COOL TC meeting minutes: 2026-06-03

Development
1

Date: 2026-06-03

Present:

  • Regina, Andras, Stephan, Adria, Noel, Caolan, Andras, Leif, Tomaz, Miklos

Completed action items

Pending action items

  • File a github issue for the canvas / slideshow cleanup (Thorsten)

Release Engineering update (Andras)

  • Monorepo status

    • CI: release branch, moving from github to gerrit: done

    • CI, still to be added on main: --enable-coplugin, fuzzer, lighthouse

      • ~~done, but needs a final check (Adria+Andras)

    • then github online.git: refer to gerrit

  • Releases (Andras)

    • CODE 26.04: cp-26.04.1-2 is tagged already

    • this week: RC3/final, to be tagged later today

      • development can continue on main branch

      • distro/collabora/co-26.04 branch will be created in case we have to release again for CODE 26.04.1 (and also for CODA)

Patch review

  • All changes

    • Week 23: 196(-24)

  • PRs to migrate from GitHub

    • Week 23: 57(-10)

  • New contributors since last week

    • Week 23: Chamodh Nethsara, David D., Fawlok, Poesty Li, Spell, Tomiri EXPT, fsadfsdfs, mnsk, scudo, teste celular, yiweoi yaw

  • Top 10 reviewers since last week

    • Week 23: Miklos Vajna (78), Noel Grandin (58), Caolán McNamara (50), Mike Kaganski (31), Stephan Bergmann (27), Tomaž Vajngerl (19), Andras Timar (15), Michael Stahl (15), Sarper Akdemir (12), Tor Lillqvist (7)

Bug reporting

Jenkins / CI update (Szymon)

  • Mohit’s monorepo test change

  • Szymon’s monitoring: “last builds” monitoring

    • need to update tool, it doesn’t work now (api call returns 403)

  • Gerrit for online main

    • Week 23: Success rate is 50%, failed builds are 25/50

  • Getting more HW from Hertzner? (Michael M)

    • Pricing may be OK – the hardware is smaller – and needs more setting up

      • 96 thread machines

    • Do we want to go there?

      • Yes, probably (Andras)

    • Currently an average job doesn’t use all the cores (Andras)

    • Lots of cores means multiple containers, not all containers should use all cores

    • There is a setup cost there, maybe bring them up when the queue is long? (Michael M)

      • Sounds like a good direction (Andras)

      • Especially if the node bring-up is automated

      • AI: look at this (Andras)

  • Latency is much better than it was a week ago (Noel)

    • more on 25.04 – not yet ported to gerrit.

    • But we did ask people to not push stuff to clear the queue for release fixes.

  • Optimizing re-building ?

    • Using the more parallel cypress building thing that Noel got up & running (Caolan)

      • plan to enable this for everyone – 5-6 minutes off the cypress build

      • just one coolwsd – and running all the cypress targets vs. that – saturates it.

      • Stresses the server more too which is good (Michael M)

    • ccache – is per-workspace (?)

      • some things so that a cypress & non-cypress build can be used (?)

      • we have a task for this – which may speed things up.

    • Currently doing lots of header file changes currently too (Michael M)

      • perhaps that reduces ?

    • Cypress is failing much less often & flatpak is quicker too (Noel)

    • Once Andras gets shared ccache together for flatpak this will help.

  • Leave it a week and lets see (Michael M)

Crash testing (Caolan)

  • 2026-05-31 278 (-269, +0)

    • crashtesting: defer dynamic-array resize while measuring optimal row height fixed, hoping to hit the bulk of those now

  • 2026-05-05 547 (-471, +286)

TTT talks

UTF8 strings for OUString (Noel)

  • Shared – profiling / research worth doing first … (Stephan)

    • tried to find if it has much of an impact on heap data in a kit process.

    • Looked like it doesn’t have a huge win – a small %age – depending on use-case

      • a large doc with many eg. Thai strings – may be bigger

      • even though lots of strings are internal ascii-only tokens

    • questioning the rationale …

  • Started looking at heap areas that might be more beneficial to look at

    • forkit process + kits forked from it – should share lots of data pages

      • but these get dirtied in one of the processes.

      • That can touch something.

      • Strings are not the main thing that dirties them

    • increasing the number of shared pages / keeping them lower.

      • Strings don’t dirty pages.

    • Very hard to get a good view of the cost of things in the COW domain

  • Had some perf profiles of do_wp_page system call in the past

    • 50Mb on fork without doing much – so a tad surprising.

    • configmgr → some sub-case that was not captured.

    • Can’t think of other big wins from the probe.

css: re-work instead ? (Michael M)

  • Not sure of all the places it goes (Stephan)

    • not just going into the language-bindings

    • the component names – at the UNO level – the .component things too

    • may be places where one gets computed from the other.

    • Lets suck it and see …

  • what instead

    • api:: instead (Quikee)

    • co:: instead? (Miklos)

    • Apple has ‘NS’ prefix for Nextstep heritage (Stephan)

    • cpoffice (Noel)

    • css – as an alias is just three characters (Miklos)

      • benefit of api:: - same length as css::

      • if we move from CO → something else – don’t re-write again.

    • Might collide with C++ things, too generic (Noel)

      • not google well.

    • CPO == cp::office (Noel)

    • no particular care one way or the other (Caolan)

    • Lets go with cpo:: (Michael)

What’s cooking

  • Regina

    • Watching SmartArt development

    • Started to collect a list of problems seen on staging-perf

    • Also a post to the forum in the Development category

    • Not new problems, but good to address them

    • Talking to Armin here

    • Also have some sample GLOX files

      • Will link them in the document, hosted on staging-perf

      • Will check with Tomaz, then put the link up in the forum

  • Stephan

    • collaboration, scripting

  • Andras

    • CODA: Options dialog and snap development

    • CODE: release related tasks, bugfixing

  • Quikee

    • Vector rendering: getting more primitives in

      • adding primitives now backed by tests so everything works nicely

      • after this – start looking at making it work for slides itself

      • Text primitives first – not doing web-fonts yet – leave to the end.

      • Currently its switched off in the product: &vector=true to the URL

        • will not disturb plans for the stable release (Andras)

    • Calc matrix formulas: UI improvements

    • Bringing the solver to COOL Calc, requires async dialogs

      • To be in the next COOL in a month

      • Need to make it more accessible first

  • Noel

    • loplugin:stringliteral and removing various bits we don’t need anymore

    • removing bits we don’t need anymore …

  • Miklos

  • Caolán

    • mystery failure of a cell being edited by one (human) user to redraw while another (automated) user constantly churns edits. On another tab, same zoom. calc-edit.ods

    • completed a11y coverage for co-26.04

    • link feature effectively completed, follow up tidy ups and tweaks in ci

  • Adria

    • Activate gerrit jobs for co-25.04, prepare agents and cleanup old workspaces

    • tracking still open PRs

  • Michael M

    • empty-document bits – getting rid of cursors before first tiles show up

    • Trying to get ruler dragging to look pretty with mouse-over tooltips idly.

    • Calc sleeping / empty cell fade-in background

Collabora Online Weekly Meeting #259