rjbs forgot what he was saying

not logged in (root) | by date | tagcloud | help | login

notes on how pod::weaver works

by rjbs, created 2009-08-20 23:29
last modified 2009-08-20 23:29
tagged with: @markup:md journal perl pod programming

Originally, I conceived of Pod::Weaver as a system that took two Pod streams and wove them together. One was the Pod that the user wrote. The other was Pod generated based on some stuff. (That's about as concrete as the idea was.) This got rewritten a bit and interleaved, and poof, you had better Pod.

Pod::Weaver was about a weaving-together of multiple threads, hence the name. That's still basically how PodPurler works, and it's fine, but it doesn't have much of a future.

The new design, which I've been working on, is much more output-centered. The program is given an output template, which it attempts to render using a number of inputs. Here's an example output template (still very much a notion and not an implemented thing):

[Abstract]
[Version]
[Synopsis]
[Description]
[Accordion / fronter]
[Attributes]
[Methods]
[Authors]
[Maintainers]
[Copyright]

So, Pod::Weaver's goal is to build a document with those sections. Each of those sections is produced by a plugin (a "weaver") that knows how to produce content given inputs. There are a few kinds of inputs. For example, some weavers need a "module info" input. The "Abstract" weaver wants to produce this Pod:

=head1 NAME

Some::Module - what it does

To do that, it needs a module name and abstract. The "Version" weaver needs similar input. "Attributes," on the other hand, could get inputs from a other places. It could look at an existing Pod document to find =head2 commands under =head1 ATTRIBUTES or it could look for =attr commands. It could also look at a compiled version of the package to inspect for metaobject information. (This will get us Moose autodoc, for example.)

Accordion sections are tricky (and were called out as such in the original grant proposal) because they require a cursor in the input Pod, and probably also a per-node marking of "already consumed." The accordion section lets you say, "anything that isn't interesting to another weaver that occurs at the corresponding point to this place in the source document." That lets your source document put a =head1 BY THE WAY... after its synopsis and have it show up in your output.

In the event that accordion sections grow to complex to implement, I have an alternate plan, which is to say something like:

[Subsection / postsynopsis]

This would weave in the contents found inside a =begin postsynopsis block, so that you could add arbitrary content there by wrapping it in begin/end markers. This avoids a lot of confusion with a pretty low cost. Still, I'd like to avoid even that low cost for authors by spending some time up front now.

So, next steps: rewrite the Pod::Weaver framework to load configuration something like that above (using Config::MVP) and emit Pod::Elemental or Pod::Eventual data. For now, I can start by only consuming the configuration and a Dist::Zilla object as input, and maybe a Module object of some sort. This can actually produce a lot of good test cases.

After that, I can implement something to find and "consume" bits of an input Pod document to get the synopsis, methods, and so on. Then I can tack on remaining on somewhere. Dealing with proper placement will be the tricky part, but it can be done.

All this should make it possible to also put of dealing with Pod nesting, which has been the most annoying thing I've had to deal with in all this code. I have ideas on how to sort it out, but I'd rather make progress than just noodle around.