[cfe-dev] Windows Lit problems again. What should we do?

[Daniel Dunbar]

On Sun, Sep 19, 2010 at 12:46 PM, Francois Pichet <pichet2000 at gmail.com> wrote:
> Can you look at the patch I posted on llvm-commit: doing a rename as a
> copy+delete fixes the problem on Windows.
>
> IMO, it is a better solution than disabling the temp code on Windows.
> Disabling the new code means clang will behave differently on Windows
> vs Unix. + introduce a platform #ifdef inside the driver.

I agree in principle, it is nice to have behavior be the same across
platforms. However, this is also intended to be a user-transparent QOI
improvement for Clang and it is closely tied to how Unix handles
files. If the underlying FS in Windows can't support the same QOI
tweak then it is reasonable to not do it.

Copy + delete seems suboptimal to me. It means Clang will be hitting
the FS twice as hard as need be.

My feeling is we should first figure out exactly why this fails, and
disabling the code is a good short term way to get the buildbot to
stop complaining while we investigate.

 - Daniel

>
> On Sun, Sep 19, 2010 at 3:08 PM, Daniel Dunbar <daniel_dunbar at apple.com> wrote:
>>
>> On Sep 19, 2010, at 12:07 AM, Francois Pichet wrote:
>>
>>> Another solution would be to modify lit to use a different temp '%t'
>>> filename for every clang invocation within the same testcase.
>>
>> This doesn't work, many test cases reuse a %t for one reason or other.
>>
>> The simple solution to all of these problems is to just disable the new tempfile code on Windows.
>>
>>  - Daniel
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>