Caolan
- tdf#159046 hang on export of math ole objects in calc
- this looks like the reason for the hang on export seen in a cool
call a while back
→ https://gerrit.libreoffice.org/c/core/+/161742
- this looks like the reason for the hang on export seen in a cool
- cool#7951 part 1 of not updating when only some settings change
to reduce invalidations when joining in calc - get wasm rebuilt and reworking for demo
- “Sessions on demo servers eat up memory and die” poke around at what might have changed with running out
of memory + SIGABRT etc - “Sessions on demo servers eat up memory and die” debug bizarro-land where cat /proc//smaps_rollup
gives different results than it appears online reads - try a bunch of different things
- in the end standalone program reproduces the same thing when
rewind and re-read the file on demo.eu, which doesn’t happen
locally - looks like a kernel bug; I suspect to be fixed by:
mm: /proc/pid/smaps_rollup: convert to single value seq_file · torvalds/linux@258f669 · GitHub
included in >= v4.19 and demo.eu is 4.15
→ Rewinding and rereading proc/smaps_rollup doubles the values on some … by caolanm · Pull Request #8002 · CollaboraOnline/online · GitHub - work on making a regression test for calc join invalidation
- tdf#159086 fix some treeview model/view mixup seen by allotropia
- “Investigate failing jobs in Tinderbox” tinderbox failure, since Oct, oops
→ fix error TS2694: Namespace 'cool' has no exported member 'TilesSecti… by caolanm · Pull Request #8007 · CollaboraOnline/online · GitHub - See if share can do long-term profiling
- need about 15G, it has 65G free at the moment, which
looks like a big enough margin to me - do a short run to see if it can email the profile, that
works, check with Andras if eating that 15G for profiling
is ok - ponder calc tab switching invalidations
- help out with a 24.2 upstream regression
→ 159138 – Hyperlink dialog is too large when there is a long text in the clipboard - launch profiling on share
→ first run for 5 days - next runs will be automatic to run 6 days a week, leaving thu
free for ad-hoc profiling of cool session etc. - I see that if you use ‘perf record -o foo’ then you can’t
get anything useful out of foo until recording is finished, but
with ‘perf record -o i > foo’ then you can work with foo while
recording is underway (https://lwn.net/Articles/381634/) which
looks like it could allow getting incremental flamegraphs while
still accumulating profiling data - experiment with ditching invalidations while switching calc tabs
- why not?
→ https://gerrit.libreoffice.org/c/core/+/162049
Mike
- “validity List shows only one item”
- Data -> Validity -> List · Issue #6956 · CollaboraOnline/online · GitHub
- Seems to send control characters unescaped; LF here
- Fixes: #6956 Escape payload strings sent in JSON by mikekaganski · Pull Request #7974 · CollaboraOnline/online · GitHub
- Problems with failing CI all around: core master, core 23.05, Online …
- https://gerrit.libreoffice.org/c/core/+/161812
- “Calc: Column and row headers get stuck on hover state”
- Repro
- ColumnHeader / RowHeader’s drawHeaderEntry
(browser/src/control/Control.RowHeader.ts, Control.ColumnHeader.ts) - Turns out that it wasn’t intentional “hover state”: the boldened text
is because the fillStyle used in fillRect is fully transparent- Then the anti-aliased text gets over-drawn, and the “half-drawn”
pixels get brighter.
- Then the anti-aliased text gets over-drawn, and the “half-drawn”
- CanvasSectionContainer has clearCanvas; it uses clearRect - use it,
too?- Doesn’t work, makes it black
- Detect fully transparent case, obtain clearColor from parent
- Make sure to clear headers' background by mikekaganski · Pull Request #7987 · CollaboraOnline/online · GitHub
- Oh - Pedro noticed, that
Fix control header entry highlighting by get-state · Pull Request #7960 · CollaboraOnline/online · GitHub handles the same - It even overlaps where mouse handling is concerned; and it makes
background non-transparent. - Reviewed and merged
- Oh - Pedro noticed, that
- “Missing display update when setting start value
of numbered list”- Regression after
cd3c16fbcb4f8e5e4c4448bc7cda96e8476d6aec - core - Gitiles - No repro in master; SwNumRule::SetInvalidRule obviously changed in
between - 154864 – Changing starting number of numbered list does nothing
- backported, tested, answered
- Regression after
- tdf#159050
- Fly portions must consume hole portions’ width…
- A fix is ready
- https://gerrit.libreoffice.org/c/core/+/161910
- “Improvements needed to DOCX fidelity”
- An “all-in-one” problematic document
- call with Miklos on this topic
Tomaz
The mouse doesn’t get activated in the cell that it needs to be
activated:
- Implemented to change the GridView window size in the LOK call to
SetClientVisibleArea- Seems this works fine as we always have a client area for a reasonable
area for a view - Needed to make sure the conversion from to pixel size is done correctly
- Seems this works fine as we always have a client area for a reasonable
- Test for the issue (revert to the initial state)
- As in the test we don’t call paintTile it is already wrong for the first
view- only paintTile sets the GridView size - to the document size
- Use a document with wide first column so it is easy to hit outside the
800px position (default size of the GridView) - Send a mouse down/up event - left size, right side of the cell
- Check what cell got select, which should always be the first cell in the
first wide column - Also implement checking in the second view - just to make sure that one
also works correctly - Was able to reproduce the issue in the test - nice
- The solution with the SetClientVisibleArea solved the issue, but we need
to set the client area in the test
- As in the test we don’t call paintTile it is already wrong for the first
- Changed the paintTile to not set the GridView window size
- as now the size is set in the SetClientVisibleArea and it should be
relatively sane value - testing it seems that there rendering tiles doesn’t seem to be
influenced - but not safe to push into a stable branch
- needs more extensive testing
- as now the size is set in the SetClientVisibleArea and it should be
Miklos
Worked on: