This is the mail archive of the
mailing list for the glibc project.
Re: run one test
- From: Carlos O'Donell <carlos at redhat dot com>
- To: DJ Delorie <dj at redhat dot com>, libc-alpha at sourceware dot org
- Date: Wed, 25 Sep 2019 17:11:57 -0400
- Subject: Re: run one test
- References: <email@example.com>
On 9/25/19 4:41 PM, DJ Delorie wrote:
> One of the small items that was mentioned at Cauldron was "how do to
> re-run just one test?"
> While it can be done with a suitable command line make invocation,
> this is a lot easier...
> I'll turn it into a real patch once we bikeshed the target name ;-)
> diff --git a/Makefile b/Makefile
> index 67ddd01bfe..c424d9ee02 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -499,3 +499,11 @@ FORCE:
> iconvdata/% localedata/% po/%: FORCE
> $(MAKE) $(PARALLELMFLAGS) -C $(@D) $(@F)
> +# Convenience target to rerun one test, from the top of the build tree
> +# Example: make onetest t=wcsmbs/test-wcsnlen
> +onetest :
> + @-rm -f $(objpfx)$t.out
> + $(MAKE) subdir=$(dir $t) -C $(dir $t) ..=../ $(objpfx)$t.out
> + @cat $(objpfx)$t.test-result
> + @cat $(objpfx)$t.out
Does this avoid rebuilding *anything* else?
One of the problems with the suggested fix on the wiki is that it can
put your tree into a bad state because of missing dependencies you
might rebuild only part of glibc and fail.
For example if $t.out depends on libpthread.so, then what happens
if you've touched a file under nptl/? Likewise for libdl.so?
I'm not saying we shouldn't do this, but that we should document the
limitations of this command.
And I suggest 'make onecheck' to match 'make xcheck' and 'make check' :-)