Speculative HTML parser

In 2020, Bocoup specified and wrote conformance tests for the speculative HTML parser. However, the specification pull request was still pending review. In February 2021, Henri Sivonen, implementer of the HTML parser in Gecko, left a few outstanding review comments that needed to be addressed. Our objective was to address the review comments for the speculative HTML parser spec change and get it merged into the HTML standard.

There were two material review comments.

The first was requesting that speculative fetches should be allowed to happen both during speculative parsing and also during normal parsing. The potential performance gain is smaller for the normal parsing case, but if parsing happens in parallel to JavaScript execution, it can still make a difference and at least Gecko already does this. We changed the specification to allow (but not require) speculative fetches during normal parsing, during element creation by the tree builder.

The second review comment was that a specific URL should not be allowed to be speculatively fetched multiple times. We changed the specification to maintain a list of speculative fetch URLs, to match Gecko’s implementation strategy.