llvm dev - Jun 2008 - [LLVMdev] clang testing?

Just for fun I added clang to the list of compilers that my random 
tester beats on.  It found a codegen bug in short order.  Would you 
folks say that clang is ready for the fairly high degree of abuse 
produced by my tool or would it be better to wait until it matures?  Thanks,

John Regehr

On Jun 27, 2008, at 1:41 PM, John Regehr wrote:
> Just for fun I added clang to the list of compilers that my random
> tester beats on.  It found a codegen bug in short order.  Would you
> folks say that clang is ready for the fairly high degree of abuse
> produced by my tool or would it be better to wait until it matures?   
> Thanks,
I say beat it up and file problem reports for it.  No better way to  
get it in tip top shape.  I'll assume you're doing C testing.  If C++,  
I'd say wait a while, as C++ is only 3-5% of the way there.  I'd  
probably say, only file 10-20 reports, and then stream more in as  
those are fixed.  This should help reduce the work that you do and  
self limit so you don't file a thousand or more bug reports, which, I  
suspect we wouldn't want.

Sounds good.  Each report requires effort to produce a minimized testcase 
and to attempt to ensure that the bug is fresh so I definitely won't be 
submitting thousands!

Thanks,

John Regehr


On Fri, 27 Jun 2008, Mike Stump wrote:
> On Jun 27, 2008, at 1:41 PM, John Regehr wrote:
>> Just for fun I added clang to the list of compilers that my random
>> tester beats on.  It found a codegen bug in short order.  Would you
>> folks say that clang is ready for the fairly high degree of abuse
>> produced by my tool or would it be better to wait until it matures?
>> Thanks,
>
> I say beat it up and file problem reports for it.  No better way to
> get it in tip top shape.  I'll assume you're doing C testing.  If
C++,
> I'd say wait a while, as C++ is only 3-5% of the way there.  I'd
> probably say, only file 10-20 reports, and then stream more in as
> those are fixed.  This should help reduce the work that you do and
> self limit so you don't file a thousand or more bug reports, which, I
> suspect we wouldn't want.
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

Good news-- after finding one nasty codegen bug in clang random testing 
over the weekend has turned up nothing else.

John Regehr