Caolan
- asan warning about missing mutex
→ lock getWakeupsArray for reading by caolanm · Pull Request #10365 · CollaboraOnline/online · GitHub - measure SharedString visitor for ref
counting reduction improvements, update, etc - autotext, experiment with trying to get autotext
files put into the jail seen/working at all - cool#10177 debug a little to see why worker thread decompression
seems laggy - debug with pranam why there is a crash with a redlined field
→ https://gerrit.libreoffice.org/c/core/+/175808 - autotext isn’t read from where one would initially
expect- i.e. a /tmp/user/user/config/autotext dir from
presets in desktop/source/app/userinstall.cxx
with userinstall::finalize
which uses utl::Bootstrap::locateUserInstallation(uri)
but we actually attempt to read from
/tmp/lue3h7e1.tmp/user/autotext
where the default autotext paths to search for come from
SubstitutePathVariables::SetPredefinedPathVariables
aState = utl::Bootstrap::locateUserData( sVal );
because preloadData sets a tmp user installation dir, and
that one gets stuck as the dir for later loads
from SynchronizeExtensionRepositories,
to comphelper::EventLogger::EventLogger
to SubstitutePathVariables singleton ctor
- i.e. a /tmp/user/user/config/autotext dir from
- howto find where the .ui warning comes from and how to
fix same - cool#10332 uncaught exception from NetworkInterface::list
→ NetworkInterface::list can throw by caolanm · Pull Request #10385 · CollaboraOnline/online · GitHub - autotext
- there are two bootstraps, rtl and unotools, unotools syncs
at SECOND_INIT with reloadData - we do have a post-preinit effort to set the bootstrap
variables to the expected dir, but PathSettings and
SubstitutePathVariables already exist pointing to the
tmp preload/preinit locations
- there are two bootstraps, rtl and unotools, unotools syncs
- Allow “F3” to get sent to core to trigger autotext
Comparing master...private/caolanm/kit_presets · CollaboraOnline/online · GitHub- but it get sent twice, which seem like something that happens
with all Function keys, bodgeable with:
→ prevent double-dispatch of uno events bound to function keys by caolanm · Pull Request #10403 · CollaboraOnline/online · GitHub
- but it get sent twice, which seem like something that happens
- autotext
- Add XInitialization support to PathSettings so it could be
reinitialized after preinit so it updates its autotext etc paths
to where they should be - get SubstituteVariables to update what path variables it
substitutes when this resync of bootstrap settings happens - tweak debug-mode kit jail creation to also install a
sample autotext - writer continue to fail to do anything sensible
- writer will refuse to access a user’s personal usertext
unless there happened to be an internal/shared autotext
which is an odd situation, apparently ancient null deref
workaround
→ https://gerrit.libreoffice.org/c/core/+/175959 - installing sample autotext now works
- next to explore checkFileInfo and demo up getting user autotext
etc from server to stick into jail. - readonly shared “org” autotext could either be put into the
kit user autotext dir as a special case, or bind-mounted over
where lo puts its “internal” one, the latter might be a better
fit for the overall idea of layered forkits. Worry about that
after getting user stuff in place.
- Add XInitialization support to PathSettings so it could be
Mike
- Calc: Misaligned shape/chart/image positioning with browser zoom
- Two-side: core + Online
- Created initial core change, fixing some rounding errors affecting placement
- Working on Online side
- Broke grid rendering (found the problem)
- The selection is misplaced
Tomaz
Analyze and close gaps to PDF-2.0
- We can output documents identifying as as PDF-2.0
- but they may not be valid PDF-2.0 documents
- a lot of features that were deprecated, which shouldn’t be written into PDF-2.0
- /Info dictionary has for the most part been deprecated
- exceptions: /CreateDate and /ModDate
- XMP metadata should be used instead
- no need to keep them in sync anymore (requirement with previous versions)
- changed the code to not write those elements, but write XMP metadata instead
- we have written XMP metadata only when we had PDF/A or PDF/UA, but it is supported from PDF 1.4
- I have enabled XMP metadata for all versions from PDF 1.4 on
- All the old encryption methods have been deprecated - PDF 2.0 has an update one
- AES-256 V6
- We use the ultra old RC4, but with 128bit key length
- Still, not good enough today
- The AES versions use a completely different algorithm.
- Implementing that.
- Factored out CryptTools (wraps NSS or OpenSSL) from oox module (used for encryption there) and put it into comphelper
- Hash - for SHA hashing was already moved there some years ago
- WIP