOKEN, NMTOKENS, anyURI, boolean, integer, string: As per including whitespace normalization as appropriate. QName: With whitespace normalization as per and according to the following definition: In the context of XProc, a QName is almost always a QName in the Namespaces in XML sense. Note, however, that p:option and p:with-param values can get their namespace declarations in a non-standard way (with p:namespaces) and QNames that have no prefix are always in no-namespace, irrespective of the default namespace. PrefixList: As a list with item type NMTOKEN, per , including whitespace normalization. XPathExpression, XSLTMatchPattern: As a string per , including whitespace normalization, and the further requirement to be a conformant Expression per or , as appropriate, or Match pattern per or , as appropriate.

A number of errors apply generally: It is a static error if the pipeline element is not p:pipeline, p:declare-step, or p:library. It is a static error if any element in the XProc namespace has attributes not defined by this specification unless they are extension attributes. It is a static error if any required attribute is not provided. It is a dynamic error if any attribute value does not satisfy the type required for that attribute. It is a static error if any element in the XProc namespace or any step has element children other than those specified for it by this specification. In particular, the presence of atomic steps for which there is no visible declaration may raise this error. It is a static error if any step directly contains text nodes that do not consist entirely of whitespace. It is a dynamic error if any option value does not satisfy the type required for that option. It is a static error if a compound step has no contained steps. It is a dynamic error if any attempt is made to dereference a URI where the scheme of the URI reference is not supported. Implementations are encouraged to support as many schemes as is practical and, in particular, they should support both the file: and http(s): schemes. The set of URI schemes actually supported is implementation-defined. It is a dynamic error if a step is unable or incapable of performing its function. This is a general error code for “step failed” (e.g., if the input isn't of the expected type or if attempting to process the input causes the implementation to abort). Users and implementors who create extension steps are encouraged to use this code for general failures. In most steps which use a select expression or match pattern, any kind of node can be identified by the expression or pattern. However, some expressions and patterns on some steps are only applicable to some kinds of nodes (e.g., it doesn't make sense to speak of adding attributes to a comment!). It is a dynamic error if a select expression or match pattern returns a node type that is not allowed by the step. If an XProc processor can determine statically that a dynamic error will always occur, it may report that error statically provided that the error does not occur among the descendants of a p:try. Dynamic errors inside a p:try must not be reported statically. They must be raised dynamically so that p:catch processing can be performed on them.

This section describes the core steps of XProc. Several of the steps defined in this specification refer to other, evolving XML technologies (XSLT, XQuery, XSL-FO, etc.). Where this specification identifies a specific version of a technology, implementors must implement the specified version or any subsequent edition or version that is backwards compatible. At user option, they may support other, incompatible versions or extensions.

A p:pipeline declares a pipeline that can be evaluated by an XProc processor. It encapsulates the behavior of a subpipeline. Its children declare inputs, outputs, and options that the pipeline exposes and identify the steps in its subpipeline. (A p:pipeline is a simplified form of step declaration.) All p:pipeline pipelines have an implicit primary input port named “source”, an implicit primary parameter input port named “parameters”, and an implicit primary output port named “result”. Any input or output ports that the p:pipeline declares explicitly are in addition to those ports and may not be declared primary. Viewed from the outside, a p:pipeline is a black box which performs some calculation on its inputs and produces its outputs. From the pipeline author's perspective, the computation performed by the pipeline is described in terms of contained steps which read the pipeline's inputs and produce the pipeline's outputs. The version attribute identifies the version of XProc for which this pipeline was authored. If the p:pipeline has no ancestors in the XProc namespace, then it must have a version attribute. See . If a pipeline does not have a type then that pipeline cannot be invoked as a step. The p:pipeline element is just a simplified form of step declaration. A document that reads:<p:pipeline some-attributes> some-content </p:pipeline>can be interpreted as if it read:<p:declare-step some-attributes> <p:input port='source' primary='true'/> <p:input port='parameters' kind='parameter' primary='true'/> <p:output port='result' primary='true'/> some-content </p:declare-step>See p:declare-step for more details.

A pipeline might accept a document as input; perform XInclude, validation, and transformation; and produce the transformed document as its output.

A for-each is specified by the p:for-each element. It is a compound step that processes a sequence of documents, applying its subpipeline to each document in turn. When a pipeline needs to process a sequence of documents using a subpipeline that only processes a single document, the p:for-each construct can be used as a wrapper around that subpipeline. The p:for-each will apply that subpipeline to each document in the sequence in turn. The result of the p:for-each is a sequence of documents produced by processing each individual document in the input sequence. If the p:for-each has one or more output ports, what appears on each of those ports is the sequence of documents that is the concatenation of the sequence produced by each iteration of the loop on the port to which it is connected. If the iteration source for a p:for-each is an empty sequence, then the subpipeline is never run and an empty sequence is produced on all of the outputs. The p:iteration-source is an anonymous input: its connection provides a sequence of documents to the p:for-each step. If no iteration sequence is explicitly provided, then the iteration source is read from the default readable port. The processor provides each document, one at a time, to the subpipeline represented by the children of the p:for-each on a port named current. For each declared output, the processor collects all the documents that are produced for that output from all the iterations, in order, into a sequence. The result of the p:for-each on that output is that sequence of documents. The environment inherited by the contained steps of a p:for-each is the inherited environment with these modifications: The port named “current” on the p:for-each is added to the readable ports. The port named “current” on the p:for-each is made the default readable port. If the p:for-each has a primary output port (explicit or supplied by default) and that port has no connection, then it is connected to the primary output port of the last step in the subpipeline. It is a static error if the primary output port has no explicit connection and the last step in the subpipeline does not have a primary output port. Note that outputs declared for a p:for-each serve a dual role. Inside the p:for-each, they are used to read results from the subpipeline. Outside the p:for-each, they provide the aggregated results. The sequence attribute on a p:output inside a p:for-each only applies inside the step. From the outside, all of the outputs produce sequences.

Within a p:for-each, the p:iteration-position and p:iteration-size are taken from the sequence of documents that will be processed by the p:for-each. The total number of documents is the p:iteration-size; the ordinal value of the current document (the document appearing on the current port) is the p:iteration-position. In the case where no XPath expression that must be evaluated by the processor makes any reference to p:iteration-size, its value does not actually have to be calculated (and the entire input sequence does not, therefore, need to be buffered so that its size can be calculated before processing begins).

A p:for-each might accept a sequence of chapters as its input, process each chapter in turn with XSLT, a step that accepts only a single input document, and produce a sequence of formatted chapters as its output. The //chapter elements of the document are selected. Each chapter is transformed into HTML and XSL Formatting Objects using an XSLT step. The resulting HTML and FO documents are aggregated together and appear on the html-results and fo-results ports, respectively, of the chapters step itself.

A viewport is specified by the p:viewport element. It is a compound step that processes a single document, applying its subpipeline to one or more subtrees of the document. The result of the p:viewport is a copy of the original document where the selected subtrees have been replaced by the results of applying the subpipeline to them. The p:viewport-source is an anonymous input: its connection provides a single document to the p:viewport step. If no document is explicitly provided, then the viewport source is read from the default readable port. It is a dynamic error if the viewport source does not provide exactly one document. The match attribute specifies an XSLT match pattern. Each matching node in the source document is wrapped in a document node, as necessary, and provided, one at a time, to the viewport's subpipeline on a port named current. The base URI of the resulting document that is passed to the subpipeline is the base URI of the matched element or document. It is a dynamic error if the match expression on p:viewport does not match an element or document. After a match is found, the entire subtree rooted at that match is processed as a unit. No further attempts are made to match nodes among the descendants of any matched node. The environment inherited by the contained steps of a p:viewport is the inherited environment with these modifications: The port named “current” on the p:viewport is added to the readable ports. The port named “current” on the p:viewport is made the default readable port. The p:viewport must contain a single, primary output port declared explicitly or supplied by default. If that port has no connection, then it is connected to the primary output port of the last step in the subpipeline. It is a static error if the primary output port is unconnected and the last step in the subpipeline does not have a primary output port. What appears on the output from the p:viewport will be a copy of the input document where each matching node is replaced by the result of applying the subpipeline to the subtree rooted at that node. In other words, if the match pattern matches a particular element then that element is wrapped in a document node and provided on the current port, the subpipeline in the p:viewport is evaluated, and the result that appears on the output port replaces the matched element. If no documents appear on the output port, the matched element will effectively be deleted. If exactly one document appears, the contents of that document will replace the matched element. If a sequence of documents appears, then the contents of each document in that sequence (in the order it appears in the sequence) will replace the matched element. The output of the p:viewport itself is a single document that appears on a port named “result”. Note that the semantics of p:viewport are special. The output port in the p:viewport is used only to access the results of the subpipeline. The output of the step itself appears on a port with the fixed name “result” that is never explicitly declared.

Within a p:viewport, the p:iteration-position and p:iteration-size are taken from the sequence of documents that will be processed by the p:viewport. The total number of documents is the p:iteration-size; the ordinal value of the current document (the document appearing on the current port) is the p:iteration-position. In the case where no XPath expression that must be evaluated by the processor makes any reference to p:iteration-size, its value does not actually have to be calculated (and the entire input sequence does not, therefore, need to be buffered so that its size can be calculated before processing begins).

A p:viewport might accept an XHTML document as its input, add an hr element at the beginning of all div elements that have the class value “chapter”, and return an XHTML document that is the same as the original except for that change. The nodes which match h:div[@class='chapter'] in the input document are selected. An hr is inserted as the first child of each h:div and the resulting version replaces the original h:div. The result of the whole step is a copy of the input document with a horizontal rule as the first child of each selected h:div.

A choose is specified by the p:choose element. It is a multi-container step that selects exactly one of a list of alternative subpipelines based on the evaluation of XPath expressions. A p:choose has no inputs. It contains an arbitrary number of alternative subpipelines, exactly one of which will be evaluated. The list of alternative subpipelines consists of zero or more subpipelines guarded by an XPath expression, followed optionally by a single default subpipeline. The p:choose considers each subpipeline in turn and selects the first (and only the first) subpipeline for which the guard expression evaluates to true in its context. If there are no subpipelines for which the expression evaluates to true, the default subpipeline, if it was specified, is selected. After a subpipeline is selected, it is evaluated as if only it had been present. The outputs of the p:choose are taken from the outputs of the selected subpipeline. The p:choose has the same number of outputs as the selected subpipeline with the same names. If the selected subpipeline has a primary output port, the port with the same name on the p:choose is also a primary output port. In order to ensure that the output of the p:choose is consistent irrespective of the subpipeline chosen, each subpipeline must declare the same number of outputs with the same names. If any of the subpipelines specifies a primary output port, each subpipeline must specify exactly the same output as primary. It is a static error if two subpipelines in a p:choose declare different outputs. As a convenience to authors, it is not an error if some subpipelines declare outputs that can produce sequences and some do not. Each output of the p:choose is declared to produce a sequence if that output is declared to produce a sequence in any of its subpipelines. It is a dynamic error if no subpipeline is selected by the p:choose and no default is provided. The p:choose can specify the context node against which the XPath expressions that occur on each branch are evaluated. The context node is specified as a connection in the p:xpath-context. If no explicit connection is provided, the default readable port is used. In an XPath 1.0 implementation, if the context node is connected to p:empty, or is unconnected and the default readable port is undefined, an empty document node is used instead as the context. In an XPath 2.0 implementation, the context item is undefined. Each conditional subpipeline is represented by a p:when element. The default branch is represented by a p:otherwise element.

A p:xpath-context element specifies the context node against which an XPath expression will be evaluated. When it appears in a p:when, it specifies the context for that p:when’s test attribute. When it appears in p:choose, it specifies the default context for all of the p:when elements in that p:choose. Only one connection is allowed and it works the same way that connections work on a p:input. No select expression is allowed. It is a dynamic error if more than one document appears on the connection for the xpath-context. The p:xpath-context element only provides the context node. The namespace bindings, in-scope variables, and other aspects of the context come from the element on which the XPath expression occurs. In an XPath 1.0 implementation, if the context node is connected to p:empty, or is unconnected and the default readable port is undefined, an empty document node is used instead as the context. In an XPath 2.0 implementation, the context item is undefined.

A when specifies one subpipeline guarded by a test expression. Each p:when branch of the p:choose has a test attribute which must contain an XPath expression. That XPath expression's effective boolean value is the guard for the subpipeline contained within that p:when. The p:when can specify a context node against which its test expression is to be evaluated. That context node is specified as a connection for the p:xpath-context. If no context is specified on the p:when, the context of the p:choose is used.

An otherwise specifies the default branch; the subpipeline selected if no test expression on any preceding p:when evaluates to true.

A p:choose might test the version attribute of the document element and validate with an appropriate schema.

A group is specified by the p:group element. In a p:try, it is a non-step wrapper, everywhere else, it is a compound step. A group encapsulates the behavior of its subpipeline. A p:group is a convenience wrapper for a collection of steps.

A try/catch is specified by the p:try element. It is a multi-container step that isolates a subpipeline, preventing any dynamic errors that arise within it from being exposed to the rest of the pipeline. The p:group represents the initial subpipeline and the recovery (or “catch”) pipeline is identified with a p:catch element. The p:try step evaluates the initial subpipeline and, if no errors occur, the outputs of that pipeline are the outputs of the p:try step. However, if any errors occur, the p:try abandons the first subpipeline, discarding any output that it might have generated, and evaluates the recovery subpipeline. If the recovery subpipeline is evaluated, the outputs of the recovery subpipeline are the outputs of the p:try step. If the recovery subpipeline is evaluated and a step within that subpipeline fails, the p:try fails. The outputs of the p:try are taken from the outputs of the initial subpipeline or the recovery subpipeline if an error occurred in the initial subpipeline. The p:try has the same number of outputs as the selected subpipeline with the same names. If the selected subpipeline has a primary output port, the port with the same name on the p:try is also a primary output port. In order to ensure that the output of the p:try is consistent irrespective of whether the initial subpipeline provides its output or the recovery subpipeline does, both subpipelines must declare the same number of outputs with the same names. If either of the subpipelines specifies a primary output port, both subpipelines must specify exactly the same output as primary. It is a static error if the p:group and p:catch subpipelines declare different outputs. As a convenience to authors, it is not an error if an output port can produce a sequence in the initial subpipeline but not in the recovery subpipeline, or vice versa. Each output of the p:try is declared to produce a sequence if that output is declared to produce a sequence in either of its subpipelines. A pipeline author can cause an error to occur with the p:error step. The recovery subpipeline of a p:try is identified with a p:catch: The environment inherited by the contained steps of the p:catch is the inherited environment with this modification: The port named “error” on the p:catch is added to the readable ports. What appears on the error output port is an error document. The error document may contain messages generated by steps that were part of the initial subpipeline. Not all messages that appear are indicative of errors; for example, it is common for all xsl:message output from the XSLT component to appear on the error output port. It is possible that the component which fails may not produce any messages at all. It is also possible that the failure of one component may cause others to fail so that there may be multiple failure messages in the document.

In general, it is very difficult to predict error behavior. Step failure may be catastrophic (programmer error), or it may be be the result of user error, resource failures, etc. Steps may detect more than one error, and the failure of one step may cause other steps to fail as well. The p:try/p:catch mechanism gives pipeline authors the opportunity to process the errors that caused the p:try to fail. In order to facilitate some modicum of interoperability among processors, errors that are reported on the error output port of a p:catch should conform to the format described here.

The error vocabulary consists of a root element, c:errors which contains zero or more c:error elements.

Each specific error is represented by an c:error element: The name and type attributes identify the name and type, respectively, of the step which failed. The code is a QName which identifies the error. For steps which have defined error codes, this is an opportunity for the step to identify the error in a machine-processable fashion. Many steps omit this because they do not include the concept of errors identified by QNames. If the error was caused by a specific document, or by the location of some erroneous construction in a specific document, the href, line, column, and offset attributes identify this location. Generally, the error location is identified either with line and column numbers or with an offset from the beginning of the document, but not usually both. The content of the c:error element is any well-formed XML. Specific steps, or specific implementations, may provide more detail about the format of the content of an error message.

Consider the following XSLT stylesheet: This stylesheet is pointless . ]]> If it was used in a step named “xform” in a p:try, the following error document might be produced: This stylesheet is pointless. ]]> It is not an error for steps to generate non-standard error output as long as it is well-formed.

A pipeline might attempt to process a document by dispatching it to some web service. If the web service succeeds, then those results are passed to the rest of the pipeline. However, if the web service cannot be contacted or reports an error, the p:catch step can provide some sort of default for the rest of the pipeline.

In addition to the six step types described in the preceding sections, XProc provides a standard library of atomic step types. The full vocabulary of standards steps is described in . All of the standard, atomic steps are invoked in the same way: Where “p:atomic-step” must be in the XProc namespace and must be declared in either the standard library for the XProc version supported by the processor or explicitly imported by the surrounding pipeline (see ).

Pipeline authors may also have access to additional steps not defined or described by this specification. Atomic extension steps are invoked just like standard steps: Extension steps must not be in the XProc namespace and there must be a visible step declaration at the point of use (see ). If the relevant step declaration has no subpipeline, then that step invokes the declared atomic step, which the processor must know how to perform. These steps are implementation-defined extensions. If the relevant step declaration has a subpipeline, then that step runs the declared subpipeline. These steps are user- or implementation-defined extensions. Pipelines can refer to themselves (recursion is allowed), to pipelines defined in imported libraries, and to other pipelines in the same library if they are in a library. It is a static error if a pipeline contains a step whose specified inputs, outputs, and options do not match the signature for steps of that type. It is a dynamic error if the running pipeline attempts to invoke a step which the processor does not know how to perform. The presence of other compound steps is implementation-defined; XProc provides no standard mechanism for defining them or describing what they can contain. It is a static error to use a declared step as a compound step.

Namespace qualified attributes on a step are extension attributes. Attributes, other than name, that are not namespace qualified are treated as a syntactic shortcut for specifying the value of an option. In other words, the following two steps are equivalent: The first step uses the standard p:with-option syntax: ]]> The second step uses the syntactic shortcut: ]]> Note that there are significant limitations to this shortcut syntax: It only applies to option names that are not in a namespace. It only applies to option names that are not otherwise used on the step, such as “name”. It can only be used to specify a constant value. Options that are computed at runtime must be written using the longer form. It is a static error if an option is specified with both the shortcut form and the long form. It is a static error to use an option on an atomic step that is not declared on steps of that type. The syntactic shortcuts apply equally to standard atomic steps and extension atomic steps.

A p:input identifies an input port for a step. In some contexts, p:input declares that a port with the specified name exists and identifies the properties of that port. In other contexts, it provides a connection for a port declared elsewhere. And in some contexts, it does both. The semantics of p:input are complicated further by the fact that there are two kinds of inputs, ordinary “document” inputs and “parameter” inputs.

The declaration of a document input identifies the name of the port, whether or not the port accepts a sequence, whether or not the port is a primary input port, and may provide a default connection for the port. An input declaration has the following form: The port attribute defines the name of the port. It is a static error to identify two ports with the same name on the same step. The sequence attribute determines whether or not a sequence of documents is allowed on the port. If sequence is not specified, or has the value false, then it is a dynamic error unless exactly one document appears on the declared port. The primary attribute is used to identify the primary input port. An input port is a primary input port if primary is specified with the value true or if the step has only a single input port and primary is not specified. It is a static error to specify that more than one input port is the primary. The kind attribute distinguishes between the two kinds of inputs: document inputs and parameter inputs. An input port is a document input port if kind is specified with the value “document” or if kind is not specified. If a connection is provided in the declaration, then select may be used to select a portion of the input identified by the p:empty, p:document, p:data, or p:inline elements in the p:input. This select expression applies only if the default connection is used. If an explicit connection is provided by the caller, then the default select expression is ignored. The p:pipe element is explicitly excluded from a declaration because it would make the default value of an input dependent on the execution of some part of the pipeline. Default values are designed so that they can be computed statically. On a p:declare-step for an atomic step, the p:input simply declares the input port. It is a static error to attempt to provide a connection for an input port on the declaration of an atomic step. An input connection has the following form: If no connection is provided for a primary input port, the input will be connected to the default readable port. It is a static error if no connection is provided and the default readable port is undefined. A select expression may also be provided with a connection. The select expression, if specified, applies the specified XPath select expression to the document(s) that are read. Each selected node is wrapped in a document (unless it is a document) and provided to the input port. In other words,provides a single document, butprovides a sequence of zero or more documents, one for each html:div in http://example.org/input.html. (Note that in the case of nested html:div elements, this may result in the same content being returned in several documents.) A select expression can equally be applied to input read from another step. This input:provides a sequence of zero or more documents, one for each html:div in the document (or each of the documents) that is read from the result port of the step named origin. The base URI of the document that results from a select expression is the base URI of the matched element or document. It is a dynamic error if the select expression on a p:input returns atomic values or anything other than element or document nodes (or an empty sequence). An input declaration may include a default connection. If no connection is provided for an input port which has a default connection, then the input is treated as if the default connection appeared. A default connection does not satisfy the requirement that a primary input port is automatically connected by the processor, nor is it used when no default readable port is defined. In other words, a p:declare-step or a p:pipeline can define defaults for all of its inputs, whether they are primary or not, but defining a default for a primary input usually has no effect. It's never used by an atomic step since the step, when it's called, will always connect the primary input port to the default readable port (or cause a static error). The only case where it has value is on a p:pipeline when that pipeline is invoked directly by the processor. In that case, the processor must use the default connection if no external connection is provided for the port.

The declaration of a parameter input identifies the name of the port and that the port is a parameter input. The port attribute defines the name of the port. It is a static error to identify two ports with the same name on the same step. The sequence attribute determines whether or not a sequence of documents is allowed on the port. A sequence of documents is always allowed on a parameter input port. It is a static error to specify any value other than true. The primary attribute is used to identify the primary parameter input port. An input port is a primary parameter input port if it is a parameter input port and primary is specified with the value true or if the step has only a single parameter input port and primary is not specified. It is a static error to specify that more than one parameter input port is the primary. The kind attribute distinguishes between the two kinds of inputs: document inputs and parameter inputs. An input port is a parameter input port only if the kind attribute is specified with the value “parameter”. It is a static error to specify any kind of input other than “document” or “parameter”. A parameter input port is a distinguished kind of input port. It exists only to receive computed parameters; if a step does not have a parameter input port then it cannot receive parameters. A parameter input port must satisfy all the constraints of a normal, document input port. It is a static error if the declaration of a parameter input port contains a connection; parameter input port declarations must be empty. When used on a step, parameter input ports are connected just like ordinary document ports. Parameter input ports always accept a sequence of documents. If no explicit connection is provided for a primary parameter input port, then the port will be connected to the primary parameter input port of the pipeline which contains the step. If no connection is provided for a parameter input port other than the primary parameter input port, then the port will be connected to an empty sequence of documents. It is a static error if a primary parameter input port is unconnected and the pipeline that contains the step has no primary parameter input port unless at least one explicit p:with-param is provided for that port. In other words, it is an error to leave a parameter input port unconnected, but any parameter passed explicitly to that port satisfies the connection requirement. This is an error: The parameter input port on the p:xslt step has no connection and the pipeline that contains the step has no primary parameter input port. This is not an error: Explicitly setting the “mode” parameter satisfies the binding requirement. No other parameters can be passed to the p:xslt step. This, also, is not an error: The parameter input port on the p:xslt step is bound to the primary parameter input port of the pipeline. The p:xslt step will receive the “mode” parameter and any other parameters passed to the pipeline. If a parameter input port on a p:pipeline is not connected, it is treated as if it was connected to an automatically created p:sink step. In other words, if a p:pipeline does not contain any steps that have parameter input ports, or if those ports are all explicitly connected elsewhere, the parameter input port is ignored. In this one case, it is not an error for an input port to be unconnected. A step which accepts a parameter input reads all of the documents presented on that port, using each c:param (either at the root or inside the c:param-set) to establish the value of the named parameter. If the same name appears more than once, the last value specified is used. If the step also has literal p:with-param elements, they are also considered in document order. In other words, p:with-param elements that appear before the parameter input may be overridden by the computed parameters; p:with-param elements that appear after may override the computed values. If a connection is manufactured for a primary parameter input port, that connection occurs logically last among the other parameters, options, and connections passed to the step. In other words, the parameter values that appear on that port will be used even if other values were specified with p:with-param elements. Users can change this priority by making the connection explicit and placing any p:with-param elements that they wish to function as overrides after the connection. All of the documents that appear on a parameter input must either be c:param documents or c:param-set documents. Consider the example in . This p:pipeline declares that it accepts parameters. Suppose that (through some implementation-defined mechanism) I have passed the parameters “output-type=fo” and “profile=unclassified” to the pipeline. These parameters are available on the parameters input port. When the XSLT step runs, it will read those parameters and combine them with any parameters specified literally on the step. Because the parameter input comes after the literal declaration for output-type on the step, the XSLT stylesheet will see both values that I passed in (“output-type=fo” and “profile=unclassified”). If the parameter input came before the literal declaration, then the XSLT stylesheet would see “output-type=html” and “profile=unclassified”. Most steps don't bother to declare parameter inputs, or provide explicit connections for them, and “the right thing” usually happens.

A c:param represents a parameter on a parameter input. The name attribute of the c:param must have the lexical form of a QName. If the namespace attribute is specified, then the expanded name of the parameter is constructed from the specified namespace and the name value. It is a dynamic error if the namespace attribute is specified, the name contains a colon, and the specified namespace is not the same as the in-scope namespace binding for the specified prefix. If the namespace attribute is not specified, and the name contains a colon, then the expanded name of the parameter is constructed using the name value and the namespace declarations in-scope on the c:param element. If the namespace attribute is not specified, and the name does not contain a colon, then the expanded name of the parameter is in no namespace. Any namespace-qualified attribute names that appear on the c:param element are ignored. It is a dynamic error for any unqualified attribute names other than “name”, “namespace”, or “value” to appear on a c:param element.

A c:param-set represents a set of parameters on a parameter input. The c:param-set contains zero or more c:param elements. It is a dynamic error if the parameter list contains any elements other than c:param. Any namespace-qualified attribute names that appear on the c:param-set element are ignored. It is a dynamic error for any unqualified attribute names to appear on a c:param-set element.

A p:iteration-source identifies input to a p:for-each. The select attribute and connection of a p:iteration-source work the same way that they do in a p:input.

A p:viewport-source identifies input to a p:viewport. Only one connection is allowed and it works the same way that connections work on a p:input. It is a dynamic error unless exactly one document appears on the p:viewport-source. No select expression is allowed.

A p:output identifies an output port, optionally connecting an input for it, if necessary. The port attribute defines the name of the port. It is a static error to identify two ports with the same name on the same step. An output declaration can indicate if a sequence of documents is allowed to appear on the declared port. If sequence is specified with the value true, then a sequence is allowed. If sequence is not specified on p:output, or has the value false, then it is a dynamic error if the step does not produce exactly one document on the declared port. The primary attribute is used to identify the primary output port. An output port is a primary output port if primary is specified with the value true or if the step has only a single output port and primary is not specified. It is a static error to identify more than one output port as primary. On compound steps, the declaration may be accompanied by a connection for the output. It is a static error to specify a connection for a p:output inside a p:declare-step for an atomic step. If a connection is provided for a p:output, documents are read from that connection and those documents form the output that is written to the output port. In other words, placing a p:document inside a p:output causes the processor to read that document and provide it on the output port. It does not cause the processor to write the output to that document.

A p:log element is a debugging aid. It associates a URI with a specific output port on a step: The semantics of p:log are that it writes to the specified IRI whatever document or documents appear on the specified port. If the href attribute is not specified, the location of the log file or files is implementation-defined. How each document or sequence of documents is represented in a p:log is implementation-defined. Pipelines are not expected to be able to consume their own logging output. The ability of a step to read the p:log output of some former step is implementation-dependent. It is a static error if the port specified on the p:log is not the name of an output port on the step in which it appears or if more than one p:log element is applied to the same port. Implementations may, at user option, ignore all p:log elements. This element represents a potential security risk: running unexamined 3rd-party pipelines could result in vital system resources being overwritten.

The p:serialization element allows the user to request serialization properties on a p:pipeline output. If the pipeline processor serializes the output on the specified port, it must use the serialization options specified. If the processor is not serializing (if, for example, the pipeline has been called from another pipeline), then the p:serialization must be ignored. The processor may reject statically a pipeline that requests serialization options that it cannot provide. The default value of any serialization options not specified on a particular p:serialization element is implementation-defined. The allowed options are defined by . It is a dynamic error if the combination of serialization options specified or defaulted is not allowed. Implementations must check that all of the specified serialization options are allowed if they serialize the specified output. If the specified output is not being serialized (because it is being returned as the result of a call from within another pipeline, for example) implementations may but are not required to check that the specified options are allowed. The semantics of the attributes on a p:serialization are described in . It is a static error if the port specified on the p:serialization is not the name of an output port on the pipeline in which it appears or if more than one p:serialization element is applied to the same port.

Variables, options, and parameters provide a mechanism for pipeline authors to construct temporary results and hold onto them for reuse. Variables are created in compound steps and, like XSLT variables, are single assignment, though they may be shadowed by subsequent declarations of other variables with the same name. Options can be declared on atomic or compound steps. The value of an option can be specified by the caller invoking the step. Any value specified by the caller takes precedence over any default value specified in the declaration. Parameters, unlike options and variables, have names that can be computed at runtime. The most common use of parameters is to pass parameter values to XSLT stylesheets.

A p:variable declares a variable and associates a value with it. The name of the variable must be a QName. If it does not contain a prefix then it is in no namespace. It is a static error to declare an option or variable in the XProc namespace. The variable's value is specified with a select attribute. The select attribute must be specified. The content of the select attribute is an XPath expression which will be evaluated to provide the value of the variable. If a select expression is given, it is evaluated as an XPath expression using the appropriate context as described in , for the enclosing container, with the addition of bindings for all preceding-sibling p:variable and p:option elements. Regardless of the implicit type of the expression, when XPath 1.0 is being used, the string value of the expression becomes the value of the variable; when XPath 2.0 is being used, the type is treated as an xs:untypedAtomic. Since all in-scope bindings are present in the Processor XPath Context as variable bindings, select expressions may refer to the value of in-scope bindings by variable reference. If a variable reference uses a QName that is not the name of an in-scope binding, an XPath evaluation error will occur. If a select expression is given, the readable ports available for document connections are the readable ports in the environment inherited by the first step in the surrounding container's contained steps. However, in order to avoid ordering paradoxes, it is a static error for a variable's document connection to refer to the output port of any step in the surrounding container's contained steps. If a select expression is given but no document connection is provided, the implicit connection is to the default readable port in the environment inherited by the first step in the surrounding container's contained steps. If there is no default readable port, the connection is treated as if p:empty was specified. It is a dynamic error if a sequence of more than one document appears on the connection for a p:variable. In an XPath 1.0 implementation, if p:empty is given or implied as the document connection, an empty document node is used as the context node. In an XPath 2.0 implementation, the context item is undefined. It is a dynamic error if the select expression makes reference to the context node, size, or position when the context item is undefined.

A p:option declares an option and may associate a default value with it. The p:option tag can only be used in a p:declare-step or a p:pipeline (which is a syntactic abbreviation for a step declaration). The name of the option must be a QName. If it does not contain a prefix then it is in no namespace. It is a static error to declare an option or variable in the XProc namespace. It is a static error to declare two or more options on the same step with the same name. An option may be declared as required. If an option is required, it is a static error to invoke the step without specifying a value for that option. If an option is not declared to be required, it may be given a default value. The value is specified with a select attribute. If a select attribute is specified, its content is an XPath expression which will be evaluated to provide the value of the option, which may differ from one instance of the step type to another. The select expression is only evaluated when its actual value is needed by an instance of the step type being declared. In this case, it is evaluated as described in except that In an XPath 1.0 implementation, an empty document node is used as the context. In an XPath 2.0 implementation, the context item is undefined. the variable bindings consist only of bindings for options whose declaration precedes the p:option itself in the surrounding step signature; the in-scope namespaces are the in-scope namespaces of the p:option itself. It is a static error to specify that an option is both required and has a default value. It is a dynamic error if the select expression makes reference to the context node, size, or position. Regardless of the implicit type of the expression, when XPath 1.0 is being used, the string value of the expression becomes the value of the option; when XPath 2.0 is being used, the value is an xs:untypedAtomic.

A p:with-option provides an actual value for an option when a step is invoked. The name of the option must be a QName. If it does not contain a prefix then it is in no namespace. It is a static error to use an option name in p:with-option if the step type being invoked has not declared an option with that name. (This error does not apply for steps in the XProc namespace when the processor is operating in forwards-compatible mode.) It is a static error to include more than one p:with-option with the same option name as part of the same step invocation. The actual value is specified with a select attribute. The select attribute must be specified. The value of the select attribute is an XPath expression which will be evaluated to provide the value of the variable. Regardless of the implicit type of the expression, when XPath 1.0 is being used, the string value of the expression becomes the value of the option; when XPath 2.0 is being used, the value is an xs:untypedAtomic. All in-scope bindings for the step instance itself are present in the Processor XPath Context as variable bindings, so select expressions may refer to any option or variable bound in those in-scope bindings by variable reference. If a variable reference uses a QName that is not the name of an in-scope binding or preceding sibling option, an XPath evaluation error will occur. If a select expression is used but no document connection is provided, the implicit connection is to the default readable port. If there is no default readable port, the connection is treated as if p:empty was specified. It is a dynamic error if a sequence of more than one document appears on the connection for a p:with-option. In an XPath 1.0 implementation, if p:empty is given or implied as the document connection, an empty document node is used as the context node. In an XPath 2.0 implementation, the context item is undefined. It is a dynamic error if the select expression makes reference to the context node, size, or position when the context item is undefined.

The p:with-param element is used to establish the value of a parameter. The parameter must be given a value when it is used. (Parameter names aren't known in advance; there's no provision for declaring them.) The name of the parameter must be a QName. If it does not contain a prefix then it is in no namespace. It is a dynamic error to use the XProc namespace in the name of a parameter. The value is specified with a select attribute. The select attribute must be specified. The content of the select attribute is an XPath expression which will be evaluated to provide the value of the variable. The values of parameters for a step must be computed after all the options in the step's signature have had their values computed. If a select expression is given on a p:with-param, it is evaluated as an XPath expression using the appropriate context as described in , for the containing step, with the addition of variable bindings for all options declared in the containing step's signature. Regardless of the implicit type of the expression, when XPath 1.0 is being used, the string value of the expression becomes the value of the parameter; when XPath 2.0 is being used, the value is an xs:untypedAtomic. All in-scope bindings for the step instance itself are present in the Processor XPath Context as variable bindings, so select expressions may refer to any option or variable bound in those in-scope bindings, as well as to any option declared in the step signature, by variable reference. If a variable reference uses a QName that is not the name of an in-scope binding or declared option, an XPath evaluation error will occur. If a select expression is used but no document connection is provided, the implicit connection is to the default readable port. If there is no default readable port, the connection is treated as if p:empty was specified. It is a dynamic error if a sequence of more than one document appears on the connection for a p:with-param. In an XPath 1.0 implementation, if p:empty is given or implied as the document connection, an empty document node is used as the context node. In an XPath 2.0 implementation, the context item is undefined. It is a dynamic error if the select expression makes reference to the context node, size, or position when the context item is undefined. If the optional port attribute is specified, then the parameter appears on the named port, otherwise the parameter appears on the step's primary parameter input port. It is a static error if the specified port is not a parameter input port or if no port is specified and the step does not have a primary parameter input port.

Variable, option and parameter values carry with them not only their literal or computed string value but also a set of namespaces. To see why this is necessary, consider the following step:The p:delete step will delete elements that match the expression “html:div”, but that expression can only be correctly interpreted if there's a namespace binding for the prefix “html” so that binding has to travel with the option. The default namespace bindings associated with a variable, option or parameter value are computed as follows: If the select attribute was used to specify the value and it consisted of a single VariableReference (per or , as appropriate), then the namespace bindings from the referenced option or variable are used. If the select attribute was used to specify the value and it evaluated to a node-set, then the in-scope namespaces from the first node in the selected node-set (or, if it's not an element, its parent) are used. The expression is evaluated in the appropriate context, See . Otherwise, the in-scope namespaces from the element providing the value are used. (For options specified using syntactic shortcuts, the step element itself is providing the value.) The default namespace is never included in the namespace bindings for a variable, option or parameter value. Unqualified names are always in no-namespace. Unfortunately, in more complex situations, there may be no single variable, option or parameter that can reliably be expected to have the correct set of namespace bindings. Consider this pipeline:It defines an atomic step (“ex:delete-in-div”) that deletes elements that occur inside of XHTML div elements. It might be used as follows:In this case, the match option passed to the p:delete step needs both the namespace binding of “h” specified in the ex:delete-in-div pipeline definition and the namespace binding of “html” specified in the divchild option on the call of that pipeline. It's not sufficient to provide just one of the sets of bindings.The p:namespaces element can be used as a child of p:variable, p:with-option or p:with-param to provide explicit bindings. The namespace bindings specified by a p:namespaces element are determined as follows: If the binding attribute is specified, it must contain the name of a single in-scope binding. The namespace bindings associated with that binding are used. It is a static error if the binding attribute on p:namespaces is specified and its value is not the name of an in-scope binding. If the element attribute is specified, it must contain an XPath expression which identifies a single element node (the input connection for this expression is the same as the connection for the p:option or p:with-param which contains it). The in-scope namespaces of that node are used. The expression is evaluated in the appropriate context, See . It is a dynamic error if the element attribute on p:namespaces is specified and it does not identify a single element node. If neither binding nor element is specified, the in-scope namespaces on the p:namespaces element itself are used. Irrespective of how the set of namespaces are determined, the except-prefixes attribute can be used to exclude one or more namespaces. The value of the except-prefixes attribute must be a sequence of tokens, each of which must be a prefix bound to a namespace in the in-scope namespaces of the p:namespaces element. All bindings of prefixes to each of the namespaces thus identified are excluded. It is a static error if the except-prefixes attribute on p:namespaces does not contain a list of tokens or if any of those tokens is not a prefix bound to a namespace in the in-scope namespaces of the p:namespaces element. It is a static error to specify both binding and element on the same p:namespaces element. If a p:variable, p:with-option or p:with-param includes one or more p:namespaces elements, then the union of all the namespaces specified on those elements are used as the bindings for the variable, option or parameter value. In this case, the in-scope namespaces on the p:variable, p:with-option or p:with-param are ignored. It is a dynamic error if the specified namespace bindings are inconsistent; that is, if the same prefix is bound to two different namespace names. For example, this would allow the preceding example to work:The p:namespaces element provides namespace bindings for both of the prefixes necessary to correctly interpret the expression ultimately passed to the p:delete step (the binding for html: is explicitly provided and the binding for h: is in-scope). The use of p:namespaces here, when all of the bindings are provided with explicit namespace declarations, is unnecessary. The bindings could simply be placed on the parent p:with-option element. We use p:namespaces here only to make the example parallel to the one which follows. The preceding solution has the weakness that it depends on knowing the bindings that will be used by the caller. A more flexible solution would use the binding attribute to copy the bindings from the caller's option value.This example will succeed as long as the caller-specified option does not bind the “h” prefix to something other than the XHTML namespace.

A p:declare-step provides the type and signature of an atomic step or pipeline. It declares the inputs, outputs, and options for all steps of that type. The value of the type can be from any namespace provided that the expanded-QName of the value has a non-null namespace URI. It is a static error if the expanded-QName value of the type attribute is in no namespace or in the XProc namespace. Except as described in , the XProc namespace must not be used in the type of steps. Neither users nor implementers may define additional steps in the XProc namespace. Irrespective of the context in which the p:declare-step occurs, there are initially no option or variable names in-scope inside a p:declare-step. That is, p:option and p:variable elements can refer to values declared by their preceding siblings, but not by any of their ancestors. When a declared step is evaluated directly by the XProc processor (as opposed to occurring as an atomic step in some container), how the input and output ports are connected to documents is implementation-defined. A step declaration is not a step in its own right. Sibling steps cannot refer to the inputs or outputs of a p:declare-step using p:pipe; only instances of the type can be referenced. The version attribute identifies the version of XProc for which this step declaration was authored. If the p:declare-step has no ancestors in the XProc namespace, then it must have a version attribute. See . For a description of psvi-required, see . For xpath-version, see . For exclude-inline-prefixes, see p:inline.

When declaring an atomic step, the subpipeline in the declaration must be empty. And, conversely, if the subpipeline in a declaration is empty, the declaration must be for an atomic step. Implementations may use extension attributes to provide implementation-dependent information about a declared step. For example, such an attribute might identify the code which implements steps of this type. It is not an error for a pipeline to include declarations for steps that a particular processor does not know how to implement. It is, of course, an error to attempt to evaluate such steps. If p:log or p:serialization elements appear in the declaration of an atomic step, they will only be used if the atomic step is directly evaluated by the processor. They have no effect if the step appears in a subpipeline; only the serialization options of the “top level” step or pipeline are used because that is the only step which the processor is required to serialize.

When a p:declare-step declares a pipeline, that pipeline encapsulates the behavior of the specified subpipeline. Its children declare inputs, outputs, and options that the pipeline exposes and identify the steps in its subpipeline. The subpipeline may include declarations of additional steps (e.g., other pipelines or other step types that are provided by a particular implementation or in some implementation-defined way) and import other pipelines. If a pipeline has been imported, it may be invoked as a step within the subpipeline that imported it. The environment inherited by the subpipeline is the empty environment with these modifications: All of the declared inputs are added to the readable ports in the environment. If a primary input port is declared, that port is the default readable port, otherwise the default readable port is undefined. If a primary output port is declared and that port has no connection, then it is connected to the primary output port of the last step in the subpipeline. It is a static error if the primary output port is unconnected and the last step in the subpipeline does not have a primary output port. The requested xpath-version must be used to evaluate XPath expressions subject to the constraints outlined in . The psvi-required attribute allows the author to declare that a step relies on the processor's ability to pass PSVI annotations between steps, see . If the attribute is not specified, the value “false” is assumed.

A p:library is a collection of step declarations and/or pipeline definitions. The version attribute identifies the version of XProc for which this library was authored. If the p:library has no ancestors in the XProc namespace, then it must have a version attribute. See . For a description of psvi-required, see ; for xpath-version, see ; for exclude-inline-prefixes, see p:inline. The steps declared in a pipeline library are referred to by their type. It is not an error to put a p:pipeline or p:declare-step without a type in a p:library, but there is no standard mechanism for instantiating it or referring to it. It is effectively invisible. Libraries can import pipelines and/or other libraries. See also .

An p:import loads a pipeline or pipeline library, making it available in the pipeline or library which contains the p:import. An import statement loads the specified IRI and makes any pipelines declared within it available to the current pipeline. It is a static error if the URI of a p:import cannot be retrieved or if, once retrieved, it does not point to a p:library, p:declare-step, or p:pipeline. It is a static error to import a single pipeline if that pipeline does not have a type. Attempts to retrieve the library identified by the URI value may be redirected at the parser level (for example, in an entity resolver) or below (at the protocol level, for example, via an HTTP Location: header). In the absence of additional information outside the scope of this specification within the resource, the base URI of the library is always the URI of the actual resource returned. In other words, it is the URI of the resource retrieved after all redirection has occurred. As imports are processed, a processor may encounter new p:import elements whose library URI is the same as one it has already processed in some other context. This may happen as a consequence of resolving the URI. If the actual base URI is the same as one that has already been processed, the implementation must recognize it as the same library and should not need to process the resource. Also, a duplicate, circular chain of imports, or a re-entrant import is not an error and implementations must take the necessary steps to avoid infinite loops and/or incorrect notification of duplicate step definitions. It is not an error for a library to import itself. An example of such steps is listed in . A library is considered the same library if the URI of the resource retrieved is the same. If a pipeline or library author uses two different URI values that resolve to the same resource, they must not be considered the same imported library.

A p:pipe connects an input to a port on another step. The p:pipe element connects to a readable port of another step. It identifies the readable port to which it connects with the name of the step in the step attribute and the name of the port on that step in the port attribute. In all cases except the p:output of a compound step, it is a static error if the port identified by a p:pipe is not in the readable ports of the step that contains the p:pipe. A p:pipe that is a connection for an p:output of a compound step may connect to one of the readable ports of the compound step or to an output port on one of the compound step's contained steps. In other words, the output of a compound step can simply be a copy of one of the available inputs or it can be the output of one of its children.

A p:inline provides a document inline. The content of the p:inline element is wrapped in a document node and passed as input. The base URI of the document is the base URI of the p:inline element. It is a static error if the content of the p:inline element does not consist of exactly one element, optionally preceded and/or followed by any number of processing instructions, comments or whitespace characters. The in-scope namespaces of the inline document differ from the in-scope namespace of the content of the p:inline element in that bindings for all its excluded namespaces, as defined below, are removed: The XProc namespace itself (http://www.w3.org/ns/xproc) is excluded. A namespace URI designated by using an exclude-inline-prefixes attribute on the enclosing p:inline is excluded. A namespace URI designated by using an exclude-inline-prefixes attribute on any ancestor p:declare-step, p:pipeline, or p:library is also excluded. (In other words, the effect of several exclude-inline-prefixes attributes among the ancestors of p:inline is cumulative.) The value of each exclude-inline-prefixes attribute is interpreted as follows: The value of the attribute is either #all, or a whitespace-separated list of tokens, each of which is either a namespace prefix or #default. The namespace bound to each of the prefixes is designated as an excluded namespace. It is a static error if the exclude-inline-prefixes attribute does not contain a list of tokens or if any of those tokens (except #all or #default) is not a prefix bound to a namespace in the in-scope namespaces of the element on which it occurs. The default namespace of the element on which exclude-inline-prefixes occurs may be designated as an excluded namespace by including #default in the list of namespace prefixes. It is a static error if the value #default is used within the exclude-inline-prefixes attribute and there is no default namespace in scope. The value #all indicates that all namespaces that are in scope for the element on which exclude-inline-prefixes occurs are designated as excluded namespaces. The XProc processor must include all in-scope prefixes that are not explicitly excluded. If the namespace associated with an excluded prefix is used in the expanded-QName of a descendant element or attribute, the processor may include that prefix anyway, or it may generate a new prefix. Consider this example: which might produce a result like this: ]]> The declaration for “c” must be present because it was not excluded. The “part” element uses the namespace bound to “b”, so some binding must be present. In this example, the original prefix has been preserved, but it would be equally correct if a different prefix had been used.

A p:document reads an XML document from a URI. The document identified by the URI in the href attribute is loaded and returned. If the URI protocol supports redirection, then redirects must be followed. It is a dynamic error if the resource referenced by a p:document element does not exist, cannot be accessed, or is not a well-formed XML document. The parser which the p:document element employs must process the external subset; all general and external parsed entities must be fully expanded. It may perform xml:id processing, but it must not perform any other processing, such as expanding XIncludes. The parser must be conformant to Namespaces in XML. Loading the document must not fail due to validation errors. Use the p:load step if you need to perform DTD-based validation. A p:document always reads from the specified IRI. In the context of a p:input, this seems perfectly natural. In the context of a p:output, this may seem a little asymmetrical. Putting a p:document in a p:output causes the pipeline to read from the specified IRI and provide that document as an output on that port. Use p:store to store the results that appear on a p:output.

A p:data reads an arbitrary resource from a URI. The resource identified by the URI in the href attribute is loaded, encoded, wrapped in the wrapper element, and returned as a document. If the URI protocol supports redirection, then redirects must be followed. The value of the wrapper attribute must be a QName. If the lexical value does not contain a colon, then the wrapper-namespace may be used to specify the namespace of the wrapper. In that case, the wrapper-prefix may be specified to suggest a prefix for the wrapper element. It is a dynamic error to specify a new namespace or prefix if the lexical value of the specified name contains a colon (or if no wrapper is explicitly specified). In other words, these two p:data elements produce equivalent wrappers: ]]> but this p:data element will raise an error: ]]> If no wrapper element is specified, the default is c:data: It is a dynamic error if the document referenced by a p:data element does not exist, cannot be accessed, or cannot be encoded as specified. Exactly how the data is encoded depends on the media type of the resource. If the resource has a content type associated with it (e.g., if the resource was retrieved with HTTP), then that content type must be used, otherwise, if the user specified a content-type on the p:data, then that content type should be assumed. If no content type was specified or is associated with the resource, the inferred content type is implementation-dependent. If the media type of the response is an XML media type or text type with a charset parameter that is a Unicode character encoding (per ) or is recognized as a non-XML media type whose contents are encoded as a sequence of Unicode characters (e.g. it has a charset parameter or the definition of the media type is such that it requires Unicode), the data must be encoded as Unicode character sequence. If the media type is not an appropriate text type, or if the processor does not recognize the media type, the content is base64-encoded. The resulting data is wrapped in an element with the name specified in the wrapper attribute (or c:data if no wrapper is specified). Implementations should add a content-type attribute to the wrapper element which indicates the specified or inferred media type of the resource (including any parameters). If the content was base64-encoded, the wrapper must have an encoding attribute which specifies “base64”. If an encoding is specified, a charset may also be specified. The character set may be specified as a parameter on the content-type or via the separate charset option. If it is specified in both places, the values must be consistent. If content-type, encoding, or charset attributes are specified on a c:data wrapper, they must not be in a namespace; if the user-specified wrapper is not c:data, then the attributes must be in the http://www.w3.org/ns/xproc-step namespace. Implementations may record additional details in extension attributes. For example, this p:identity step: might produce output like this: <c:data xmlns:c="http://www.w3.org/2007/03/xproc-step" content-type="text/plain"> AL,Alabama AK,Alaska AZ,Arizona … </c:data> Whereas this pipeline fragment: produces a single img element with an inline graphic (represented using a data: URI): <img src="data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAABmJLR0QA/wD/AP+gvaeTAAAACXBI WXMAAAsTAAALEwEAmpwYAAAAB3RJTUUH1QsEDycHOWjhgQAAAB10RVh0Q29tbWVudABDcmVhdGVk IHdpdGggVGhlIEdJTVDvZCVuAAABjklEQVQ4y42Tv0tbcRTFz3t5kuQFwUVFEYIGU6q4mU0RlCDR qYO2g4gdghUyKSFDwD/AwS2IoIMSHOLgpOLg4CIOWdy76GBoqQUJLlr66aCW6LuoF+7wvefcc3/w vZLgpbe0wOoqn/Ro5TKRVIpLi6tkkpoFpFJUj48JSVI2y4LnBTm+D1paYqa5OQiGw5DN8nV7m1hv L99f4o4Dk5Mc6Pwcd2yMXauL/n5+FAqsWdXjca5LJbokSXt7fOzu5toSaWsLxpqaIJcjr0ZbXGQh HDaWZPjQEGc7O3jPBC4u8EZHOXkruaODPxsbDMuylRWSra3cviYwO0upMcdtfNRqGq/X5esVu7nR BxMol4knEtTfGsH3+bu8zLdnydUqXjrN/nsWKEEiwdXmJp3/BYpFctFokOi6EIsF444DU1McSZLW 1+nq6eGnVWlkhNrcHBXXNUe5z+f5rEyGA8cxD+pufp7BYtH+yhL09fFbkYjd4vQ0W08jFgp8se7l wY3gwAC/KhXanwROTwlNTHBocf8BTQe7F9dc2bMAAAAASUVORK5CYII="/> Some steps, such as p:xquery and p:validate-with-relax-ng, are designed to process non-XML inputs. If a base64-encoded input occurs in such a context, it should be decoded before processing. In this way, for example, an XQuery document can be read with p:data and passed to the p:xquery step without regard to how the data was encoded.

A p:empty connects to an empty sequence of documents.

A p:documentation contains human-readable documentation. There are no constraints on the content of the p:documentation element. Documentation is ignored by pipeline processors. See .

A p:pipeinfo contains ancillary information for steps in the pipeline. There are no constraints on the content of the p:pipeinfo element, see .

Errors in a pipeline can be divided into two classes: static errors and dynamic errors.

A static error is one which can be detected before pipeline evaluation is even attempted. Examples of static errors include cycles and incorrect specification of inputs and outputs. Static errors are fatal and must be detected before any steps are evaluated. For a complete list of static errors, see .

A A dynamic error is one which occurs while a pipeline is being evaluated. Examples of dynamic errors include references to URIs that cannot be resolved, steps which fail, and pipelines that exhaust the capacity of an implementation (such as memory or disk space). If a step fails due to a dynamic error, failure propagates upwards until either a p:try is encountered or the entire pipeline fails. In other words, outside of a p:try, step failure causes the entire pipeline to fail. For a complete list of dynamic errors, see .

Several of the steps in the standard and option step library can generate dynamic errors. For a complete list of the dynamic errors raised by builtin pipeline steps, see .

When handling imports, an implementation needs to be able to detect the following situations, and distinguish them from cases where multiple import chains produce genuinely conflicting step definitions: Circular imports: A imports B, B imports A. Re-entrant imports: A imports B and C, B imports D, C imports D. One way to achieve this is as follows: The step type exports of an XProc element, against the background of a set of URIs of resources already visited (call this set Visited), are defined by cases. The step type exports of an XProc element are as follows: p:pipeline, p:declare-step A singleton bag containing the type of the element p:library The bag-merger of the step type exports of all the element's children p:import Let RU be the actual resolved URI of the resource identified by the href of the element. If RU is a member of Visited, then an empty bag, otherwise update Visited by adding RU to it, and return the step type exports of the document element of the retrieved representation all other elements An empty bag The changes to Visited mandated by the p:import case above are persistent, not scoped. That is, not only the recursive processing of the imported resource but also subsequent processing of siblings and ancestors must be against the background of the updated value. In practice this means either using a side-effected global variable, or not only passing Visited as an argument to any recursive or iterative processing, but also returning its updated value for subsequent use, along with the bag of step types. Given a pipeline library document with actual resolved URI DU, it is a static error if the step type exports of the document element of the retrieved representation, against the background of a singleton set containing DU as the initial Visited set, contains any duplicates. Given a top-level pipeline document with actual resolved URI DU, it is a static error if the bag-merger of the step type exports of the document element of the retrieved representation with the step type exports of its children, against the background of a singleton set containing DU as the initial Visited set, contains any duplicates. Given a non-top-level p:pipeline or p:declare-step element, it is a static error if the bag-merger of the step type exports of its parent with the step type exports of its children, against the background of a copy of the Visited set of its parent as the initial Visited set, contains any duplicates. The phrase "a copy of the Visited set" in the preceding paragraph is meant to indicate that checking of non-top-level p:pipeline or p:declare-step elements does not have a persistent impact on the checking of its parent. The contrast is that whereas changes to Visited pass both up and down through p:import, they pass only down through p:pipeline and p:declare-step. The bag-merger of two or more bags (where a bag is an unordered list or, equivalently, something like a set except that it may contain duplicates) is a bag constructed by starting with an empty bag and adding each member of each of the input bags in turn to it. It follows that the cardinality of the result is the sum of the cardinality of all the input bags.