modularize tests - please review

[Sean Silva]

On Tue, Mar 19, 2013 at 2:50 AM, Daniel Dunbar <daniel at zuster.org> wrote:

>
>
>
> On Mon, Mar 18, 2013 at 7:01 PM, Sean Silva <silvas at purdue.edu> wrote:
>
>>
>> On Mon, Mar 18, 2013 at 8:06 PM, Thompson, John <
>> John_Thompson at playstation.sony.com> wrote:
>>
>>>  Thanks, Sean.****
>>>
>>> ** **
>>>
>>> I’ve fixed the issues mentioned, except for this one:****
>>>
>>> ** **
>>>
>>> **Ø  **+# RUN: modularize %s -x c++ 2>&1 | FileCheck %s****
>>>
>>> ** **
>>>
>>> It seems “|&” mentioned in the docs doesn’t work on Windows, but “2>&1”
>>> does.  Both forms seem to work on Linux, though.  I see “2>&1”  is used in
>>> several clang tests.****
>>>
>>> **
>>>
>>
>> Daniel, can you confirm that the advice regarding |& vs. 2>&1 on <
>> http://llvm.org/docs/TestingGuide.html#writing-new-regression-tests> is
>> out of date with respect to lit's current behavior? If it is out of date,
>> I'd like to remove it.
>>
>
> That information is just wrong. Lit has never supported "|&", but 2>&1
> does work (and should have worked for a long time).
>
> To the best of my current recollection, the only redirections that don't
> work also fail with an error inside the shell parser or shell execution (of
> course, the latter only happens when using the internal shell execution as
> opposed to the one that shells out to bash).
>
>  - Daniel
>
>
>> -- Sean Silva
>>
>
>
Ok, I purged that incorrect documentation in r177403.

-- Sean Silva
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130319/c457cfc9/attachment.html>