Make ListFormatter allocation datagen test not allocate

[sffc]

It seems that ListFormatter invokes a nontrivial validation of the regex-automata bytes that involves memory allocations and a BTreeMap. We should find a way to avoid this, either by validating with no allocations or by performing GIGO. This may involve patching the regex-automata crate.

[sffc]

So it looks like to avoid the BTreeMap we just need to disable the alloc feature on regex automata.

https://github.com/BurntSushi/regex-automata/blob/a4900a9ba0d14b27f2ed7d782b296f54d300b376/src/dfa/sparse.rs#L1372

It says:

        // In order to validate everything, we not only need to make sure we
        // can decode every state, but that every transition in every state
        // points to a valid state. There are many duplicative transitions, so
        // we record state IDs that we've verified so that we don't redo the
        // decoding work.
        //
        // Except, when in no_std mode, we don't have dynamic memory allocation
        // available to us, so we skip this optimization. It's not clear
        // whether doing something more clever is worth it just yet. If you're
        // profiling this code and need it to run faster, please file an issue.
        //
        // ---AG

I think we should disable the alloc feature. If, in the future, we discover a bottleneck (ideally one that a client reports), we can file an issue. There are a few ways to make it faster, one of which is to GIGO it.

CC @robertbastian

[robertbastian]

We already deactivate alloc, however the zero-copy test runs in the datagen crate, with datagen's dependencies, which all have std (datagen actually, which can't work without regex-automata/alloc) enabled. It actually has a hard dependency on datagen, because it uses DatagenProvider to convert from buffers to concrete data payloads. We'd have to redesign how this tests gets its deserialization implementations. @Manishearth.

[sffc]

Hmm, interesting. If it's an issue we could probably just submit an upstream PR splitting this out into a different feature we can disable.

[robertbastian]

Upstream will likely tell us to fix our features. It would be good to find a way to run the zero-copy test without the datagen feature enabled.

[Manishearth]

We can at least ask, I guess.