r187374 - PR16715: Fix assert in verifier: only mark call to 'operator new' as 'builtin' if

[Nick Lewycky]

On 15 August 2013 15:08, Eli Friedman <eli.friedman at gmail.com> wrote:

> On Mon, Jul 29, 2013 at 2:30 PM, Richard Smith <richard at metafoo.co.uk>wrote:
>
>> On Mon, Jul 29, 2013 at 2:00 PM, Eli Friedman <eli.friedman at gmail.com>wrote:
>>
>>> On Mon, Jul 29, 2013 at 1:14 PM, Richard Smith
>>> <richard-llvm at metafoo.co.uk> wrote:
>>> > Author: rsmith
>>> > Date: Mon Jul 29 15:14:16 2013
>>> > New Revision: 187374
>>> >
>>> > URL: http://llvm.org/viewvc/llvm-project?rev=187374&view=rev
>>> > Log:
>>> > PR16715: Fix assert in verifier: only mark call to 'operator new' as
>>> 'builtin' if
>>> > corresponding 'operator new' was actually emitted as a function marked
>>> 'nobuiltin'.
>>>
>>> There's a similar bug involving LTO; consider the following:
>>>
>>> a.cpp:
>>> using size_t = decltype(sizeof(0));
>>> void *operator new(size_t) __attribute__((alias("something")));
>>>
>>> b.cpp:
>>> int *pr16715 = new int;
>>>
>>> If I link these two together with llvm-link, I get exactly the same
>>> error.
>>>
>>> Overall, this seems like a very fragile constraint; maybe we can relax
>>> it somehow?
>>
>>
>> It would seem reasonable to me for the verifier to allow the 'builtin'
>> attribute on calls to functions declared without 'nobuiltin'. Patch
>> attached.
>>
>
> Ping.
>

Ping?

I just hit this exact problem, the proposed patch seems right to me. Please
commit?

Nick
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130901/48baba48/attachment.html>