29501 2016-02-23 16:34:12 +0000 [xp31] syntactic problems around the colon of a MapConstructorEntry 2016-04-25 21:34:53 +0000 1 1 1 Unclassified XPath / XQuery / XSLT XPath 3.1 Candidate Recommendation PC All RESOLVED FIXED P2 normal --- 1 mike jonathan.robie andrew_coleman jmdyck josh.spiegel public-qt-comments oldest_to_newest 125237 0 mike 2016-02-23 16:34:12 +0000 The list of non-terminal symbols in A.2.2 does not include colon, which is present in the syntax of MapConstructor. It would seem to make sense to make it a delimiting terminal symbol, and to treat it like "." and "-" as an exception case (see last para of the section) where a symbol separator is needed if the character follows an NCName. Whether it also needs a symbol separator if it follows a prefixed QName is an open question. These choices would invalidate a number of test cases recently added to the MapConstructor test set: for example MapConstructor-029 uses map{*:b:b} where not only the parser but also the human reader can otherwise be easily misled into an incorrect reading. 125238 1 josh.spiegel 2016-02-23 16:44:56 +0000 I agree that a constraint should be added to require a separator when ":" follows NCName. It seems ":" is already a delimiting symbol - it is identified in the list as "(colon)". However, I don't see why it should require a separator if ":" follows a QName. If we don't do this, then MapConstructor-029 should still pass. i.e. map{*:b:b} Is equivalent to: map{(*:b) : b} 125239 2 mike 2016-02-23 16:48:48 +0000 Benito pointed out that I didn't read carefully enough. The list does include ":" (written out as (colon), for some reason). The other comment, about treating it like "." and "-", remains. We do have a note in 3.11.1.1 Note: In some circumstances, it is necessary to include whitespace before or after the colon to ensure that this grammar is correctly parsed; this arises for example when the MapKeyExpr ends with a name and the MapValueExpr starts with a name. But it's a Note, it's very unspecific about what the exact rules are, and it's in the wrong place. 125240 3 mike 2016-02-23 17:00:21 +0000 MapConstructor-029 is actually quite interesting, because the wildcard *:b is not one terminal symbol, but three (even though whitespace is not allowed). If it were one terminal symbol, then it would be chosen in preference to "*" by the "longest token" rule. But it isn't, so one could legitimately argue: Given the input *:b:b (a) the first token is unambiguously "*" (b) the second token is unambiguously ":" (c) the third token can be either the NCName "b" or the QName "b:b". Both are "consistent with the EBNF" in the sense that there is a grammatical production that accepts this token as a continuation - a Wildcard accepts the NCName, while a MapConstructorEntry accepts the QName. The QName is longer, so this should be chosen, meaning that the construct parses as map{*: (b:b)} 125242 4 josh.spiegel 2016-02-23 17:47:56 +0000 Yes, it seems that simply requiring a separator when colon follows NCName also breaks this wildcard. foo:* 125312 5 josh.spiegel 2016-03-01 19:36:56 +0000 Setting aside "how" for a moment, I would like the specification to clarify the meaning of these queries: (1) map{a:b:c} (2) map{*:b:c} (3) map{a:*:c} The W3C applets currently identify these key expressions: (1) a:b (2) *:b (3) a:* My implementation gives the same answers. 125314 6 mike 2016-03-01 21:21:33 +0000 I agree that if these three expressions are allowed, then they should have the meanings described. I was very surprised by the discovery that the "longest token" rule currently leads to *:a:b parsing as * : a:b But I think my preference is probably to impose restrictions requiring the use of whitespace - I think that reduces the risk of users writing something they didn't intend. Again, I'm not sure exactly what the rule should be. Part of it is probably that (like "-" and ".") ":" as a token is not allowed immediately after an NCName (including an NCName that is part of a QName). That leaves questions about ":" after "*". We might find that the grammar works better if we define wildcards using composite tokens "*:" and ":*", or even if we make wildcards (like QNames) into single tokens. 125315 7 josh.spiegel 2016-03-01 21:55:34 +0000 > We might find that the grammar works better if we define wildcards using composite tokens "*:" and ":*", I like this. It would be a simple fix and have a small editorial impact. > or even if we make wildcards (like QNames) into single tokens. My implementation arrives at the answers in comment 5 because it already does this. I don't this distinction mattered until map constructors were added (notice, ws:explicit is in effect for Wildcard). 126041 8 andrew_coleman 2016-04-22 08:31:08 +0000 At the meeting on 2016-04-19, the WG discussed the options presented in https://lists.w3.org/Archives/Public/public-xsl-query/2016Mar/0035.html (action A-635-03). It was decided to adopt option 2b. The change to appendix A, the applet and the test cases will be tracked by action A640-02