Now one can open another document after closing the first one. Turns
out that throwing an exception to return through the call stack to
SocketPoll::poll() is not necessary after all. But doing
document.reset() as before a86508d815 is
essential.
Change-Id: I248df78bd9b0d3f0962df2126ca394cb746542b8
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90456
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Tor Lillqvist <tml@collabora.com>
But opening a second document now hangs.
Sigh, the plumbing in the mobile apps is so extremely fragile. But
that is to be expected when turning a multi-process structure (where
one class of processes exit as soon as they have done their job) into
a single process running forever.
Change-Id: I0fdb751f44e16efb42843189969e049bf14816f0
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90443
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Tor Lillqvist <tml@collabora.com>
Downloading pdf file does not work because the kit process is not jailed
Change-Id: I1e67840eb58997f6de10948c8d8e260888abe326
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90338
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Henry Castro <hcastro@collabora.com>
The 'exit' command from DocBroker to Kit is processed
immediately. It set the termination flag and destroyed
the Document instance in Kit. Unfortunately, this is
highly problematic. There are a number of races here.
Normal disconnection procedure from DocBroker is to
first send 'disconnect'. This tells Kit to unload
and close the view in question, and when 0 views
are left, it simply exits the process.
However, since 'disconnect' is processed in the
message queue, the 'exit' (and indeed the socket
disconnection that follows 'exit') may be handled
before the message queue is drained, and so the
'disconnect' wouldn't have been processed yet.
The end result of these two races is that Kit
tries to exit the Core main loop, which fails
a number of assertion (f.e. that no LOK Windows
remain, which is presumably a leak check, which
we care not about as we don't attempt to cleanup
the sidebar).
The fix here is to process 'exit' immediately,
because DocBroker is gone, there is absolutely
nothing for Kit to do anymore.
Change-Id: I5c09fcfdb1713f4e0b56b717c747d919d0c6728f
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90356
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
Essentially we want to be able to separate low-level socket code
for eg. TCP vs. UDS, from Protocol handling: eg. WebSocketHandler
and client sessions themselves which handle and send messages
which now implement the simple MessageHandlerInterface.
Some helpful renaming too:
s/SocketHandlerInterface/ProtocolHandlerInterface/
Change-Id: I58092b5e0b5792fda47498fb2c875851eada461d
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90138
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Open Writer, insert a multi-MP JPEG, select it.
Before:
ChildSession::renderShapeSelection: finished in 74.67 ms
After:
ChildSession::renderShapeSelection: finished in 67.33 ms (90.17% of baseline)
Change-Id: I495e9fcf0c42b0bcfc7987a402a13f5d8664b1e1
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90291
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
Open Writer, insert a multi-MP JPEG, select it.
Before:
ChildSession::renderShapeSelection: finished in 81.33 ms
After:
ChildSession::renderShapeSelection: finished in 74.67 ms (91.81% of baseline)
This is with an --enable-symbols core with a -O2 online, with libstdc++.
The cost on the Android profile with its libc++ looked even more,
spending time in the std::vector dtor.
Change-Id: I50af2e13fd24569dc32304420b8f3e70d15803eb
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90262
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Allows comparing tokens with C strings without a heap allocation. Do the
same when comparing two tokens from two different StringVectors.
And use it at all places where operator ==() has an argument, which is a
StringVector::operator []() result.
Change-Id: Id36eff96767ab99b235ecbd12fb14446a3efa869
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90201
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Sometimes it is very useful to have one "lokit" process,
to focus on a 100% reproducible bug, and not worry
that server pre-spawn several processes.
Change-Id: I414a8145b53a0601a282cba9c245833f5d07f404
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/89999
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Henry Castro <hcastro@collabora.com>
WebSocketHandler handles this complexity for us now, and for the
forseeable future. Simplify to prepare for larger re-factor.
Change-Id: I73b919885adc358cb6502a13492cdac85c34459c
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/90059
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
This is meant to reduce lots of small allocations and instead have
pointers into the single string for the various tokens instead.
This has a few requirements, though:
1) It's no longer OK to modify the tokens, changing their length would
invalidate the start/length of other tokens. Rework
DocumentBroker::load() to avoid such mutation.
2) The iterators no longer expose zero-terminated strings, so
Poco::cat() doesn't work anymore: add an own cat() instead and use that
in e.g. ChildSession. The own cat() has the benefit that it won't read
past the end of the array if the begin index is out of bounds to add
more safety.
(This nicely works towards killing Poco usage in general.)
3) If zero-terminated strings for all individual tokens is needed, a
copy has to be made, as done in spawnProcess().
(For all of these requirements, the build fails if there are problems.)
Change-Id: Iea40e4400e630b2d669f5c72aea85cb40edf9a2c
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/89711
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
The bulk of this commit just changes std::vector<std::string> to
StringVector when we deal with tokens from a websocket message.
The less boring part of it is the new StringVector class, which is a
wrapper around std::vector<std::string>, and provides the same API,
except that operator[] returns a string, not a string&, and this allows
returning an empty string in case that prevents reading past the end of
the underlying array.
This means in case client code forgets to check size() before invoking
operator[], we don't crash. (See the ~3 previous commits which fixed
such crashes.)
Later the ctor could be changed to take a single underlying string to
avoid lots of tiny allocations, that's not yet done in this commit.
Change-Id: I8a6082143a8ac0b65824f574b32104d7889c184f
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/89687
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
Particularly configuration layers so we can tweak mobile config
easily.
Add core source files from configmgr for breakpointing convenience in
the iOS project. Add loolkitconfig.xcu to the iOS app bundle. Use
${BRAND_BASE_DIR} instead of a compile-time LOOLWSD_CONFIGDIR literal
on iOS (because there is no compile-time constant path to the app
bundle). No "registry" directory directly in the app bundle any longer
on iOS, a corresponding change in core.git moved that stuff to be
under "share", like on other platforms.
Change-Id: I6672efc0505abf27297c4758118a20992b10ceb3
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/88765
Tested-by: Michael Meeks <michael.meeks@collabora.com>
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
This dates back to 2016 and our pre non-blocking and pre-unipoll state.
It is no longer necessary - a single thread reads all data from the
socket and feeds events into the Kit process; much cleaner.
Change-Id: I46ad6806a1e0cdbb0e5cf4ea5d3e5e5078d3391a
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/88741
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Amazingly more time was spent allocating, wiping and freeing these
vectors on Android than the actual rendering of pixels in a profile.
Change-Id: I49ea093816eba0f4187613ab6c8dc24d8dcba75b
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/86335
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Use a fully reliable uniqueness check, rather than a hash, and get
simpler ids as a bonus. Fetch view data from the session itself
rather than passing it in too.
Change-Id: Ibcd625156b5a98eb280e35d6537b5c8c026d0197
Reviewed-on: https://gerrit.libreoffice.org/c/online/+/86150
Reviewed-by: Mert Tümer <mert.tumer@collabora.com>
Tested-by: Mert Tümer <mert.tumer@collabora.com>
It is painful to check and search manually the PID to attach the LOKit
process when exists several pre-spawned waiting to load a document.
This patch helps to attach the debugger when the LOKit process is about
to load a document then send the "signal SIGUSR1" to resume it.
Change-Id: I3b15bd522c6ef3ef57dc3453b457dcf91f2661b9
Reviewed-on: https://gerrit.libreoffice.org/85430
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Henry Castro <hcastro@collabora.com>
Reviewed-by: Henry Castro <hcastro@collabora.com>
The failing scenario was reproducible on my old phone:
* start editing the document from the shell
* add a word
* leave the editing using the '<' in the top left
* start editing the document again => boom!
In the normali, non-crashing case, the order of destruction is that
~ChildSession is called first (and calls ChildSession::disconnect() that
consequently calls Document::onUnload() via _docManager->onUnLoad(...).
The Document::onUnload() then deregisters callbacks and all is fine.
The in the above described crashing case, the Document is destroyed
before ChildSession and calls resetDocManager() for all sessions.
Before this patch, this meant that the _docManager was set to nullptr,
leading to situation that later, in the ~ChildSession, the call of
onUnload() was avoided, the callbacks were not unregistered, and later,
on the next document load, the app crashed because it called stale
callbacks.
I suspect the change might be useful even for the non-mobile case, but
not 100% sure, so rather do it MOBILEAPP-only.
Change-Id: I279a160ccaab3080e84fe0437ed72684331b6e13
Reviewed-on: https://gerrit.libreoffice.org/84588
Reviewed-by: Jan Holesovsky <kendy@collabora.com>
Tested-by: Jan Holesovsky <kendy@collabora.com>
lok clients can request to complete a function name partially typed in
the formula input box.
Change-Id: If8e4485c5ed9f91a594dfcec04e0c0b10becdcd0
Reviewed-on: https://gerrit.libreoffice.org/83985
Reviewed-by: Marco Cecchetti <marco.cecchetti@collabora.com>
Tested-by: Marco Cecchetti <marco.cecchetti@collabora.com>
View switching should not cause the sidebar UX to re-build at all. So
pass the view-id into the render call so we can avoid this.
Change-Id: I194c1841df2cedfcdf82defb2712d20715d7ff1d
Reviewed-on: https://gerrit.libreoffice.org/84131
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>
This patch is a follow-up for overflowing
watermarks patch
Change-Id: I8613736dc4fea49a22ae29aba209c95b1ea93557
Reviewed-on: https://gerrit.libreoffice.org/83635
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>
There are a few things acceptable/safe in a
signal handler, and taking locks is not one of them.
This replaces the logic with a simple counter that
serves the purpose just as well.
If we get a double signal, we log and ignore.
Change-Id: If589c18492468c120d00c213805467bcbba05d27
Reviewed-on: https://gerrit.libreoffice.org/83150
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
Based on information from UNO ViewRowColumnHeaders messages, this draws
the grid lines with Leaflet's SVG polylines, in a background pane.
To prevent visual glitches on touchscreens, this also reverts parts of
26aad36771ebd5a304b763912184acd2ea26dfdf - that introduced a regression
that prevented 'dragend' events from being fired when inertia-dragging.
Instead, inertia drag is disabled in the L.Map constructor.
[ Miklos: cherry-pick these early as they prevent automatic merging. ]
Change-Id: I86c68214822a0a2c2f8ced82ccf5a6c201e2cf04
Particularly those used on Android...
Change-Id: I47bf9692f5e99ba30140d698558472201168a198
Reviewed-on: https://gerrit.libreoffice.org/82302
Reviewed-by: Jan Holesovsky <kendy@collabora.com>
Tested-by: Jan Holesovsky <kendy@collabora.com>
The following flags are affected:
ShutdownRequestFlag
TerminationFlag
DumpGlobalState
Since it's common to grep for all places
that set or reset these global flags, it
makes more sense to have explicit functions
for each operation. Now we have set and reset
accessors where appropriate and get is reserved
for read-only access.
This changes the getters to only return
the boolean value of these flags rather than
a reference to the atomic object, now that
they are read-only.
Also, a few Mobile-specific cases were folded
either with other Mobile-specific sections, or
they were now identical to the non-Mobile case
and therefore deduplicated, making the code
cleaner and more readable.
Change-Id: Icc852aa43e86695d4e7d5962040a9b5086d9d08c
Reviewed-on: https://gerrit.libreoffice.org/81978
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
Add a "singleton" class method to DocumentViewController to return the
(as for now) singleton DocumentViewController.
Change-Id: I0b8a8def558cfe7f9469b6062a86311dfa63f549
Reviewed-on: https://gerrit.libreoffice.org/82007
Reviewed-by: Tor Lillqvist <tml@collabora.com>
Tested-by: Tor Lillqvist <tml@collabora.com>
(cherry picked from commit 2807f907d947549a17c5bae586b85d412d552a09)
Reviewed-on: https://gerrit.libreoffice.org/82021
Sometimes core renderes with sub-pixel differences
(the crosshair at the corners of the Writer pages
show line anti-aliasing differences). This causes
failure of the tests that count the tile deduplication.
We now tolerate when we get an unchanged tile twice,
assuming it was due to such a rendering difference,
but we re-trigger another change and this time we
don't expect any extra tiles, no more than two
variations of the anti-aliased crosshair was
observed.
We also move some duplicate code into utility
functions to improve readability and reuse.
Reviewed-on: https://gerrit.libreoffice.org/81196
Reviewed-by: Andras Timar <andras.timar@collabora.com>
Tested-by: Andras Timar <andras.timar@collabora.com>
(cherry picked from commit 9603597fd1aaecb27893792cfd2d243e450b58b8)
Change-Id: I1a66732dd3443bfbd770d8dc65721571dfa08615
Reviewed-on: https://gerrit.libreoffice.org/81572
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>