Updating rquickjs from 0.6.1 to 0.8.1

[jeffcharles]

I'm going to document the issues I ran into trying to update rquickjs here.

1. Turning on the SIMD implementation for JSON seems to cause at least one test262 failure
2. Fuel increases in tests
3. We still need to investigate if there are breaking changes to the QuickJS bytecode so we know whether we need to cut new major versions of the javy and javy-plugin-api crates and increment the plugin version.

The test262 failure I'm seeing is in the file crates/javy/test262/test/built-ins/JSON/stringify/value-bigint.js and the error message is Expected a TypeError to be thrown but no exception was thrown at all.

For the fuel increases:

Test	Target fuel	New fuel	Increase %
test_encoding	258197	270087	4.6%
test_exported_functions	61404	63518	3.4%
test_identity	47773	50760	6.3%
test_empty	22590	24554	8.7%
test_exported_default_arrow_fn	39004	40038	2.7%
test_exported_default_fn	39951	40933	2.5%
test_fib	64681	71302	10.2%
test_logging_with_redirect	35007	37709	7.8%
test_logging_without_redirect	36641	37592	2.6%
test_recursive_fib	69306	76848	10.9%
test_readme_script	254503	271684	6.8%
test_str	146027	154404	5.7%

[Sytten]

We changed to quickjs-ng under the hood. if you need help feel free to join the discord @jeffcharles (https://discord.gg/pyMppMngYU). Also I am extracting the serde serialization you guys have built (thanks for that) so that everybody can use it: https://github.com/rquickjs/rquickjs-serde
FYI it looks like you might not be running the tests for this crate since the test test_map_does_not_support_non_string_keys doesnt work.

[Sytten]

@jeffcharles @saulecabrera I finished extracting the crate and cleaning it up. It's live https://crates.io/crates/rquickjs-serde
If you want to be included as a maintainers of the crate let me know, I think it would be good to continue improving on it in the shared repo.

[saulecabrera]

I think it would be good to continue improving on it in the shared repo.

I'm not opposed to this idea, if we're going this route though, I think it might make sense to consider consuming the one that you have published from this project, to avoid having to maintain two potentially diverging versions of the same code.

[jeffcharles]

The increase in fuel use is the main blocker for updating rquickjs for Javy. The first issue I outlined with the failing test is likely resolvable and the third issue is just a matter of incrementing the major version in the import namespace for the default plugin.

[saulecabrera]

@Sytten would you be up to opening a PR to make the switch to use the published crate instead? Just note that I won't be merged right away, as we first need to figure out how to upgrade rquickjs without incurring in the performance overhead, however, I think it's probably fine to get that process started.

[Sytten]

There wont be an easy fix to improve performances. quickjs-ng is not as fast as the original quickjs but has a lot of fixes / better in general. Now that quickjs is again active rquickjs might start supporting it again but seems unlikely since there are some APIs we added to quickjs-ng that dont exist in the base version.

As for porting, I am not using Javy personally and I have very limited time, so maybe I can PR once you have decided to want to move to rquickjs 0.9.

We are actively trying to be added as maintainers of rquickjs. Lots of companies are using it, I am contributing mainly to the llrt project of AWS on node like modules like fs, fetch, child process, net, etc. Some of which you might want to adopt too.

[Sytten]

Unrelated but I am trying to remove the class id checks in this serde and you might want to chime on my comment here: quickjs-ng/quickjs#758 (comment)