Server crash in polyglot tests: fake_archive_exc_t

[Tryneus]

Version: rethinkdb 1.7.3-492-gcfefee-dirty (GCC 4.6.3)
error: Error in ../src/errors.cc at line 110:
error: Uncaught exception of type "fake_archive_exc_t"
error: Backtrace:
error: Wed Jul 31 20:59:31 2013

       1: ../../build/release/rethinkdb() [0x93b712]
       2: ../../build/release/rethinkdb() [0x936750]
       3: ../../build/release/rethinkdb() [0x93696f]
       4: +0xb5846 at 0x2ad278926846 (/usr/lib/x86_64-linux-gnu/libstdc++.so.6)
       5: +0xb5873 at 0x2ad278926873 (/usr/lib/x86_64-linux-gnu/libstdc++.so.6)
       6: +0xb596e at 0x2ad27892696e (/usr/lib/x86_64-linux-gnu/libstdc++.so.6)
       7: ../../build/release/rethinkdb() [0x6a7e02] mailbox_t<void ()(rdb_protocol_t::read_response_t)>::read_impl_t::read(read_stream_t*) + 2146
       8: ../../build/release/rethinkdb() [0x90d0ee] mailbox_manager_t::mailbox_read_coroutine(int, unsigned long, string_read_stream_t*) + 238
       9: ../../build/release/rethinkdb() [0x43a91e] coro_t::run() + 46
error: Exiting.

(Notes added for function names)

It looks like there's a problem with rdb_protocol_t::read_response_t deserialization. I think I've run into something similar before that when an uninitialized value was serialized, it would crash the receiving side. Perhaps some recent change to rdb_protocol_t::read_response_t has allowed for this.

[Tryneus]

I have an idea for how to track this whole class of bugs down, will be implementing it shortly.

[Tryneus]

Well, that didn't work out too well. I may have a fix for this instance, though. This error is rather repeatable, it shows up after a few iterations using:

while [ $? == 0 ]; do make py_polyglot TEST=sindex-api; done

[Tryneus]

Implemented and up in review 790.

[Tryneus]

This has been approved and merged to next in commit 070e117. Will be in release 1.8.