Caolan
- some memory wins to reduce the size of each kit
→ https://gerrit.libreoffice.org/c/core/+/183015
→ https://gerrit.libreoffice.org/c/core/+/183074
→ https://gerrit.libreoffice.org/c/core/+/183076 - mem leak in coolwsd in make run
- make run mode fd leak by caolanm · Pull Request #11367 · CollaboraOnline/online · GitHub
- tiny mem leak with --with-infobar-url
→ fd leak with --with-infobar-url by caolanm · Pull Request #11376 · CollaboraOnline/online · GitHub - SIGUSR1 crashes on demo.eu
→ DocumentBroker::dumpState seen to crash by caolanm · Pull Request #11379 · CollaboraOnline/online · GitHub - high cpuload on demo.eu could be side effect of LOG_INF and
smap dumping when there is a lot of memory used? - check what the “resting” size of coolwsd should be, dumpState
numbers don’t seem to correspond to anything close to what is used- dump FileHash content capacity in dumpState
- 200M in local build is from FileHash
→ Private/caolan/explain some unaccounted memory by caolanm · Pull Request #11383 · CollaboraOnline/online · GitHub
- use after free on DocumentPoll
→ extend DocumentBroker lifecycle until preset callbacks no longer need it by caolanm · Pull Request #11386 · CollaboraOnline/online · GitHub for
that theory - ofz html use after free
→ https://gerrit.libreoffice.org/c/core/+/183193 - TTT on profiling, including adventures in tcmalloc to see live
allocations - play around on demo.eu with tcmalloc and coolwsd to see if there
are insights to be got that way
→ Ubuntu 18.04 tcmalloc puts a lot into LazyFree/MADV_FREE
Chris
-
cool#11304 – Cache tile canvases (merged)
- Follow-up: cool#11364 – Don’t cache null canvases in TilesMiddleware.CanvasCache
-
cool#11371 – Fix double resize on CanvasSectionContainer
-
Investigated, worked on and submitted cool#11389 - Fix TileManager.garbageCollect
-
Reviews: cool#11358, cool#11387
Mike
- Loading times for Writer documents increases nonlinear for many bookmarks in tables
- Had a very mysterious compiler bug
- Estimated
Tomaz
OAuth 2.0 support for webdav
- Implementing the HTTP server using boost::beast
- kind-of complex as it’s async, which makes the execution less deterministic
- issue with shutdown, which we want to do after we get the necessary information
- make server code separate, so it can be tested independently
- put into tools
Miklos
Worked on Track changes: ability to reinstate a change · Issue #11357 · CollaboraOnline/online · GitHub