Re: glibc 2.2: math test failures

[Andreas Jaeger]

Michael Deutschmann <address@hidden> writes:

> > > 2.2.1 is not yet releases.  Once it is I meant.
> 
> Seems I was distracted by other things at the time 2.2.1 and even 2.2.2 was
> released.  But I've finally gotten around to recompiling and retesting.

I don't remember your exact bug report.

> The math errors are gone (or at least suppressed).  However, the test 
> failures related to Linux 2.2 features (LFS and one AIO test) still 
> continue.  You should suppress these tests on Linux 2.0 systems, or at 
> least mentioned it in your FAQ.  (I haven't retested the 386).

I don't see those failures with Linux 2.2 in glibc 2.2.3.  Which
version did you test?  If it still fails, please send some more
details, we tried to fix the failing tests.

> (While you're at it, you could also mention that at least one test,
> tst-chmod.out, fails intermittently when the test is run on one computer 
> using NFS to access the build tree of another.  I'm pretty sure that is 
> all NFS's fault.
> 
> Yes, I know NFS is slow, but sometimes I'm forced to use it to run tests
> on a computer without enough free diskspace for a glibc build.)
> 
> Also, you never answered my question about the math tolerance.  How high
> does the worst-known error on a math function have to get before you'll fix
> the problem at nontrivial cost rather than just document it?

There's no general rule.

> Or did you mean, by your complaints of not enough interested programmers,
> that you would insist on < 0.5 ULP if you only had the capacity?  (so
> if I submitted a replacement function that was provably correct but
> 1,000,000 times slower than the buggy status quo, it would be accepted.)

We prefer less than 1 ULP without any slowdowns ;-).

Andreas
-- 
 Andreas Jaeger
  SuSE Labs address@hidden
   private address@hidden
    http://www.suse.de/~aj