16:00
<Michael Ficarra>
https://p.ocmatos.com/blog/jsse-a-javascript-engine-built-by-an-agent.html
16:06
<jschoi>

Array.fromAsync was the final fix

Ah yes, Array.fromAsync, the final frontier…

16:08
<Michael Ficarra>
I'm assuming it's because the plan it created roughly ordered recent things by when they were added to the language
16:11
<jschoi>
Its PLAN.md was apparently generated “from the ECMAScript spec and test262 submodules”, so I wonder how that ordering happened. Not that it matters much.
16:51
<bakkot>
I wonder if it has garbage collection
16:51
<bakkot>
if I were given the task to implement a spec-compliant JS engine with no further qualifications I definitely would not add garbage collection
16:58
<jschoi>

Claude picked a reasonable feature order, but it made some architectural decisions early on (like how to handle generators, how to structure the GC, how to deal with WTF-8 strings) that caused expensive rework later.

17:15
<Ashley Claymore>
was the answer for how to structure the GC: "use a noop"
17:21
<Jesse>
https://github.com/pmatos/jsse/blob/main/src/interpreter/gc.rs
17:27
<Ashley Claymore>
I do think this is a great benchmark. As the definition is very fixed. It will be interesting what would happen given the same challenge 1 year from now
17:27
<Ashley Claymore>
1 year is enough time for models to have this repo in their training set
17:28
<Michael Ficarra>
it seems like it attempts GC before every statement: https://github.com/pmatos/jsse/blob/c490bc3fee8c4c8e6b2452f4acacb60d84b02a39/src/interpreter/exec.rs#L112
17:28
<Michael Ficarra>
which seems both too coarse and too fine