Original bug ID: 6274 Reporter:@trefis Assigned to:@zoggy Status: closed (set by @xavierleroy on 2015-12-11T18:27:40Z) Resolution: open Priority: normal Severity: feature Category: ocamldoc Tags: patch Monitored by:@gasche@hcarty
Definition of object types, while syntactically similar to record types, are not "handled" by ocamldoc, i.e. the documentation one associates to fields does not appear in ocamldoc output.
It would be nice if it did.
I'm not sure how much this is related to #4347 .
I probably have a naïve approach here, however it does seem to work without breaking anything and I am not yet "deeply sad to the point of not using ocamldoc at all" (to quote jm).
You will notice that I introduce some code duplication with the handling of records (although it might be hard to notice, considering how much code duplication there already is) which should be easily factorisable.
However I wanted to gauge people's interest before investing more time on that; and it can always be cleaned afterward.
The patches are currently on the 4.01.0 branch, but I don't think rebasing them to trunk would pose any problem.
The text was updated successfully, but these errors were encountered:
I just did a review of the patch (medium precision, I would say). It consists of three commits:
trefis@759a72a collects comments occurring after method fields in a definition of an object type
trefis@68fc982 collects method fields of object type definitions in the Odoc_type.t_type structure, and then pretty-prints them (in various backends) with more structure than just outputting the source text, rather like record definitions.
trefis@501f15c fixes a bug with the implementation approach of commit (2)
I think (commit 1) is reasonable, but I am less convinced by the overall design of (commit 2): a new "type kind" is added to the type Odoc_type.type_kind, which previously precisely mirrored the type_kind of the typedtree (type synonym, variant or record), to store the information needed by the backends.
I think it is wrong to break the correspondence between the typedtree "type kind" and ocamldoc's "type kind" (for example ocamldoc uses type_kind to decide whether to print parameters variance information, and I suspect this is broken by this change). A better design, I think, would be to add the idea that, besides the type_kind, useful printing information may come from the type manifest (the type-expression after the equal sign, if any), and have a function like
val manifest_structure : Types.type_expr -> Odoc_type.ty_manifest option
where ty_manifest would be a new type storing information for the type expressions we decide to print in a special way when they occur on the right of an equal sign -- (only) objects types for now. At printing points you could check both the kind and the manifest, and decide on both (eg. giving priority to the ty_manifest when it exists).
I think most of the current implementation could be reused, the idea is only to preserve the tight correspondence between typedtree's and ocamldoc's type_kind.
Other than that the patch looks sensible, and I don't see any reason not to include this extra feature.
I tried to follow your proposition:
I first reverted the two previous commits, i.e. "2" and "3" in your comment, and then introduced [Odoc_type.type_manifest] as suggested (all this could be squashed before being pushed to the svn if you want a cleaner history).
I tried to make this commit as little intrusive as I could, so there is a lot of code redundancy which I then remove in the two next commits.