Re: Build broken (was: [PATCH 02/11] Generate pass-instances.def)

[David Malcolm <dmalcolm at redhat dot com>]

On Wed, 2013-07-31 at 18:00 +0200, Jan-Benedict Glaw wrote:
> On Wed, 2013-07-31 11:31:35 -0400, David Malcolm <dmalcolm@redhat.com> wrote:
> > On Wed, 2013-07-31 at 11:36 +0200, Jan-Benedict Glaw wrote:
> > > On Wed, 2013-07-31 10:34:10 +0200, Jan-Benedict Glaw <jbglaw@lug-owl.de> wrote:
> [breakage with mawk]
> > > > It seems this does only happen on one of the three running build
> > > > clients. That one is using `mawk' instead of `gawk', what the two
> > > > other builders (which are not affected) use.
> > > 
> > > The substr() was wrong, awk starts all its indices with 1, while the
> > > script used 0. gawk ignores this, mawk starts the substring at 1, but
> > > counts from 0 upwards.
> > 
> > A thousand apologies, and thanks for fixing this - I guess I owe you a
> > $FAVORITE_BEVERAGE.  I had only tested with gawk (with and without -c)
> > and with busybox awk.  I've now installed mawk and nawk on my dev box.
> 
> That's why I'm running the build robot :)  Once I've got some
> time[tm], I'll spend it a small web frontend to look into the basic
> information (which builds worked/failed, show build logfile, show git
> log between working/non-working version.)

BTW, have you seen buildbot?  i.e. http://buildbot.net/

MIT-licensed and Python-based.

It's in Fedora, and I see that it's in Debian:
http://packages.debian.org/unstable/buildbot

(With my Python hat on, we use that for verifying every commit of
CPython builds and passes the testsuite, across multiple build
environments, and it's fairly easy for 3rd-parties to hook in their own
build slaves into a build farm, for the less common envs).

> > Are that any other awks I should be testing with - and is this
> > information captured somewhere for reference?
> 
> I don't know of any--I'm actually not a regular awk user at all.
> And actually testing specific awk features isn't codified IMHO. And
> while indices start at 1 (so the code was buggy in that way), it might
> also be true that the implementation-defined behavior of virtually all
> awk variants is to silently s/0/1/. Then, this would be an additional
> glitch (not bug) in my mawk.

It sounds like your awk expertise is greater than mine, though :)

[...snip my testing...]

> > Have I messed up my testing above, or is something else going on?  What
> > version of mawk are you using?
> 
> Your testing looks fine to me. My Debian "unstable" mawk is slightly
> older than yours: 1.3.3-17

Looking at mawk's CHANGES file I see this entry:

20090726
[...snip...]
	+ modify workaround for (incorrect) scripts which use a zero-parameter
	  for substr to ensure the overall length of the result stays the same.
	  For example, from makewhatis:
		filename_no_gz = substr(filename, 0, RSTART - 1);

so perhaps that's it.

Thanks again.
Dave