[PATCH] Mimic MSVC whitespace collapse for incompatible token pasting

[Reid Kleckner]

Yeah, looks like it shouldn't be modified.

I still think a local fix would be better. As a reader, it's hard to
convince myself that the comment is correct:
+  // MSVC: If previous token was pasted, this must be a recovery from an
invalid
+  // paste operation. Ignore spaces before this token to mimic MSVC output.
+  // Required for generating valid UUID strings in some MS headers.

Why is it that we only come here for invalid token pastes?

On Fri, Apr 10, 2015 at 2:06 PM, Will Wilson <will at indefiant.com> wrote:

> Actually looking at the ownership semantics for the TokenLexer::Tokens
> array, I'd say it has to remain immutable. So modifying RHS looks like a
> no-go...
>
> - Will.
>
> On 10 April 2015 at 23:03, Will Wilson <will at indefiant.com> wrote:
>
>> Hi Reid,
>>
>> RHS is a const reference to the Tokens array. So RHS is only mutable if
>> we make some ugly changes... I haven't tried it, but I can if you think
>> it's worth the fallout?
>>
>> - Will.
>>
>> On 10 April 2015 at 21:59, Reid Kleckner <rnk at google.com> wrote:
>>
>>> Sure, this is http://llvm.org/pr23071.
>>>
>>> Can this be handled locally by this code?
>>>           // If we're in microsoft extensions mode, downgrade this from
>>> a hard
>>>           // error to an extension that defaults to an error.  This
>>> allows
>>>           // disabling it.
>>>           PP.Diag(Loc, PP.getLangOpts().MicrosoftExt ?
>>> diag::ext_pp_bad_paste_ms
>>>                                                      :
>>> diag::err_pp_bad_paste)
>>>               << Buffer;
>>>
>>> Any reason we can't do RHS.clearFlag(Token::LeadingSpace) here in
>>> -fms-extensions mode?
>>>
>>> On Fri, Apr 10, 2015 at 12:45 PM, Will Wilson <will at indefiant.com>
>>> wrote:
>>>
>>>> Hi Reid,
>>>>
>>>> Attached is a patch to handle a corner case supported by MSVC. I
>>>> decided to attempt to workaround the behaviour in clang after encountering
>>>> it in public MS headers for XAudio where clang would fail to generate a
>>>> valid UUID due to the UUID components being combined with the '-' UUID
>>>> separators. Clang was attempting to recover but would preserve the leading
>>>> whitespace from the tokens after each failed paste leading to spaces
>>>> creeping into the UUID and causing an error in the __declspace(uuid())
>>>> parsing.
>>>>
>>>> Reference: Microsoft DirectX SDK (June 2010)\Include\XAudio2.h(51)
>>>>
>>>> Test case included and tested against latest trunk.
>>>>
>>>> Let me know if it looks acceptable!
>>>>
>>>> Cheers,
>>>> Will.
>>>>
>>>
>>>
>>
>>
>> --
>> *Indefiant *: http://www.indefiant.com
>> Home of Recode : Runtime C++ Editing for VS
>>
>
>
>
> --
> *Indefiant *: http://www.indefiant.com
> Home of Recode : Runtime C++ Editing for VS
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150410/4728569d/attachment.html>