Re: [PATCH] git-send-email: fix get_maintainer.pl regression

[Thomas Adam <thomas@xteddy.org>]

On Wed, Nov 22, 2017 at 09:05:41AM +0000, Alex Bennée wrote:
> My hacky guess about GIT's perl use calls is:
> 
>    find . -iname "*.perl" -or -iname "*.pm" -or -iname "*.pl" | xargs grep -h  "use .*::" | sort | uniq | wc -l
>    88

So let us concentrate just on git-send-email.perl for now.  In the
Module::Extract::Use module (which happens to be what corelist uses), there's
an example script called 'extract_modules' which will statically analyse a
perl file and tell you the following information:

% perl extract_modules ./git-send-email.perl
Modules required by ./git-send-email.perl:
- Authen::SASL
- Cwd (first released with Perl 5)
- Email::Valid
- Error
- File::Spec::Functions (first released with Perl 5.00504)
- File::Temp (first released with Perl 5.006001)
- Getopt::Long (first released with Perl 5)
- Git
- Git::I18N
- IO::Socket::SSL
- MIME::Base64 (first released with Perl 5.007003)
- MIME::QuotedPrint (first released with Perl 5.007003)
- Net::Domain (first released with Perl 5.007003)
- Net::SMTP (first released with Perl 5.007003)
- Net::SMTP::SSL
- POSIX (first released with Perl 5)
- Sys::Hostname (first released with Perl 5)
- Term::ANSIColor (first released with Perl 5.006)
- Term::ReadLine (first released with Perl 5.002)
- Text::ParseWords (first released with Perl 5)
- strict (first released with Perl 5)
- warnings (first released with Perl 5.006)

Therefore, we have the following modules which are not standard:

- Email::Valid
- Error
- Git
- Git::I18N
- IO::Socket::SSL
- NET::SMTP::SSL

Looking at the code for git-send-email.perl, it seems most of those are
eval()d at the point they're needed, which seems in many cases to be fallback
responses to something we've written, or a means of ensuring we don't need to
explicitly handle the case of it not being present at run-time.

> Should the solution be to just make Mail::Address a hard dependency and
> not have the fallback?

This seems like a slight on ensuring a running script which may or may not
have additional functionality depending on which modules are installed.  Given
the pretty good state of packaging across those platforms which Git runs on, I
would argue we're now in a much better position to explicitly check for
non-core modules at BEGIN{} time, and moan loudly if they're not installed.

-- Thomas Adam