> ac> As Michael well knows, supplemental information, such as which specific 
> ac> branch of a test passed or failed can be included in paren in the test 
> ac> message.  Any analysis tools comparing test results needs to accomodate 
> ac> this convention.
> 
> I don't know any such convention.

Michael, you and I had an e-mail exchange about this very issue.  The 
end result, last time, was no change.

> I'm looking at a typical gdb.sum.
> There are 1523 instances of '(...)' in 112 different test scripts.
> Most of these are part of the test name.

To split hairs, I can see two cases:

- two runs within identical environments
I can see 'diff -u' reasonably working here.

- two runs within different environments
After paren stripping the results should be identical (or close two it).

For instance sizeof.exp contains various tests to check that sizes are 
sane.  The actual sizes found are included in the output.  That's fine 
since if the numbers were to change between runs the test results are 
pretty sunk.

Thanks to the advent of PIE (position independant code) it's now 
possible for the PC/FP to change between "100% identical" runs.  So for 
this specific case, what ever.

David, I've attached a script (repost).  It's what Michael tables 
started out using but have likely since been completly rewritten.

Andrew


> If you want to promulgate such a convention, put it in the docs,
> and I'll start submitting patches to change all these test scripts,
> or committing them as obvious fixes.
> 
> A better path would be to pick a character sequence which is unused,
> such as '// ...'.
>