RE: test-binary-io unit test failure under MingW + fix patches

[dave]

Bruno,

I wanted to follow-up with you with the latest information.  I used the --create-testdir option per your last mail and the test-binary-io unit test passes for me as well.  However, when I create a new project manually (using the same directory as I used with the --dir option alongside --create-testdir) and add the entries gnulib-tool calls for to configure.ac and Makefile.am after performing the module import, the unit test fails.

Given the above, the problem must be the result of a configuration error somewhere in the setup of the new manual project versus the one created via the --create-testdir option which ultimately leads to the failure.  It's really quite odd given the nature of the failure and the workarounds that seem to resolve it (inserting a sleep or using a separate temp file as opposed to stdout).  I have not had the time to ascertain which automake/autoconf setting/settings ultimately result in the issue, but I did want to provide details that the failure doesn't seem to be specific to the unit test.

Thanks for your responses.

David

 

-------- Original Message --------
Subject: Re: test-binary-io unit test failure under MingW + fix patches
From: Bruno Haible <address@hidden>
Date: Wed, March 09, 2011 5:16 pm
To: address@hidden
Cc: address@hidden

Hello David,

> This should now be in plain text.

It is. Thanks.

> Here are the requested details about my specific setup. It is a
> standard MinGW install (mingw-get-inst-20110211.exe) using the
> pre-packaged repository catalogs and MSYS.
> ...
> - 50GB NTFS partition on a local SATA drive
> ...
> GCC: gcc version 4.5.2 (GCC) [Target: mingw32]
> LD: GNU ld (GNU Binutils) 2.21
> Make: GNU Make 3.81 (i686-pc-msys)
> MinGW version from _mingw.h: 3.18

This is very bizarre. I can't reproduce the failure.

I created a testdir with
$ ./gnulib-tool --create-testdir --with-tests --dir=/tmp/testdir poll,
transferred that testdir to a Windows XP machine (also NTFS file system),
and the test-binary-io.sh test succeeds, both in MSYS (with an older mingw,
running inside the MSYS provided 'rxvt') and Cygwin.

What should we do about this now? Given that MSYS is not a recommended
environment (because of the hacks that it does in the command-line argument
processing, and because of the pipe handling bugs it has when running inside
rxvt), I'd propose to just add a comment to test-binary-io.sh stating that
this test may fail on MSYS. What do you think?

Bruno
--
In memoriam Otto Freundlich <http://en.wikipedia.org/wiki/Otto_Freundlich>;