metacpan / cpan-changes Goto Github PK

The current spec for change logs is so loose that pretty much anything is a "valid" changelog file. This makes it hard to interpret changelogs. Since the spec is so loose, many things that are valid according to the spec are unparseable by the code (at least not into any reasonable format).

The spec also doesn't match well with the existing changelogs in the wild. Different "group" formats are common, as well as deeper nestings than just one level of groups. The data model of the module doesn't support this, but that is easy enough to fix.

I don't think treating it as a "spec" is really very useful.

I would like to include Test::CPAN::Changes in the xt directory of my distribution https://metacpan.org/module/VM::JiffyBox
However I have problems.

First the links to the relevant files:

• Testfile: https://github.com/tim-schwarz/VM-JiffyBox/blob/develop/xt/changes.t
• Changesfile: https://github.com/tim-schwarz/VM-JiffyBox/blob/develop/Changes

The problem occurs because of the date-format. However I use Dist::Zilla for this. The dates are created how dzil thinks they should be.

When I now run the test-suit like so prove -Ilib xt/changes.t, I get errors similar to this one for each entry:

xt/changes.t .. 1/4 Date "2013-05-27 16:44:00 Europe/Berlin" is not in the recommend format at /home/perl/perl5/perlbrew/perls/perl-5.16.3/lib/site_perl/5.16.3/Test/CPAN/Changes.pm line 28.

However this stuff is just verbose... prove tells me that tests passed: All tests successful.

I do think it would be great if the defaults of dzil would play nice with the standards, set by this lib. But I don't know at which point we should introduce any Changes. Is the standard here just to strict? Or does dzil rely on false stuff.

I notice that some cpan authors include something like [user] on a changes line when attributing an author as contributing a feature. I was thinking it'd be nice to have a specification for this that way it can also be parsed out.

I use the RSS feed to see what's new on CPAN. Sometimes I don't know by a dist's name whether it will be interesting, and would like to see the abstract. (Like, what's "Pithub" all about anyway?)

The abstract should be in the META.* file.

Until now I've abused the fact wrapping is implemented in terms of Text::Wrap, and coerced Text::Wrap to have some less insane defaults.

Some of those are outlined here: https://rt.cpan.org/Ticket/Display.html?id=95736

$Text::Wrap::break='[\x09\x20]';
$Text::Wrap::huge='overflow';

And both of those are important because the first prevents unicode NONBREAK-SPACE (0xA0) from being wrapped on, and the second prevents long tokens ( like URI's ) from being uselessly snapped in two.

The last one I do personally is localise

$Text::Wrap::columns = 120

Because under 80 columns I find far too restrictive for any serious amount of text.

And this ability has been clobbered by 087f3b7 , where it localises it at a scope closer than I can access.

Would you be open to a pull request that broke load_string into a CPAN::Changes::Parser class with several methods?
Something that would be easier to subclass in case there are times where a custom (non-spec-conforming) parser may be desired?

For example, separate methods for detecting matching lines for versions and groups
as well as handling them would allow a lot more flexibility (and also make unit testing easier, etc).