This is the mail archive of the
mailing list for the binutils project.
Re: RFC: Allow the suppression of expensive linker and gas tests
- From: Ben Elliston <bje at au1 dot ibm dot com>
- To: Nick Clifton <nickc at redhat dot com>
- Cc: binutils at sourceware dot org
- Date: Fri, 09 Nov 2007 09:57:12 +1100
- Subject: Re: RFC: Allow the suppression of expensive linker and gas tests
- References: <email@example.com>
> I saw the time for my full test run drop from 1 hour 24 mins to 27
> minutes which was great. So my question is, does anyone have any
> objection to applying this patch and allowing these tests to be
> suppressed if the user explicitly requests it on the command line ?
Making the testsuite conditional in this way bothers me a bit. I
realise that this is something that the user needs to explicitly request
via their environment, but if the majority of binutils developers decide
that it's a good thing to do, we will effectively stop running those
tests regularly (in which case, what is their value?)
When you run the testsuite for 100+ different targets, that strikes me
as the most appropriate time to run the testsuite in the most extensive
way possible. It is then that we would be most likely to discover
subtle regressions for obscure targets.
If the testsuite takes too long to run, then let's address that by
scaling back the magnitude of the worst offending test cases. I don't
think giving us the means to produce sets of test results that are not
directly comparable is the right approach.