We need to ensure we don’t corrupt parser cache if there are indeed divergences between edit and read views and catch regressions before deploy.
Possible test idea:
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | ssastry | T332931 Read views HTML and Edit views HTML for a page should come from the same ParserCache entry | |||
| Declined | ssastry | T332934 Add tests to ensure we catch ParserCache divergences between edit and read view code paths |
Discussed with @daniel and we'd like this test code to live in the ParsoidCachePrewarmJob (T328034) and to randomly sample 1% of pages or so, and for those randomly selected pages generate "the same" output except using ParserOutputAccess instead of ParsoidOutputAccess and verify that the contents are identical, logging if not. (Also some unit tests comparing ParserOutputAccess and ParsoidOutputAccess for a more deterministic test, but these might miss interesting corner cases.)
We figured out that after all the patches and other testing, we didn't need to do this anymore.