net: do not close socket on EWOULDBLOCK, as this can happen during cursor iteration

[tommilligan]

Reason for the change

When calling .next(wait=False) on a cursor, ReqlTimeoutError is raised if the query would block. This is expected and documented behaviour.

However, introduced in 9fbd6f2, this will also close the underlying socket, preventing .next() from being called again on the cursor.

This is unexpected behaviour; we would like to be able to continue listening for new changes in a cursor, even after one request has timed out.

Description

• net: Do not close underlying socket connection if EWOULDBLOCK received
• tests/integration: Add test case that reproduces the issue (fails against master)

Code examples

See the added test for an example.

Checklist

• Unit tests created/modified N/A
• Integration tests created/modified

References

Documentation detailing the behaviour of next and wait can be found here: https://rethinkdb.com/api/python/next/

[gabor-boros]

@tommilligan Very nice catch! Thanks 🙏

[gabor-boros]

I'll release this today/tomorrow

[tommilligan]

No problem - to be honest it's only something we've run into when writing integration tests, as that's where we generally set wait=False. But I guess it would've turned up sooner or later at runtime.