[llvm-commits] [llvm] r62619 - /llvm/trunk/test/CodeGen/Generic/2008-07-29-EHLabel.ll

[Dale Johannesen]

On Jan 20, 2009, at 5:55 PMPST, Chris Lattner wrote:

> On Tue, 20 Jan 2009, Duncan Sands wrote:
>
>> Hi Chris,
>>
>>> Don't bother running the assembler, we don't know that it will be  
>>> configured
>>> for whatever llc defaults to.  This fixes PR3363
>>
>> the whole point of this testcase is to run the assembler!  The  
>> original
>> bug resulted in assembler being output that didn't assemble.
>
> Ok, how can we solve this issue?  The test harness isn't set up to  
> know
> about *any* assembler, much less one for a particular target.  Can't  
> this
> use 'grep' like anything other test?

Probably.  While that looks like a bug I might have introduced, the  
analysis,
fix, and testcase are all Duncan's; I'll let him answer.

>  While yes, the output didn't
> assemble, there was a specific reason that it didn't assemble, why not
> just grep for that?

Because checking that the file assembles is a much stronger and more  
valuable test.
Tests that embalm a bug that existed at one point are fairly useless;  
while duplicate bugs can
creep back in, that's rare.  (In years of working on gcc, a large  
majority of the regressions we
found in the testsuite were things that had nothing to do with the  
original purpose of the test.
The testsuite's main virtue is volume, not the contents of the  
individual tests.)