https://bugs.documentfoundation.org/show_bug.cgi?id=172902
lechinski <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |NEW Ever confirmed|0 |1 Assignee|[email protected] |[email protected] |desktop.org | --- Comment #2 from lechinski <[email protected]> --- Confirming on master (27.2.0.0.alpha0, ec75111a57a2, Linux) and on 26.8.1.1; also seen in production with 25.2.3. Not ODT specific: same with a DOCX source, converted headless to DOCX and to PDF. Synthetic DOCX with N distinct 4000x3000 JPEGs, soffice --headless --convert-to, peak RSS of soffice.bin (VmHWM), master: photos -> docx -> pdf 20 1257 MiB 1328 MiB 50 2414 MiB 2582 MiB 94 4112 MiB 4332 MiB Roughly 39 MiB per photo, i.e. every image stays decoded. Setting GraphicAllowedIdleTime=0 alone changes nothing on master (94 -> docx: 4114 MiB), because nothing triggers a reduction: since 324f2e135427 MemoryManager reduces memory only from maSwapOutTimer, which never fires without an event loop. Patch restoring the synchronous reduction in registerObject(), as before 24.8, with a unit test: https://gerrit.libreoffice.org/c/core/+/213290 With the patch and GraphicAllowedIdleTime=0, 94 photos -> docx peaks at 1213 MiB. With the default of 10 s, a short conversion barely improves, since few graphics have been idle long enough. Tomaž: would it make sense to ignore GraphicAllowedIdleTime (or use a much shorter one) when running headless / without an event loop? I kept that out of this patch so it's a pure regression fix, but I'm happy to do it as a follow-up if you agree. -- You are receiving this mail because: You are the assignee for the bug.
