Installing Perl module with nmake

[Jim Keenan]

This posting does not concern Perl code. It concerns the use of 'nmake' to
install a Perl module on a Windows98 system. If you deem this off-topic or
want to rant about why I am still using a Win98 box, stop reading here.

I have developed a number of Perl modules in a Linux environment using the
standard MakeMaker approach. Once these modules have been tested and
installed on Linux, I send the tarball to my Win98 box and repeat the
process with 'nmake' (v1.50). Heretofore, have never had any difference in
results, and two of these modules have been uploaded to CPAN where they
passed all tests on both *nix and Windows.

Tonight I followed the same procedure with a new module. Everything fine on
Linux. On Windows, when I called 'nmake test', I got a screen full of
warning messages and indications that most of the tests failed. But I
couldn't get a clear reading on the errors, because the DOS-prompt doesn't
permit me to scroll backwards far enough.

I got this far with the previous version of the module a couple of days ago.
For fun, I went ahead and called 'nmake install' -- and the module installed
correctly!

Tonight I tried a different approach. I shifted to the Cygwin BASH prompt
after calling 'perl Makefile.PL' and 'nmake'. I then called 'nmake test' --
and this time all 200 tests passed, just as they had done on Linux! I then
proceeded to 'nmake install', which, as before, proceeded correctly.

So has anyone seen this problem before? 'nmake test' gives 1 (wrong) result
from the DOS-prompt and a different (correct) result from the Cygwin
BASH-prompt.

Jim Keenan

 

[Koms Bomb]

This posting does not concern Perl code. It concerns the use of 'nmake'
to

install a Perl module on a Windows98 system. If you deem this off-topic or
want to rant about why I am still using a Win98 box, stop reading here.

I use both Win98 and Win2k

Tonight I followed the same procedure with a new module. Everything fine on
Linux. On Windows, when I called 'nmake test', I got a screen full of
warning messages and indications that most of the tests failed. But I
couldn't get a clear reading on the errors, because the DOS-prompt doesn't
permit me to scroll backwards far enough.

'nmake test | more' can pause the scrolling or 'nmake test > result.txt' can
redirect the message to a file, so you can read and analyse them.

 

[Jay Tilton]

: Tonight I followed the same procedure with a new module.

What was that module? Is it available on CPAN? A reader (like me) who is
interested in reproducing the bad result could use that information.

: Everything fine on
: Linux. On Windows, when I called 'nmake test', I got a screen full of
: warning messages and indications that most of the tests failed. But I
: couldn't get a clear reading on the errors, because the DOS-prompt doesn't
: permit me to scroll backwards far enough.

The output from nmake can be redirected to a file or piped into a pager.

: I got this far with the previous version of the module a couple of days ago.
: For fun, I went ahead and called 'nmake install' -- and the module installed
: correctly!

That's not especially surprising.

: Tonight I tried a different approach. I shifted to the Cygwin BASH prompt
: after calling 'perl Makefile.PL' and 'nmake'. I then called 'nmake test' --
: and this time all 200 tests passed, just as they had done on Linux! I then
: proceeded to 'nmake install', which, as before, proceeded correctly.
:
: So has anyone seen this problem before? 'nmake test' gives 1 (wrong) result
: from the DOS-prompt

Yes. I've seen it when the test procedure depends on a unix-like shell
that can redirect STDERR.

: and a different (correct) result from the Cygwin
: BASH-prompt.

Never tried. If the the test procedure isn't windows-savvy, I'm not very
confident that the module will work correctly in a windows environment.

 

[Sisyphus]

Tonight I tried a different approach. I shifted to the Cygwin BASH prompt
after calling 'perl Makefile.PL' and 'nmake'. I then called 'nmake test' --
and this time all 200 tests passed, just as they had done on Linux! I then
proceeded to 'nmake install', which, as before, proceeded correctly.

Leads me to believe that there may be some shell commands in the test
scripts (or even in the module itself) on which command.com chokes. Iirc
(and I'm pretty sure I do in this instance) it's not hard to get
command.com to choke on shell commands. A simple '2>&1' is often enough
to completely bamboozle it.

Cheers,
Rob