| Age | Commit message (Collapse) | Author |
|---|
|
Some of the exclusions were too aggressive. Restrict them to only the
important classes, which exposes some more places this plugin applies.
Change-Id: I1b2d1fb24391adc71ed0984f94168f61a149479f
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/165154
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
|
|
Change-Id: I024653154c51389bb27f3e94b422ff7fc1c9b46b
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/135296
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
|
|
As reported by Julien, once the mediawiki extension is installed, the
xHandler in framework::DispatchProvider::implts_searchProtocolHandler()
points to an UNO component implemented in Java, and we crash in
dynamic_cast<>(), at least on Linux with gcc 7.5 and libstdc++.
This dynamic_cast<>() call was added in commit
c0fa456436947a5c167c652d19a884064b43c03d (tdf#149261 sdext: fix crash on
starting the presenter console for the 2nd time, 2022-05-26), to allow
the presenter console to opt out from protocol handler caching. It was
expected that the proxy object created for a Java UNO component would
simply return nullptr when we try to dynamic_cast<>() it down to a C++
interface.
Fix the problem by moving the interface to an UNO one: this side-steps
the dynamic_cast<>() crash at the price of introducing an UNO interface,
which is not meant to be part of the public UNO API (but at least it's
not published).
It may still make sense to improve the bridges/ code at some stage to
not crash in dynamic_cast<>() on generated Java proxy objects.
Change-Id: Iaac44515339e0dc15dddc3be45ef7dee7331e47a
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/135114
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
Reviewed-by: Caolán McNamara <caolanm@redhat.com>
Tested-by: Caolán McNamara <caolanm@redhat.com>
|
|
This started with commit 3f768cddd28a2f04eb1ffa30bed4474deb6fbfc4
(framework: avoid re-creating protocol handler instances all the time,
2022-05-02). In case there are 2 monitors, then one monitor shows the
slideshow, the other shows the presenter console. The presenter
console's protocol handler in
sdext::presenter::PresenterProtocolHandler::Dispatch::Dispatch() has
mpPresenterController->GetWindowManager() as an empty reference on the
2nd time the presentation starts.
The above commit started to cache protocol handler instances at a frame
level for performance reasons, and this is meant to be safe in general,
but the presenter console's window manager is re-created between
slideshow runs, so it depends on framework/ code to re-create the
protocol handler all the time, which is problematic here.
Fix the problem by introducing a framework::CacheInfo interface that
allows protocol handler implementations to signal if they want to avoid
being cached.
This should be good enough for now, but if later it turns out that there
are too many broken protocol handlers out there, then we can consider
flipping the default and only cache handlers which explicitly opt in for
this behavior. This is not done in this commit.
Change-Id: Ic159813b1b339540bc8c4e780c4d6d7d2d4d2445
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/135020
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
Tested-by: Jenkins
|
|
Once you install an extension that adds its own protocol handlers (e.g.
<https://github.com/niocs/ProtocolHandlerExtension>), DispatchProvider
re-creates this protocol handler every time the custom menu gets opened
or a command gets dispatched.
This allows the dispatch provider to avoid managing the lifecycle of
those protocol handler instances, but in case the constructor of those
handlers is expensive, this leads to performance problems.
Introduce a map of handler instances in DispatchProvider to avoid
unnecessary re-creation and re-initialization: these instances get the
same XFrame anyway (the DispatchProvider is owned by the frame), so this
is meant to be safe.
No testcase for this -- the problem is only visible if you have an UNO
service registered in the global UNO service registry, but by the time
our cppunit tests run, that is already a fixed list, so this would be
hard to test from code.
Change-Id: I6d69906a795a2d5a67706002d635b6cb3091b856
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/133706
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
Tested-by: Jenkins
|
|
Change-Id: Ied2683a0b8a1bab1a7594da1e9bdbd3cb753552c
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/124370
Tested-by: Jenkins
Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
|