Hello! I'm unable to join the Zoom -- it is asking for a password

Same problem for me

<Vlad> Sorry, our IT folks changed the setup. Will post new link here

<Vlad> or try this password: 834384

<Garret> Hey, does anyone know the password for the zoom meeting?

<scribe> scribeNick: Persa_Zula

updates on action items

<myles> https://github.com/w3c/PFE-analysis/pull/35

myles: noticed some things worth calling out. 1. even without changes that he added - patch subset - compiles 601 files. -- range request method has 2 files

Vlad: is it because of harfbuzz?

myles: that's one of the 10 dependencies

Garret: harfbuzz accounts for most compilation work

myles: harness was created differently - range request operates on glyphs. the form of the corpus is a set of codepoints - can't faithfully convert from one to the other

Garret: hoping to publish with glyph IDs with the codepoints
... when converting the datasets, we're running the text from a shaper to get the glyph sets

myles: for this PR, ran the codepoints through the cmap table - fine for CJK, but probably not fine for other languages
... the corpus associates a set of character with the font - codepoints to font tuples -- doesn't work for the range request corpus. had to add a default font that would be used for un-tagged characters.
... have some preliminary results -- not final results.
... (sharing PFE initial results graphs)
... total cost function - range request is more than double of patch-subset method. but still compared to what we have today is much better

Vlad: has this data been compared with the output of the previous runs of the framework?

myles: not yet

Garret: whole font method is so bad that we had to put a cap. cost function is capped for whole font which is why it looks similar to unicode range
... we'll have to look at the cost function - 2G results look wrong

Vlad: can you send that to the list?

myles: will send PDF to the list with preliminary results
... confidence in the cost function is very low, we need to look at that before we decide these results are accurate
... not sure that the optimal results are straightforward, we may need to re-review it

Garret: we might want to try to have both ideas for optimal subset shown

myles: last piece on the harness -- we need experimental validation to validate the relationship between the request graph and the total time that's gathered. some of the assumptions might not be right
... things that need done on Pull #35: 1. when calculating what the range request should be, it's done without actual bytes of each glyph, can't get good gzip compression ratio

Garret: brotli is available in the framework

myles: file compression is not build into it yet. not sure whether it would help or not
... bar graphs shown are on a small corpus. when ran with a large corpus, patch-subset was failing
... will sync with Garret to debug

<myles> https://github.com/litherum/StreamableFonts

<Vlad> https://docs.google.com/document/d/1aShjR0-UVoQcu1R9GGPF_Qa3bViOdqYVlk7HM07IdGU/edit

Garret: summary of PFE simulation results doc
... Version 2.3.7 of harfbuzz is released and a new version of results will be sent out after analysis is re-run

Vlad: timeline for remaining development

Garret: biggest piece we're missing is the dataset -- should be able to get those in about 2 weeks if things go well
... for patch/subset, there is some optimization to explore -- not critical but would make it more fair in analyzing costs
... publishing the dataset with glyph ids - working on this internally for publishing it

<scribe> ACTION: Garret to 1. publish dataset with glyph ID data

<trackbot> Created ACTION-221 - 1. publish dataset with glyph id data [on Garret Rieger - due 2020-06-15].

<scribe> ACTION: Garret to 2. explore optimizing patch-subset approach

<trackbot> Created ACTION-222 - 2. explore optimizing patch-subset approach [on Garret Rieger - due 2020-06-15].

<scribe> ACTION: myles to integrate updates into streaming framework

<trackbot> Created ACTION-223 - Integrate updates into streaming framework [on Myles Maxfield - due 2020-06-15].

This is scribe.perl Revision of Date 
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00)

Succeeded: s/dbeug/debug/
Present: jpamental Persa_Zula myles Garret
Found ScribeNick: Persa_Zula
Inferring Scribes: Persa_Zula

WARNING: No meeting chair found!
You should specify the meeting chair like this:
<dbooth> Chair: dbooth

Found Date: 08 Jun 2020
People with action items: garret myles

WARNING: Input appears to use implicit continuation lines.
You may need the "-implicitContinuations" option.


WARNING: IRC log location not specified!  (You can ignore this 
warning if you do not want the generated minutes to contain 
a link to the original IRC log.)