We can't call the callbacks (as the comments correctly indicated),
but we can certainly delete the transfer objects, since they
rightfully belong to us.
Should fix #19242, #19241, #19180 and possibly other recent regressions.
Also fixes a long-standing memory leak from this scenario.
This reverts commit 6211ae2575.
It breaks multiple applications that worked fine on BeOS,
so clearly this tweak was correct after all, or at least mostly correct.
Change-Id: I3b228ba2119dd33067b8defedd6db4ae9ead9124
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8537
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Haiku-Format: Haiku-format Bot <no-reply+haikuformatbot@haiku-os.org>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Based on X512's original change, but with more modifications:
* Don't use _ in macro parameter names to appease GCC 2.
* Get rid of ioctl_args struct. We don't need it anymore,
and just adding parameters of the same types has the
exact same ABI on x86 and x86_64, so this doesn't break
any existing compiled code.
* Add (void*) cast to third parameter.
* Enable for _KERNEL_MODE also.
Change-Id: Id4ad8b85f54836fd26dc6278226954d0a081d5f0
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8533
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
This will allow unistd.h to define it even in _KERNEL_MODE.
Change-Id: I6669c670353dda7cc202017d707a6bce3b5519e0
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8561
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
PadBlocker wasn't handling key repeats, so once key repeat started
kicking in, it would register it as keypress and consequently block any
B_MOUSE_DOWN event that followed.
Fixes #19238 and possibly #17821.
Change-Id: I1dccca5a31021866a61aa95bbeecfb0ba37facac
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8532
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Reviewed-by: Axel Dörfler <axeld@pinc-software.de>
Left top position will be set to zero.
It is a common pattern to define `BRect` value with only width and height so it allow to simplify code a bit.
Change-Id: Ie14644843324c9e5bcc55e7cfbd557a8884559d3
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8535
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Avoid declaring random friend classes in public header. Allow to access
private methods from arbitrary source if needed.
Change-Id: Iac2cf0ca59e483aa0657e3fe1fc47080c661cf8b
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8534
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Continued work to normalize the data model
in the application.
Change-Id: I6dc6d17cff866b18f56431056958092c8bdc3a19
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8524
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Two minor behavioral changes: EINVAL is now returned on invalid
times in more places, and clock_settime now rounds up instead of down.
Otherwise all should be as before.
* A new icon to indicate native apps in HaikuDepot. It's simply the
Overlay_leaf on a blue disk.
* Slight change to HaikuDepot_Installed to add a black outline to its
green disk. That makes is better distinguishable on similar coloured
backgrounds.
* Made the green of HaikuDepot_Installed a bit less glaring, using a
slightly light green that shown in the old Haiku Color Palette [1]
[1] https://www.haiku-os.org/files/downloads/2007-03-20_haiku-color-palette.png
Change-Id: I5c29d65e05561fafe5d84c173ddaf083281d391b
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8526
Reviewed-by: Adrien Destugues <pulkomandy@pulkomandy.tk>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
We reschedule the interrupt transfers every time they succeed,
so we need to cancel them on failure to attach in order to not
leave one lingering (and later KDL).
Fixes #17549 and related tickets.
This way, modules can decide to do different things based on having
all the ancillary data available. In particular, the UNIX module will
now post only one message header for all the FDs, even if they came
from multiple sets of ancillary data.
This should fix "Message needs unreceived descriptors" from the Chromium
IPC code (which is used by Firefox).
Same as for the scatter/gather pool. This also changes CCBs
to use condition variables instead of semaphores for completion,
which means they're now C++-only. A few C files still include
<SCSI.h>, but none use CCBs directly, so this works out fine.
A quick check with a compile benchmark didn't show a performance regression.
All object_caches use locked memory by default; and in fact
the CACHE_UNLOCKED_PAGES flag isn't even implemented at present.
So the locked_pool system is redundant here.
The passed request is the one triggering the auto-sense, not
the auto-sense itself.
Resetting the "buffered" flag on the wrong request meant that
if a buffer had been used for the triggering request, it would
be leaked, and subsequently we would get KDLs when trying to
unlock memory that wasn't actually locked at all.
Fixes #19218 and related tickets.
