Currently, each pipeline creates its own document and, at various stages, the results of those parses need to be adopted by the main document.
We expect some performance gains by eliminating that work.
See https://gerrit.wikimedia.org/r/#/c/385312/
Also, there're a few instances where we have a dummyDoc to avoid string concatenation which should make use of it.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | None | T179082 Use one ownerDocument for the entire parse | |||
| Open | None | T221790 Parsoid extension API should use DOM fragments, not documents |
I think generally the pattern should be:
var df = env.ownerDocument.createDocumentFragment();
var tempBody = env.ownerDocument.createElement('body');
df.append(tempBody);
tempBody.innerHTML = "some string to parse";as opposed to:
var newDoc = domino.createDOMImplementation().createHTMLDocument(); newDoc.documentElement.innerHTML = "some string to parse";
This uses Document.createDocumentFragment for storage of a DOM tree which isn't directly linked into the document itself.
https://gerrit.wikimedia.org/r/c/mediawiki/services/parsoid/+/490654 identifies where all the docs are currently created.