Perl PortGroup fails to reinplace non-ascii Makefiles

[mojca (Mojca Miklavec)]

When building p5-yaml on 10.8 and 10.9, sed fails with

DEBUG: Executing reinplace: /usr/bin/sed {/^CCFLAGS *=/s/$/ /} < /opt/local/var/macports/build/_opt_mports_dports_perl_p5-yaml/p5.16-yaml/work/YAML-0.90/Makefile >@ file14
DEBUG: sed: RE error: illegal byte sequence

(see ​https://build.macports.org/builders/buildports-mavericks-x86_64/builds/2454) because the Makefile contains a non-ascii character (in Latin 1 encoding, invalid UTF).

This comes from the perl PortGroup:

# CCFLAGS can be passed in to "configure" but it's not necessarily inherited.
# LDFLAGS can't be passed in (or if it can, it's not easy to figure out how).
post-configure {
    fs-traverse file ${configure.dir} {
        if {[file isfile ${file}] && [file tail ${file}] eq "Makefile"} {
            ui_info "Fixing flags in [string map "${configure.dir}/ {}" ${file}]"
            reinplace "/^CCFLAGS *=/s/$/ [get_canonical_archflags cc]/" ${file}
            reinplace "/^OTHERLDFLAGS *=/s/$/ [get_canonical_archflags ld]/" ${file}
        }
    }
}

Suggestions from IRC:

neverpanic:

set LC_ALL to C I think that's a common issue with ill-formatted UTF-8 files /usr/bin/sed changed on 10.8 and above You can either patch that to be valid UTF-8 or set configure.env LC_ALL=C in the Portfile It is kind of weird to use the locale settings for that. I mean you could set the correct locale for the file, but then any messages to your terminal would be broken…

Dar1us:

it refuses to process files which aren't validly encoded given the current LC settings .. and by default that is UTF-8

[cooljeanius (Eric Gallager)]

Cc Me!

[danielluke (Daniel J. Luke)]

see also: ​https://lists.macosforge.org/pipermail/macports-dev/2012-August/019993.html

(yes the perl portgroup should be updated)

[anddam (Andrea D'Amore)]

The portgroup cannot possibly know what encoding will the produced Makefile use.

Port p5-yaml is using ExtUtils::MakeMaker that in turn relies on Pod::Man. Pod::Man can produce utf8 Makefiles, ​search utf8 in in this page but seems that MakeMaker doesn't have any option to specify that, see ​this request.

Since the portgroup is patching Makefiles we could sanitize the file by adding a new variable specifying a charset to convert from. In this case the portfile writer would specify something like

perl5.makefile_encoding    iso8859-1
perl5.setup         YAML 0.90

and the post-configure phase would take care of convert the file to UTF-8 (or any other target encoding depending on the darwin release) before reinplaces.