[PATCH] Catch malformed inline assembly clobber constraints

[Akira Hatanaka]

On Wed, Sep 3, 2014 at 6:49 PM, Eric Christopher <echristo at gmail.com> wrote:

>
>
> On Wed, Sep 3, 2014 at 6:06 PM, Akira Hatanaka <ahatanak at gmail.com> wrote:
>
>> On Wed, Sep 3, 2014 at 5:32 PM, Eric Christopher <echristo at gmail.com>
>> wrote:
>>
>>> The testcase doesn't need either the triple or the data layout for an
>>> llvm-as test.
>>>
>>>
>> OK, I'll remove the triples and data layout.
>>
>>
>>> Also, I tried it, which llvm_unreachable? (I mean, it's obviously doing
>>> something wonky so it's not necessarily a bad patch, but I can't duplicate
>>> whatever problem you're seeing).
>>>
>>>
>> Sorry, my comment was misleading. If you compile the test case with "llc
>> -march=aarch64", llc terminates at MachineValueType.h:386.
>>
>>
> I see it now, thanks. I'd tried an older version of llc and it didn't fail
> oddly enough.
>
> I'm not sure this is the right place for the check, could probably do the
> clobber error checking up above where it's set? Any reason to put it down
> here?
>
>
I was just trying to avoid checking whether I==E, but looking at it now, it
might be better to move the check up.

I am assuming you are suggesting moving the check to line 93 where the
prefixes are parsed.


> Thanks!
>
> -eric
>
>
>
>>
>>
>>> -eric
>>>
>>>
>>>
>>>
>>> On Wed, Sep 3, 2014 at 1:45 PM, Akira Hatanaka <ahatanak at gmail.com>
>>> wrote:
>>>
>>>> The attached patch adds a check to InlineAsm::ConstraintInfo::Parse
>>>> which checks that '{' immediately follows '~' in the inline assembly
>>>> clobber constraint. This enables catching the error in the test case
>>>> earlier by LLParser and printing a more helpful diagnostic, rather than
>>>> terminating when an llvm_unreachable is triggered.
>>>>
>>>> This patch fixes rdar://problem/14206559.
>>>>
>>>>
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>>
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140903/eddaae1e/attachment.html>