Re: [Module-build-general] Fwd: Module::Build Win32 patches

[Ken W. <ke...@ma...>]

> From: "James Freeman" <jfr...@ta...>
> Date: Sun Feb 16, 2003  11:59:39  AM US/Central
> To: <ke...@ma...>
> Subject: Module::Build Win32 patches
> Reply-To: <jfr...@ta...>
>
> Hi Ken
>
> Module::Build is not very Win32 friendly.

Actually Module::Build itself is very Win32 friendly, it takes pains to 
do things in a cross-platform way everywhere.  Module::Build::Compat, a 
compatibility layer that builds a pass-through Makefile is clearly 
*not* very cross-platform yet, so your patches should help get the ball 
rolling in that area.

However, your patch creates new problems: by using the full path name 
to the Makefile and Build scripts in the Makefile, the shell-quoting 
can-of-worms is now open.  Consider what will happen if the full path 
is:

  '/home/ken/ foo/Makefile'

or

  '/System Folder/foo/Makefile'

or any number of more malicious variations.  The second path above will 
render an OS X system quite useless when "rm -rf $makefile" is run.

I'm not sure whether the right way to do this is to figure out shell 
quoting in a cross-platform way (*not* looking forward to this job), or 
to get relative paths working in a cross-platform way (would be much 
safer).  What's your feeling on the matter?


> I have patched the sources of all
> the Modules to use more portable code. Mainly this is not using 
> ./Build as
> the shell command but rather perl /path/to/Build as set by file spec. 
> Fixed
> bareword error Cwd::cwd as well which was causing errors.

Why would that cause errors?  Is there some version of Cwd.pm that 
doesn't have cwd()?


> Now a little less broken but failure to write a .packlist makes 
> uninstallation difficult.

Right, a .packlist isn't written.  Uninstallation is broken anyway, 
there should be a better mechanism designed rather than perpetuating 
the broken one.  It's a goal of the project.


> Another bug is that even though I incremented the $VERSION nmake 
> install
> fails to recognise this and do a reinstall. Here are the highlights of 
> where
> it is currently. I have attached a tar.gz of the patched distribution 
> for
> you to run diff over.

It would be helpful to submit patches against the current CVS tree, if 
you have access to a CVS client and a version of diff.  Some of the 
Win32 errors in your output are fixed there already, actually.  At 
least the "Can't unlink file lib/XSTest.c" one.

Thanks again for the patches.

  -Ken