Re: libm-test rewrite

[Andreas Jaeger <aj at suse dot de>]

>>>>> Ulrich Drepper writes:

Uli> Andreas Jaeger <aj@suse.de> writes:
>> I've put on ftp.suse.com/pub/people/aj/libm-test.diff.gz [1] a patch for
>> glibc 2.2 which adds the rewritten libm-tests. 

Uli> Many thanks to Andreas.  Good job.  I feel much better about this form
Uli> for the test suite.  I haven't given it a full run yet but I managed
Uli> to create the ULPs file for x86.

Glad to hear.  Btw. I don't think the make rules are fixed yet.  Just
rm all test files from math and you notice while generating the *.d
files:
test-float.c:34: libm-test.c: No such file or directory

Uli> There are only three more things to do as far as I'm concerned:

Uli> - write documentation on how to generate initial error files for a new
Uli>   architecture

Ok, I'll do that.

Uli> - add a mode where the test program complains even if the error is smaller
Uli>   than the one from the error file.  This should be enabled for all
Uli>   development and test releases.

Should we even complain about an error of 1 ulp?

Uli> - somehow generate from the error files a table where for each function
Uli>   and each architecture the maximal error is listed.  I've mentioned
Uli>   before that this is a requirement for a Unix library.

Do you want to have one file which contains all the information?  I
guess I can come up with something.  There's only one problem: The
tests are pointwise and not exhaustive - the maximum error given is a
minimal maximum error.

Andreas
-- 
 Andreas Jaeger   
  SuSE Labs aj@suse.de	
   private aj@arthur.rhein-neckar.de