[llvm-commits] [llvm] r156240 - /llvm/trunk/lib/Support/Process.cpp

[Daniel Dunbar]

On Sun, May 6, 2012 at 1:46 AM, Chandler Carruth <chandlerc at google.com> wrote:
> On Sat, May 5, 2012 at 10:36 AM, Duncan Sands <baldrick at free.fr> wrote:
>>
>> > Won't this give us really bad random numbers on system with no
>> > arc4random (linux)? I'm thinking of multiple invocations of clang in the
>> > same second all getting the same "random" file name. Does Linux have another
>> > source of pseudorandom numbers?
>>
>> If this is about getting unique temporary files, shouldn't tmpfile or
>> mkstemp be
>> used?
>
>
> Yes, and it seems to have caused: PR12743
>
> This is *not* the way to get temporary files. There are proper OS APIs for
> that we should be using, not RNGs....

I tend to agree, however this is not what I changed. I just replaced
factored out the horribleness of how the RNG was used (both
implementations were poor, for different reasons). The seeding problem
should be fixed now, thanks to chapuni's r156252.

I would support someone eliminating the current API and using the
proper OS API, but it requires changing the fs API because of the way
unique_file was defined.

That said, why is using the OS API so much better than using a well
seeded RNG? I'm not familiar with the Linux implementation.

 - Daniel

>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>