Re: test-vc-cvs [was: IRIX failures]

[Eric Blake]

On 09/02/2010 08:02 AM, Tom G. Christensen wrote:

> > zsh -vx ./test-vc-list-files-cvs.sh
> >
> Sure, output here:
> http://jupiterrise.com/tmp/zsh_vx_test-vc-list-files-cvs.out

Thanks.  I definitely see a couple of problems.

First, there's no 'emulate sh' command anywhere in the trace, and zsh 
when NOT in POSIX mode does weird things (such as failure to do field 
splitting).  I guess part of my init.sh patching today will be fixing 
init.sh to properly sanitize zsh when bash is skipped would be 
appropriate (zsh installed as /bin/sh already behaves in POSIX mode, so 
you are probably the first case of someone actually hitting zsh as the 
preferred fallback shell).

Next, your trace showed the cause of the first message:

+mktempd_:51> template_length_=+mktempd_:1> wc -c
+mktempd_:51> template_length_='            16'
+mktempd_:52> nx_=+mktempd_:1> echo gt-init.sh.
+mktempd_:52> nx_=+mktempd_:1> wc -c
+mktempd_:52> nx_='            12'
+mktempd_:53> nx_=+mktempd_:1> expr '            16' - '            12'
non-numeric argument

So wc is outputting space-padded results, and expr doesn't like the 
leading spaces.  And once expr dies, the rest of the temporary directory 
creation goes downhill.

I also noticed (in trying this myself) that directly running the script 
doesn't quite work if build-aux is not in PATH (that aspect is 
automatically taken care of when run via the Makefile).  But at this 
point, let's wait until some more gnulib patches are in, and I make 
another m4 snapshot, to see if we caught everything on this go-around.

> echo $BASH_VERSION
> + echo '3.1.16(1)-release'
> 3.1.16(1)-release
> test $(echo y) = y || exit 1
> cho y
> ++ cho y
> bash: line 2: cho: command not found
> + test = y
> bash: line 2: test: =: unary operator expected
> + exit 1
>
> Definitely something funny going on there.
> Inserting a space so it's $( echo y) and the test passes.

Oh my.  Your bash installation is quite broken - no wonder the script 
bypasses it.  Is this a self-built bash?  If so, you may want to upgrade 
to 3.1.17 (and/or 3.2, 4.0, or 4.1), and hope that the rebuild fixes 
whatever was broken about your 3.1.16.  But looking at bash31-017, that 
only touches array expansion, so it's not directly related to your weird 
command substitution behavior.  But at least it's not a bug in gnulib.

If it were a more recent bash version, I'd suggest reporting it to the 
bash list; but since it is rather old, it may be that upgrading to a 
newer bash is all you need.

--
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org