fix(codec): Properly decode partial DATA frames

[cpcloud]

Closes #76

TODO:

• Add mock to reproduce in a unit test

Motivation

Reproduced from #76:

When receiving data by polling the body, it's possible to end up in a
situation where the polling results in a buffer of size N where N >= length, but also that N < 5 + length.

Roughly the flow is like this:

1. Tonic polls for data, gets N bytes, where length <= N < 5 + length
2. Tonic sees that N bytes have been read, and if N >= length, _advances the buffer by 5 bytes
3. Advancing the buffer reduces the buffer length by 5, now we have that length - 5 <= N < length
4. Since we need length bytes for the proto, prost fails with a buffer underflow

Solution

The solution that I've implemented here is to check that self.buf--the buffer
that holds the bytes for decoding--has enough data to decode both the tonic
header and the message. If the check returns false, wait for more data,
otherwise decode the bytes

[lovesegfault]

This is also not accurate, I think. It's Apache OR MIT dual-licensed.

[lovesegfault]

As an addendum, the right thing to do if license headers are desired is to use SPDX format ones, otherwise the LICENSE in the repo works just fine.

[LucioFranco]

@cpcloud so if I understand correct the issue is that we may not have the initial 5 bytes from the buffer handed to us from hyper? Thus the buf.advance(5) fails?

[cpcloud]

@LucioFranco No, the issue is that we'll receive the header + a partial body. For example:

> read 10005 bytes from the body (tonic header + 3 bytes from protos + 9997 bytes of the byte array)
  > advance 5
  > 10000 left over, but the message length is actually 10003
  > BEFORE: try to decode
  > AFTER: check that we've read enough to contain the entire message and if not, poll the body again

We need to ensure that self.buf has enough of the body to decode the full message including the header.

[LucioFranco]

Ok so after our discussion on discord I am happy with this solution. I hate to ask this but would it be possible to try and make this change with the smallest footprint possible which I think is just the second if statement in readbody. And instead of introducing a new example service can we add a test, there should already be tests for the codec. In the mock body simulate that the frames are split.

[LucioFranco]

and thank you so much for figuring this out! It is super helpful, I'd like to get this change in before the next release this week. Let me know if you need anything.

[cpcloud]

@LucioFranco Looks like the decode test never actually reads anything! 😂 Each poll_data() call is checking whether 0 > 100.

[cpcloud]

I'll add a fix in this PR.

[LucioFranco]

LGTM! Thank you so much for digging into this!

[cpcloud]

Yeah! Thanks for reviewing + merging!