Make list_formatter data *almost* zero-copy

[Manishearth]

Unfortunately, this cannot be 100% non-allocating due to the DFA::from_bytes() used for verification. I'm not sure what to do about that, technically the data is still zero-copy (and this can be done in a crab-bake-kosher way) but our tool will still complain.

Might be worth allowlisting it for allocating exactly 1408 bytes.

Part of #1678

[sffc]

Unfortunately, this cannot be 100% non-allocating due to the DFA::from_bytes() used for verification. I'm not sure what to do about that, technically the data is still zero-copy (and this can be done in a crab-bake-kosher way) but our tool will still complain.

Perhaps we can make an upstream contribution to make the verification be zero-alloc.

Might be worth allowlisting it for allocating exactly 1408 bytes.

Alternatively, could we add a flag to weaken the assertion from total bytes the same to current bytes the same, allowing for intermediate allocations during the deserialization process?

[Manishearth]

I was thinking about that, I kinda prefer us obviating all allocations but I'm not against sticking to the "current" thing.

As far as upstream goes, I had a look, the problem is that they need to allocate a BTreeMap to do the check. Maybe there's a way around it.

[sffc]

I was thinking about that, I kinda prefer us obviating all allocations but I'm not against sticking to the "current" thing.

We should enforce "no allocations whatsoever" by default, but switching to "no net allocations" could be an alternative to the allowlist.

[Manishearth]

sounds good, I've already updated that PR so that it measures both (but still uses total)

We should get that landed and iterate on that allowlist here

[robertbastian]

Ah good catch, we wouldn't have noticed this if we just checked for borrow annotations.

[Manishearth]

(or get this landed first and add the allowlist there, either works for me)

[jira-pull-request-webhook]

Notice: the branch changed across the force-push!

• tools/datagen/src/bin/verify-zero-copy.rs is now changed in the branch

View Diff Across Force-Push

~ Your Friendly Jira-GitHub PR Checker Bot

[Manishearth]

@sffc We now have two allowlists here

[jira-pull-request-webhook]

Notice: the branch changed across the force-push!

• components/list/Cargo.toml is now changed in the branch
• components/list/src/string_matcher.rs is now changed in the branch
• experimental/list_formatter/Cargo.toml is no longer changed in the branch
• experimental/list_formatter/src/string_matcher.rs is no longer changed in the branch

View Diff Across Force-Push

~ Your Friendly Jira-GitHub PR Checker Bot