Render numpydoc strings from a template#77
Conversation
|
Ping @amueller |
bddf80d to
9c232e4
Compare
|
Wow that's much better than what I had in mind. |
|
Btw, I'm not entirely happy with how the |
|
Related to gh-78, which was about having the section order fixed or not. Having a fixed default ordering which is enforced by default but can be overridden/customized if you really want or need to would make sense to me. |
|
can you add |
|
Very clean and simple implementation, I'd be fine with adding this. |
|
sphinx requires jinja2...?
…On 29 November 2016 at 19:41, Ralf Gommers ***@***.***> wrote:
Very clean and simple implementation, I'd be fine with adding this.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#77 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AAEz61jDgHqsKeGWvf4xOov0VLOZ8bhuks5rC-UogaJpZM4K3jJf>
.
|
|
I like the template implementation, but also: shouldn't attributes be rendered in the position suggested by default? This is the order in which is appears in the docstring and according to the standard. |
I know, but explicit > implicit and all that. You normally put all your direct dependencies in |
|
Of course. |
|
I run as far away from dependencies as I can :) |
|
I've added the jinja2 dependency, specifying the same version as current sphinx, though I assume an earlier version would work fine. |
Without checking: IIRC there's a long standing mismatch between docs and implementation for at least |
|
Looks OK to me. @pv Do you have any thoughts on the section ordering? |
|
I'd really like to finally get scikit-learn using numpydoc directly. @amueller and I consider not having Attributes in its incumbent position a blocker. Can we please do something about it? |
|
No objections, so in it goes. Thanks @jnothman, all! |
BUG: fix installing numpydoc, broken with missing templates after gh-77.
In scikit-learn, we would like to start tracking the upstream numpydoc (scikit-learn/scikit-learn#7355). However, we currently render Attributes closely after Parameters, due to their significance, etc. Numpydoc renders Attributes after examples, notes, references, etc. which we find unideal.
This patch allows us to add
templates/numpydoc_doscstring.rstcontaining:to achieve the sought reordering. It is clearly also much more flexible than that, but I do not have (many) grand designs as yet.