This patch concerns the case where we successfully create a read-only directory
with FILE_DELETE_ON_CLOSE. Currently we return STATUS_CANNOT_DELETE (having
cleared O_CREAT). This patch would change that to return success.
However, tests show that this is incorrect. Actually FILE_DELETE_ON_CLOSE should
always fail on a read-only file, including when that file is created by the same
call.
These were committed without any explanation, but they do seem to fix Cygwin
unlink(), and it's been observed that many patches in Wine-Staging, whether
labeled or not, were specifically for Cygwin.
The relevant functionality was fixed a different way upstream, namely:
91e442b060
There is no record at all of what this was for.
However, it's not hard to make a good guess. The effect of the patch is to skip
walking the TEB chain if the faulting %esp is *above* the first element in the
TEB chain.
The most obvious thing this protects against is the case where the application
switched stacks and the new stack happened to be at a higher address. Without
this patch, we would walk through the whole TEB chain, since all of its entries
would be below the target frame we are unwinding to. But they wouldn't actually
be "inner to" it, and so we'd incorrectly hit Wine try/catch blocks.
The most notable such try/catch block is the unhandled exception filter itself,
and it would necessarily have been triggered by any such exception if no other
blocks were.
One can further speculate that this patch, like many others in Wine-Staging, was
written for Cygwin, which is known to switch stacks.
Besides Wine commits c22aa54e9977785eafcd7cc3811116e5f4dd2da8, and other more
targeted workarounds to specific functions, the workaround introduced by this
patch was obviated by a similar, but more complete and holistic, workaround
upstream, namely 8fe95d29d32533e8fa28383c0211555eb71ea6c1.
Thus this patch has been, in almost the simplest sense, upstreamed. Remove it.
This seems to be another one of those cases where Sebastian added a patch for an
application [1], then never submitted the patch upstream, and then someone else
submitted a different patch upstream [2], but Sebastian didn't bother checking
if his version was actually still necessary, and instead assumed it was (or just
decided it was better) and rebased the patch [3], and of course still never
submitted it upstream.
In this case the patch is for Terragen 4. That application does not, according
to my testing, actually depend on success from SetThreadIdealProcessorEx().
The patch does not add a correct implementation, is trivial to recreate, and is
easy to re-debug (the function still prints a FIXME), so it's not adding any
value. Remove it.
[1] a12dca03ce
[2] 980754bff7
[3] 0c46d1e8a2
Despite Michael Müller's claim that all patches in wine-staging actually fix
something [1], I've come across several patch sets over the years that seem to
be related to some contemporaneous work but don't actually fix any application
themselves (e.g. wine-staging commits 5d8901ac21, ba9a7a6a74, probably most of
e353590528; I think there are plenty of other examples as well.)
This patch appears to fall into this category. The upstream commit it was
written in response to was bc68b30d20.
The application in question is buggy. It uses OpenFile(), but compares the
return value to 0 instead of -1. The open in question is the first in the
program's run. The problem occurs if the DOS handles are unassigned, in which
case the valid handle 0 will be returned, and the program will interpret it as
failure, hit some broken code path, and crash.
bc68b30d20 fixes this by ensuring that the DOS standard handles are always
valid, and therefore OpenFile() will always return at least 5. This seems to
match what happens on Windows. I can reproduce this fix; I didn't go to the
trouble of building its parent, but reverting that patch in current Wine does
make the program crash the exact same way (comparing to the +relay log helpfully
provided in the bug report).
Sebastian probably saw this commit, thought that "well, there's multiple ways
for a handle to be invalid", wrote this patch catching the additional ones, and
for some reason never submitted it upstream.
Thing is, these handles come from the server, and they're guaranteed to be
either valid or zero. As evidence cf. the duplicate_handle() calls in the
new_process request handler, which were present even at the time. Hence this
patch isn't doing anything, so remove it.
[1] https://www.winehq.org/mailman3/hyperkitty/list/wine-devel@winehq.org/message/YGKVQN2N537MXAVSMLHX5IV4XCEWKBVY/
The tests are left alone since they aren't particularly important and will need to be rewritten anyway.
They should probably be somewhere other than input.c, since they don't actually relate to input.
