Status report: consultancy team, week 47, 2023

Status reports
1

Caolan

  • we need at least 20G to link wasm
  • fight to get wasm to serve through nextcloud
  • try NextCloud 28 which has the option to enable
    wasm
  • add Julius patch to richdocuments to serve the
    necessary headers
  • seem to need to add extra Cross-Origin-Resource-Policy
    at least to some places to have a chance looking
    at the various failure warnings
  • wasm appears to at least start, but fails to open
    a document. Progress.
  • seem to be missing ACCESS_TOKEN etc stuff, presumably
    that should be available in the html that launches wasm
  • We generate the cool.html from cool.html.m4 and the
    wasm case is currently like the mobile case and we strip
    all that out?
  • leave it in, tweak things to pass everything we need to
    the wasm, add more server headers around the place
  • try with wasm “forced” on and modified nextcloud
    → seems to work
  • non wasm case then fails to work at all
  • Could not parse URL ReferenceError: RPMGetInnerMostURI is
    not defined
    → if the richdocuments server headers needed for wasm are
    enabled, then it looks like online has to have matching
    headers for the non-wasm case too, doesn’t need
    script-src:unsafe-eval unsurpringly I guess, but does seem
    to need
    Cross-Origin-Opener-Policy: same-origin
    Cross-Origin-Embedder-Policy: require-corp
    Cross-Origin-Resource-Policy: cross-origin
    to avoid cryptic NS_ERROR_DOM_CORP_FAILED error visible
    in firefox network page on non-wasm failed load
  • firefox works with wasm from nextcloud, but chrome doesn’t
  • only using a http not https locally, possibly relevant
  • update wasm/optimization estimates
  • merge various outstanding patch reviews
  • bisect oss-fuzz crash to latest armin change
    → https://gerrit.libreoffice.org/c/core/+/159685
  • Demo up a “Long-Run Profiling Script” with the intent to get
    length profiles on demo.eu
    → roughly 100M an hour of data with:
    perf record -F1 --call-graph dwarf,65528
    so approx 16G for 7 days and there is 100G available on demo.eu
    Plausible. demo.eu account has aws privs to email productivity-dev
    already for crashreports. Test a short run from console, then
    another from crontab
  • set something to run for 6.9 days and start every 7 days
  • try some other things with wasm and chrome and nextcloud
  • debugging missing tiles on reconnection
  • maybe we have two problems
    → setting tileid to 1 doesn’t look right to me?
    → viewids, from dark/light mode setting?, look odd too
  • Lets merge what wasm stuff we have to online master and try with a
    real install on perf-staging rather than mess around with a
    self-build

Mike

  • Exporting multiple consecutive spaces to ReqIF-XHTML
    • I implemented that according to our import code - we do honor a space
      before
      in ReqIF; but not a space after
      ; so now we emit
      ‘xml:space=“preserve”’ in the latter case, but not the former
    • Explained that in the commit message
  • Improvements for the LanguageTool integration
    • The fix itself seems easy, but how to test?
    • Moreover: which characters should be encoded? RFC 3986 defines
      Reserved Characters to include both ‘@’ and ‘+’; but LT needs one of them
      encoded, while another may be unencoded. No clear specs … Reading some
      Java code.
    • OK, settled on encoding all reserved - it also works, and ‘@’ is
      accepted in both forms
  • Tried to repro “Slow Saving Due to Bookmarks / References”
    • Created a macro
    • The document with 1000 bookmarks saves normally, using both LO 7.6.3.2
      and CO 23.05.4.2.
    • Asked for a reproducer.
  • LanguageTool: tdf#156662 - not using proxy
    • A bit of research - it is also a problem in other places
    • Filed tdf#158352
  • Unexpected content when exporting image anchored to paragraph
    • Working / debugging / minimizing…
    • Caused by a wrap setting
    • sw/source/filter/html/htmlflywriter.cxx, see
      OOO_STRING_SVTOOLS_HTML_O_clear
    • Also there’s a bug in HtmlWriter::characters - made
      https://gerrit.libreoffice.org/c/core/+/159932

Tomaz

First-page header and footer from and to DOCX:

  • Move header/footer tests common class in a separate commit
    • easier to see what needed to change
  • Got most test fixed
    • one issue is that export now writes even, odd and first files into the
      OOXML
    • generally not a problem as the files are empty or identical
    • but need to find the reason why this happens later
  • Only one problem persists with uiwriter test, which crashes
    • but only when the next test is run
    • seems like something statically allocated is not cleaned correctly for
      some reason
    • need to find out the reason

Miklos