This fixes a bug where the window size limits were not set correctly
causing the window not to be hidden properly in some cases while
Deskbar is in auto-hide mode. This bug was introduced in hrev53585:
Update window resize size limits.
A couple of other auto-hide related bugs were also fixed:
Hide TBarView in constructor if auto-hide is on. This is needed to
size and position the window correctly on Deskbar startup in auto-
hide mode.
Always Check fTime->IsHidden() from the perspective of fTime instead
of the parent view because we were getting false positives that the
clock was hidden in auto-hide mode which caused the replicants not to
realign themselves around the clock on Deskbar startup. The clock
thought it was hidden because the parent view was hidden but that's
not what is needed here.
Bail out of BarView::MouseMoved if resizing. This fixes a bug where
if you resized the window in auto-hide mode once the window had become
as wide as possible dragging beyond the window hidden area slop limit
would confusingly cause the window to hide itself in the middle of your
resize operation.
Fixes#15067 better. Fixes problems related to #8641 and #9469.
Change-Id: I58de02e0cdd4e4cdccc15594992f11bf8c7f3a26
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2252
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
This way, it is now actually impossible to submit a request to a
pipe that is in the process of being destroyed, solving several
race conditions.
Should fix the remaining KDL in #15710.
In d49942867b (part of hrev52813),
I moved invocation of the TERM_BEFORE/TERM_AFTER functions to
runtime_loader. Unfortunately, I only updated one location where
they should be invoked, and runtime_loader had two.
After this commit, there is one function that they are called from,
which is itself invoked from the appropriate places.
Noted by ttcoder & mmlr in #14047.
* As we were clearing out the endpoint structures before detaching
the xhci_endpoint from the Pipe, it was possible for SubmitRequest
to be called while we were being destroyed and then use the half-
torn-down endpoint, leading to crashes. Instead, we now call
SetControllerCookie first, so that SubmitRequest will thus
fail.
* Lock the endpoint after calling StopEndpoint, for the same
reasons that this is now the order of operations in CancelQueuedTransfers.
Probably fixes#15710, and maybe other issues.
It was only actually used to determine if a device was disabled
or not, which can be determined by checking the "slot" field
is "0" instead, as 0 is an invalid slot number.
No functional change (intended.)
On modern x86, one can use __rdtscp to get the current cpu in userland.
Change-Id: I1767e379606230a75e4622637c7a5aed9cdf9ab0
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2248
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
Change-Id: Idc50cfee16db5f7c0dbea625a7739ad42b2d7bf0
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2249
Reviewed-by: Alex von Gluck IV <kallisti5@unixzen.com>
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
* "xorriso doesn't support much of the mkisofs options"
* Dropping the hfs map made our bootloader no-longer get
mapped as :tbxi which means qemu + Apple devices refused
to boot it.
* Switch to genisoimage which can generate the correct image.
(xorriso -as genisoimage doesn't work either)
Change-Id: Id71b97796f5004f6ddf6244644c51a5787f9bf8e
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2241
Reviewed-by: Alex von Gluck IV <kallisti5@unixzen.com>
Both made many valuable contributions.
Thanks, Kyle and Ilya!
Change-Id: Iaf819f37319df45050c6892ddc2066901f93f7eb
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2247
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
This workarounds the start delay for sound output.
This can be reverted when sound output is available right after startup
tested on emulation.
Change-Id: I2efb5d46ab25dcda15602bba70e728ae9e34537a
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2238
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
hrev53379 clears the buffer cache for disconnected clients, and also delete buffers.
This is too early (see #15263, media_addon_server crash), and should only happen
after the buffer is recycled. This can be resolved by abusing the fFlags field of
BBuffer to mark the buffer for deletion, and mark the buffer to be reclaimed.
Some BBuffers don't reside in the SharedBufferList, so we have to mark them as to
be reclaimed. For those in the SharedBufferList, call a new RemoveBuffer(), which
can check whether the buffer is still to be reclaimed. For reclaimed BBuffers,
delete them right away, others can be marked for deletion.
fixes#15606#15263, possibly #15433
Change-Id: I66e94138e7e10a40d4c48e2ac042f816c79f5aab
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2245
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
Reviewed-by: X512 <danger_mail@list.ru>
Reviewed-by: Ryan Leavengood <leavengood@gmail.com>
This patch is part 1 of 3 with the goal of having a working
integration test harness for BHttpRequest. In this patch the existing
test cases were expanded and fixed for HTTP. In followup patches the
test harness will be updated to support HTTPS and reverse proxies.
Before this patch the tests for BHttpRequest had hard dependencies on
the external services httpbin.org and portquiz.net. These tests
eventually stopped working because the owner of those services made
changes, causing the assertions in these tests to fail.
The goal of these patches is to make a test harness that allows for
the same kinds of end-to-end integration tests but without any
external dependencies.
The test suite now includes a Python script called testserver.py which
is a HTTP echo server of sorts. When it receives a request, it will
echo the request headers and request body back to the client as a
text/plain response body.
The TestServer class manages the lifecycle of this testserver.py
process. Each test case calls Start() on the server to start a new
instance, and then it is shut down when the destructor is called. On
each invocation a random port is assigned by the kernel in TestServer,
and that socket file descriptor is provided to the child testserver.py
script.
Authorization tests are supported, currently implementing Basic and
Digest auth. If the test server receives a request for a path
/auth/<auth-scheme>/<expected-username>/<expected-password>, then the
appropriate authorization scheme will be employed. For example, if
/auth/basic/foo/bar is used as the path, then the server will expect
the Authorization header to contain an appropriate Basic auth
payload.
The tests now perform a bit more validation than before, validating
the expected HTTP headers and response body is returned from the
server.
The following tests are not fixed yet or were removed:
* PortTest was removed entirely since I'm not sure of the point of this
test, and that functionality seems to be covered by the existing tests
anyway.
* HTTPS tests are not functional yet, but will be in a followup
patch. THis requires updating testserver.py to generate a
self-signed TLS cert if --use-tls is provided.
* ProxyTest was disabled before this patch, but can be enabled in a
followup patch by providing a reverse proxy in the test harness.
Change-Id: Ia201ef4583b7636c61e77072a03db936cb0092be
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2243
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
StartWatching() must be called for each notification type.
Change-Id: I34957af526a0af557a86eef0c3de5722f3503ca5
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2244
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
'Restart media services' in Media Preferences works better.
Change-Id: Ifbffdbd81ee851ae3e7d3dfd14f3d5f41ac155ce
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2240
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
Move NULL check for 'fReply' and 'fRequest', since they might be NULL.
Change-Id: Icc42b8f24f406d6752c25f4203d6ebe3f6ba0d97
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2168
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
Rework the layout of the installer window, do not try to put everything
in a single grid but instead use a mix of grid and groups as
appropriate, which keeps things simpler.
Make the width of the window more font sensitive.
Have the package view properly resize with the window, and the items in
it still draw properly (checkbox and name on the left, size on the
right)
The menu fields for source and destination don't need to extend to the
whole window width, so allow them to be smaller.
Fixes#5785
Change-Id: If0683f610b9e7f3629d51d25c1d8e8b00c101156
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2185
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
KPaths are most commonly used in the VFS syscall paths,
and so they are typically user_memcpy'd to/from userland.
In the "to" case this is not really necessary (but it should
be so small of a performance difference as to not matter),
but in the "from" case, we must always clear the buffer
we received from the allocator, so as not to leak information
to userland.
Part of #14961.
It is unused, and it leads to an unexpected entry in the Input
preferences.
Change-Id: I68f31c3ae6baca7be1a5d6070dc1ebcf04522dad
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2208
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
This is used in Find window and also in Installer.
Remove some unused code (I think leftovers from Tracker InfoWindow
refactoring).
Change-Id: Ic0dd07e06c11b3839adbe5b8ef9598a5b16171a6
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2234
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
The driver had been using 16 bit reads and write (introduced by Axel in
2006), but the hardware documentation (released in 2008) says the
register is 32bit. Use 32bit since that makes the code identical to
what's used for later devices.
If 16bit is needed, it's only for the pre-2008 devices.
If a transfer completed before the StopEndpoint did,
HandleTransferComplete would try to lock the endpoint,
resulting in a deadlock, and so StopEndpoint would
always time out in this case.
May help with #15161, #15141, #15215, #15416, #15657.
It seems that many controllers do not report status changes following
the hand-off from the BIOS, so we were never probing those ports
and reporting the devices attached to them.
The BSDs do something similar (although it seems to only do this
on the first explore, not always.)
This fixes booting off of USB3 ports from most Intel controllers,
as well as a variety of others with similar behavior.
Fixes#15000.
OpenSSL says we should retry when a non-blocking read finds no data is
pending. But in that case we should not retry immediately, because the
operation should be non-blocking.
Before the refactor to use Event Data TRBs, this was necessary
to detect short packet conditions. Now that we use Event Data
TRBs, the hardware will send us an Event Data event even on
Short Packet conditions, so this is redundant.
Fixes log spam of "TRB ... not found in the endpoint", as
this was causing duplicate event postings. Also potentially
fixes instability seen when those messages were present.
There is one efi_block_io_protocol per disk and one per partition.
All we need to do is find the disk ones and let Haiku find
bootable partitions.
There is a special case for a device with one fixed partition
which does not have one for disk, but it is unlikely we will
ever want to boot from such a device.
Fixes#15587.
Change-Id: I915870d6d3b19947bc58b32a969f9f89d2d2245d
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2232
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
And remove Mouse, Keyboard and Touchpad.
Userguide and localizations will need to be updated.
Change-Id: I4543b2b63367cd13562c542610bad34b5934b103
Reviewed-on: https://review.haiku-os.org/c/haiku/+/2210
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
