[llvm] r318361 - Convert a use of createUniqueFile to TempFile::create.

[Davide Italiano via llvm-commits]

FWIW, after the recent changes to bugpoint I see my screen spammed
with these messages:

*** Attempting to reduce testcase by deleting instructions:
Simplification Level #1
Checking instruction:   %tmp5 = phi i64 [ %tmp13, %bb9 ], [ 0, %bb
]Failed to delete temp file Bad file descriptor

Checking instruction:   %tmp6 = phi i64 [ %tmp13, %bb9 ], [ 0, %bb
]Failed to delete temp file Bad file descriptor

Checking instruction:   %tmp7 = icmp eq i64 %tmp6, 0Failed to delete
temp file Bad file descriptor

Checking instruction:   %tmp10 = phi i1 [ false, %bb8 ], [ true, %bb4
]Failed to delete temp file Bad file descriptor

Checking instruction:   %tmp11 = or i1 %tmp7, %tmp10Failed to delete
temp file Bad file descriptor

Checking instruction:   %tmp12 = zext i1 %tmp11 to i8Failed to delete
temp file Bad file descriptor

Checking instruction:   %tmp13 = xor i64 %tmp5, 5Failed to delete temp
file Bad file descriptor

They're harmless, although annoying.
They seem related to your file handling changes, but I haven't
bisected to confirm. Can you please take a look?

Thanks (and happy thanksgiving),

--
Davide

On Wed, Nov 22, 2017 at 2:09 PM, Rafael Avila de Espindola
<rafael.espindola at gmail.com> wrote:
> Davide Italiano <davide at freebsd.org> writes:
>
>> On Wed, Nov 22, 2017 at 12:03 PM, Rafael Avila de Espindola via
>> llvm-commits <llvm-commits at lists.llvm.org> wrote:
>>> Should be fixed now. Thanks a lot for the bug report.
>>>
>>> Cheers,
>>> Rafael
>>>
>>
>> Any chance we can have some regression testing for this (even if
>> enabled only on Windows?)
>
> It depends.
>
> The issue in bugpoint was that it was calling discard twice. Given the
> design of TempFile, that should work.
>
> One patch fixed TempFile and that included a unit test. With that change
> the code in bugpoint became just redundant and I removed it in a
> followup.
>
> If the issue is testing bugpont itself on windows, we need to have
> passes that crash so that bugpoint can be tested. On unix we do that by
> having opt load BugpointPasses.so. On windows we would need to have an
> option to link those passes into opt.
>
> Cheers,
> Rafael



-- 
Davide

"There are no solved problems; there are only problems that are more
or less solved" -- Henri Poincare