Follow up to Bug 1053898. DAMP shows a small regression (3% on Linux, might be bigger on other platforms) for the test custom.inspector.expandall.manychildren. We should try to profile and see if this can be corrected. For the record expandall.manychildren expands a node that contains 100 children (the maximum displayed by the markup view).

Damp results on windows show a bit more regression than what was detected on Linux try pushes: - 3% on custom.inspector.expandall.manychildren - 3% on custom.inspector.expandall.balanced - 8% on mutations

The current patch attached is an improvement I dropped from the shadow DOM queue because it was not mandatory. But locally it seems to slightly improve the performance of children() calls, so let's see if we pick up an improvement on CI as well: https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=68979bd71976dcc326270262490614282f1480a3&newProject=try&newRevision=0f1f5bbe3a3945aa4d82ed581f8a1f0834ddc41b&framework=1

I have been trying several improvements on the children() method, with no success so far. Even filter caching doesn't improve the results for the expand* tests. Looking at profiles, I now think that the expand* tests are stressing the client more than the server, so maybe the regression is more linked to client changes than server changes. started a few talos pushes: - baseline https://treeherder.mozilla.org/#/jobs?repo=try&revision=692aaed754008a4cfa8b41b7377fe524b022f18a - server only https://treeherder.mozilla.org/#/jobs?repo=try&revision=b34573ac9fe9b99f66f4cc9a3ae96f981f75e7d8 - server + client https://treeherder.mozilla.org/#/jobs?repo=try&revision=d96e90a73f5d4b066810283be27a7d3a76c29c88

We have improved performance for custom.inspector.expandall.manychildren and custom.inspector.expandall.balanced. We don't plan to do further work.