[llvm-commits] Asan interception library patch (issue 6010049)

[Kostya Serebryany]

Please hold on with this patch.
I'd really like to understand the consequences but have not time until late
next week (traveling, etc).
It would help if you could make a test for this new functionality.

--kcc

On Thu, Apr 12, 2012 at 7:22 AM, <samsonov at google.com> wrote:

> On 2012/04/12 14:21:26, samsonov wrote:
>
>> http://codereview.appspot.com/**6010049/diff/1/interception.h<http://codereview.appspot.com/6010049/diff/1/interception.h>
>> File interception.h (right):
>>
>
>  http://codereview.appspot.com/**6010049/diff/1/interception.h#**newcode78<http://codereview.appspot.com/6010049/diff/1/interception.h#newcode78>
>> interception.h:78: # define DECLARE_WRAPPER(ret_type, convention,
>>
> func, ...)
>
>> Naming was not self-explanatory already and now it's even worse :( As
>>
> I get it,
>
>> you use DECLARE_WRAPPER(...func...) to define "func" that is:
>> 1) alias to an interceptor if instrumented code doesn't define its own
>>
> "func"
>
>> 2) breaks linkage if instrumented code does define its own "func"
>> Why not call this macro smth like
>>
> "ADD_GUARD_AGAINST_CUSTOM_**DEFINITION" or smth
>
>> like that.
>>
>
> I mean, "breaks in runtime"
>
>
>
>  http://codereview.appspot.com/**6010049/diff/1/interception.h#**newcode91<http://codereview.appspot.com/6010049/diff/1/interception.h#newcode91>
>> interception.h:91: # define WRAP(x) interception_wrap_##x
>> This fails on our output tests: the top stack frame of allocation now
>>
> says
>
>> "interception_wrap_malloc" instead of "malloc". Can we use some other
>>
> naming?
>
>  http://codereview.appspot.com/**6010049/diff/1/interception.h#**newcode96<http://codereview.appspot.com/6010049/diff/1/interception.h#newcode96>
>> interception.h:96: __attribute__((weak,
>>
> alias("interception_wrap_"#**func), \
>
>> alias(WRAPPER_NAME(func)) ?
>>
>
>  http://codereview.appspot.com/**6010049/diff/1/interception.h#**newcode97<http://codereview.appspot.com/6010049/diff/1/interception.h#newcode97>
>> interception.h:97: visibility("default")))
>> You can use INTERCEPTOR_ATTRIBUTE instead of visibility("default")
>>
>
>  http://codereview.appspot.com/**6010049/diff/1/interception_**linux.h<http://codereview.appspot.com/6010049/diff/1/interception_linux.h>
>> File interception_linux.h (right):
>>
>
>
> http://codereview.appspot.com/**6010049/diff/1/interception_**
> linux.h#newcode32<http://codereview.appspot.com/6010049/diff/1/interception_linux.h#newcode32>
>
>> interception_linux.h:32: (void*)&func, (void*)&WRAP(func))
>> I think that
>> && (void*)&func == (void*)&WRAP(func)
>> would be better.
>> Anyway, it's worth explaining this magic in the comment here or in
>> interception.h
>>
>
>
>
> http://codereview.appspot.com/**6010049/<http://codereview.appspot.com/6010049/>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120412/882006c8/attachment.html>