[PATCH] clang-cl: Store the linker options for the ASAN runtime libraries in the obj files as opposed to pass it to the linker on the command line

[Reid Kleckner]

On Thu, Jul 10, 2014 at 1:27 PM, Ehsan Akhgari <ehsan.akhgari at gmail.com>
wrote:

> On Thu, Jul 10, 2014 at 4:20 PM, Reid Kleckner <rnk at google.com> wrote:
>
>> On Thu, Jul 10, 2014 at 7:04 AM, Ehsan Akhgari <ehsan.akhgari at gmail.com>
>> wrote:
>>
>>> Hmm, thinking more about this patch, this requires you to know whether a
>>> given object file is going to be linked into a DLL when building the C++
>>> source so that you can pass -LD to clang-cl.  That may be not easy to get
>>> right depending on the build system, and when things go wrong you might
>>> need nasty -nodefaultlib arguments to the linker.  Maybe this is not the
>>> right approach after all?
>>>
>>
>> Oh, true.  I forgot about the DLL thunk library.
>>
>
> OK, I think I'll drop this patch then.  Sorry for wasting your time!
>
>
>> Honestly, I think we need to completely revisit our CRT interceptor
>> approach so we can support MD, maybe MDd, and avoid the extra DLL thunk
>> stuff.  I don't have brilliant ideas for fixing this at the moment, though.
>>  :(
>>
>
> Do you know where I should look to figure out why the existing strategy
> has been employed?  If I understand the reasons things were done this way
> better, perhaps I can figure out a way to reuse Firefox's malloc
> replacement techniques with -MD/MDd in the ASAN runtime...
>

I have to defer to Timur, here.  I think the Firefox malloc replacement
techniques are probably the way to go, but ASan has many more interceptors
than just malloc&co. It hooks all the str* functions, for example.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140710/6f3a97cf/attachment.html>