21:38
<Michael Ficarra>
at the TG5 call this morning, some EPFL researchers presented their work mechanising the RegExp section of the spec: https://github.com/epfl-systemf/Warblre
21:39
<Michael Ficarra>
they did this to show equivalence to their linear matching algorithm
21:40
<Michael Ficarra>
and at the same time proved that our spec doesn't have any assertion failures or out-of-bound accesses in that section (🎉)
21:41
<Michael Ficarra>
I was thinking we should invite them to an editor call to see if we could use the mechanisation to help us make refactorings, proving that they are semantics-preserving
21:41
<Michael Ficarra>
but first we should probably think what kind of refactoring we'd like to make to that section
21:42
<Michael Ficarra>
I know we're unhappy with it (especially @bakkot), but I don't know if there's any concrete plans for what we'd want
21:43
<Michael Ficarra>
another idea is that we could take the linear matching algorithm and derive the spec factoring from that (given we know it's semantically identical)
21:45
<bakkot>
my current unhappiness with the section is about the observable complexity, not the editorial specification
21:45
<bakkot>
we dealt with most of the glaring editorial issues
21:46
<Michael Ficarra>
it's still definitely weird
21:46
<Michael Ficarra>
like these are weird https://tc39.es/ecma262/multipage/text-processing.html#sec-countleftcapturingparenswithin
21:48
<Michael Ficarra>
also, they skipped Annex B and based their work on ES2024, but I don't think either of those is significant for this work