Port new fuzzer to v1

[bolshoytoster]

Closes #663.

[alejandro-vaz]

this really requires the new CI as well

[bolshoytoster]

I'm waiting for #677 to merge first.

[alejandro-vaz]

hey @bolshoytoster do you have a name?? just so I can name you without pinging you directly to GitHub??

[bolshoytoster]

I'm Taylor :)

[alejandro-vaz]

okay #677 merged now, we can continue

[bolshoytoster]

I'll be completely honest: I don't know how CI works and I don't really know how to debug it. If I copy over checks.yml and fuzzing.yml from v2 will it just work?

[bolshoytoster]

Are we porting #675 to v1?

[alejandro-vaz]

yeah also #675

a few notes:

• we have to downgrade MSRV
• callgrind needs a dev-dep or something like that
• everything else seems to magically work

[alejandro-vaz]

missing file now

I think you can copy-paste the callgrind one

[bolshoytoster]

I copied the file over. I just had to convert to v1 syntax and remove the from_elem bench since that's only in v2.

[bolshoytoster]

Miri checks not taking 10 minutes is quite refreshing :)

[alejandro-vaz]

hahaha new miri checks are peak

the workflow of MSRV still needs to be downgraded otherwise the first 2018-edition version

[alejandro-vaz]

no no that edition was okay

[alejandro-vaz]

revert that

[alejandro-vaz]

I meant this

[bolshoytoster]

What's v1's MSRV? I can't seem to find it anywhere.

[alejandro-vaz]

since v1 doesn't have an explicit MSRV but has edition set to 2018, it means the MSRV is the first version with the 2018 edition

[bolshoytoster]

I've used 1.36, since the old workflow uses it. This causes an issue because for some reason the MSRV check is trying to read fuzz/Cargo.toml (maybe because of the workspace). It might be fixed if I change the edition back to 2018.

[alejandro-vaz]

yeah then change the fuzzing edition to 2018 if necessary

[alejandro-vaz]

???

it was 2018 not 2021

[bolshoytoster]

Oh, iai-callgrind uses edition = "2021".

Do you reckon making it optional is enough?

[alejandro-vaz]

what if we just disable callgrind for v1 and remove the complexity altogether

[bolshoytoster]

Yeah, it doesn't look like you can have optional dev-dependencies anyway.

It would've been nice to be able to directly compare callgrind bench results between v1/v2, but I think the regular bechmarks are good enough.

[alejandro-vaz]

tomorrow I'll review your PR that I'm now dead tired

I'll also open a ton of issues

by the way, msrv fails because the associated constant wasn't stabilized then, we have to use the module constant in std

[bolshoytoster]

Thanks.

It looks like the previous workflow disabled tests on MSRV (#288). Should I keep them disabled or see if they'll work now?

[alejandro-vaz]

let's have them enabled better

[bolshoytoster]

The test_size test seems to fail because MSRV doesn't do niches.

I'm not really sure what I can do.

I can't use #[cfg(version(..))] because it's unstable.

You don't have to respond immediately - I'm going to bed.