qt6-bb10

History

Marc Mutz ca216b7bed QDuplicateTracker: shed memory_resource on clear() The pmr::monotonic_buffer_resource ("pmr::mbr") strategy can only allocate buffers; deallocation is a no-op. I chose it for QDuplicateTracker because the tool was only used one-off: build it, then destroy it. There were no other operations. That changed when `090c7e3262` added clear(). Calling pmr::unordered_set::clear() will deallocate the nodes (but not the bucket array), but due to pmr::mbr, nothing is actually deallocated, so memory that was freed by the container isn't actually reused by it, later. IOW: repeated grow-clear-grow-clear cycles would forever grow the monotonic_muffer_resource, using more and more memory and thus appear to be leaking it. This isn't exactly optimal behavior, so try to shed the pmr::mbr's extra memory with a call to mbr::release(). Unfortunately, C++17 originally failed to nail down the semantics of pmr::mbr::release(), prompting LWG 3120 in response to observed implementation divergence. In particular, MSSTL, at the time of LWG 3120 filing, didn't appear to allow to allocate() after release(). In our tests, it does allow it, it just never falls back to the original (stack) buffer, and, looking at the implementation, it doesn't look like it can do so anytime soon (it doesn't remember the buffer, and everything is inline, so they would seem to need a BC break to fix it. We tried to work-around the problem by hard-resetting 'res' by going through a destroy-recreate cycle. pmr::mbr is neither copy- nor move-constructible, so that blunt instrument is, unfortunately, necessary (std::optional would be an alternative, but has overhead that affects all platforms). This crashed, though, so just call release() and handle MSSTL's failure to reuse the initial buffer as a QoI issue. For all platforms, before release(), we also need to make sure that `set` no longer holds any references into it, and, since clear() doesn't shed capacity, we need to use the C++11 version of the swap-trick (swap is actually UB here, because the allocators differ): move-assign a default-constructed set. Amends `090c7e3262`. Fixes: QTBUG-132945 Pick-to: 6.5 6.2 Change-Id: I4796806e427602255439dcb1518aa9b661c7933e Reviewed-by: Thiago Macieira <thiago.macieira@intel.com> (cherry picked from commit 9f4325e67354ce6c4c98e7a206f17729b378dc04) Reviewed-by: Qt Cherry-pick Bot <cherrypick_bot@qt-project.org> (cherry picked from commit d2c4f6df21629495ce6fe32653037b35a989d6f4)		2025-01-31 14:06:20 +00:00
..
auto	QDuplicateTracker: shed memory_resource on clear()	2025-01-31 14:06:20 +00:00
baseline	test: baseline: Call finalizeAndDisconnect	2024-11-22 20:18:50 +00:00
benchmarks	Fix wrong value for 'noon' in tst_bench_qdatetime	2025-01-20 09:11:02 +00:00
global	…
libfuzzer	Complete color space toICC write	2024-05-31 16:24:50 +02:00
manual	Make iconbrowser manual test directly openable in Creator	2025-01-22 23:04:14 +00:00
shared	De-duplicate the disabling of crash dialogs in our unit tests	2025-01-20 21:30:42 +00:00
testserver	Add REUSE.toml files	2024-11-07 08:38:49 +01:00
CMakeLists.txt	Rid of 'special case' markers	2023-04-13 18:30:58 +02:00
README	…

README

This directory contains autotests and benchmarks based on Qt Test. In order
to run the autotests reliably, you need to configure a desktop to match the
test environment that these tests are written for.

Linux X11:

   * The user must be logged in to an active desktop; you can't run the
     autotests without a valid DISPLAY that allows X11 connections.

   * The tests are run against a KDE3 or KDE4 desktop.

   * Window manager uses "click to focus", and not "focus follows mouse". Many
     tests move the mouse cursor around and expect this to not affect focus
     and activation.

   * Disable "click to activate", i.e., when a window is opened, the window
     manager should automatically activate it (give it input focus) and not
     wait for the user to click the window.