On 23/10/02 13:09 -0400, Mitchell N Charity wrote:
> It looks like %INC's concept of a canonical filename differs
> from File::Spec's. And may get pretty weird.
>
> Here is a version of the patch that was suggested:
>
> --- Inline.pm Wed Oct 23 12:21:00 2002
> +++ Inline.pm_new Wed Oct 23 12:21:00 2002
> @@ -428,8 +428,10 @@
>
> my @pkgparts = split(/::/, $o->{API}{pkg});
> my $realname = File::Spec->catfile(@pkgparts) . '.pm';
> - my $realpath = $INC{$realname}
> - or croak M27_module_not_indexed($realname);
> + my $realname_for_inc = $realname;
> + $realname_for_inc =~ s/\\/\//g if $^O =~ /Win32/i;
> + my $realpath = $INC{$realname_for_inc}
> + or croak M27_module_not_indexed($realname_for_inc);
>
> my ($volume,$dir,$file) = File::Spec->splitpath($realpath);
> my @dirparts = File::Spec->splitdir($dir);
>
>
> $^O is a pattern match rather than eq, because some versions of perl
> seem to have gotten it wrong (or so a quick grep for $^O in the change
> log suggests).
>
>
> I have not tested this under Windows. Just linux.
>
> Dos may require something similar.
>
> It looks like VMS, and perhaps os2, are also playing games.
> Related files include
> perl.c (eg, S_incpush)
> vms/vms.c (eg, ...tounix...)
> os2/os2.c (eg, perllib_mangle)
> I only grepped about briefly, so I'm no doubt missing things.
> As a hack, it might be worth seeing what unixify() (from
> VMS::Filespec??) in File::Spec::VMS does.
>
> The Right Thing(tm) is to determine whether there is any portable way
> to assemble filenames for %INC. A post to perl5-porters?
I pinged the p5p on this last night. They said that the keys of %INC always
use Unix separators. So the right thing to do is use
File::Spec::Unix->catdir() for %INC keys. That's what I did in TRIAL8.
Cheers, Brian
