[devdocsgjs/main: 1261/1867] PyTorch 1.6.0+ doc structure support

[Andy Holmes <andyholmes src gnome org>]

commit cf852768ed086f24f2514712585f0d54daed8463
Author: Peiran Yao <xavieryao me com>
Date:   Fri Nov 20 13:04:54 2020 -0700

    PyTorch 1.6.0+ doc structure support
    
    Document structure has changed since PyTorch 1.6.0.

 lib/docs/filters/pytorch/clean_html.rb |  9 +++++++++
 lib/docs/filters/pytorch/entries.rb    | 18 ++++++++++++++----
 2 files changed, 23 insertions(+), 4 deletions(-)
---
diff --git a/lib/docs/filters/pytorch/clean_html.rb b/lib/docs/filters/pytorch/clean_html.rb
index dd19c3e0..465ae3a5 100644
--- a/lib/docs/filters/pytorch/clean_html.rb
+++ b/lib/docs/filters/pytorch/clean_html.rb
@@ -2,9 +2,18 @@ module Docs
   class Pytorch
     class CleanHtmlFilter < Filter
       def call
+        breadcrumbs = at_css('.pytorch-breadcrumbs')
+        type_name = breadcrumbs.css('li')[1].content
+
         @doc = at_css('.pytorch-article')
         # Show katex-mathml nodes and remove katex-html nodes
         css('.katex-html').remove
+
+        # pass type_name to following filters as a new node
+        node = Nokogiri::XML::Node.new 'meta', doc
+        node.content = type_name
+        doc.child.before node
+
         doc
       end
     end
diff --git a/lib/docs/filters/pytorch/entries.rb b/lib/docs/filters/pytorch/entries.rb
index f63926b4..ba92e222 100644
--- a/lib/docs/filters/pytorch/entries.rb
+++ b/lib/docs/filters/pytorch/entries.rb
@@ -2,6 +2,15 @@ module Docs
   class Pytorch
     class EntriesFilter < Docs::EntriesFilter
       def get_name
+        # retrive the name in breadcrumb from the auxiliary node
+        name_in_breadcrumb = doc.child.content
+        doc.child.remove
+
+        # hard-coded name replacements, for better presentation.
+        name_replacements = {
+          "Distributed communication package - torch.distributed" => "torch.distributed"
+        }
+
         # The id of the container `div.section` indicates the page type.
         # If the id starts with `module-`, then it's an API reference,
         # otherwise it is a note or design doc.
@@ -9,7 +18,9 @@ module Docs
         if doc.element_children[1]['id']&.starts_with? 'module-'
           /\Amodule-(.*)/.match(doc.element_children[1]['id'])[1]
         else
-          at_css('h1').content
+          name_in_breadcrumb = name_in_breadcrumb.delete_suffix(' >')
+          name_in_breadcrumb = name_replacements.fetch(name_in_breadcrumb, name_in_breadcrumb)
+          name_in_breadcrumb
         end
       end
 
@@ -18,9 +29,8 @@ module Docs
       end
 
       def include_default_entry?
-        # If the page is not an API reference, we only include it in the index when it
-        # contains additional entries. See the doc for `get_name`.
-        doc.element_children[1]['id']&.starts_with? 'module-'
+        # Only include API references, and ignore notes or design docs
+        !subpath.start_with? 'generated/' and type.start_with? 'torch'
       end
 
       def additional_entries