Re: avoiding races in IPDL protocols

On 7/15/09 4:04 PM, Chris Jones wrote:

>> Yeah, that defeats the point of having a protocol state system.
>>
> 
> Not really.  It's a question of error handling --- no races would have
> been tolerated in any case.  The question is whether races are dynamic
> errors (boo!), static errors of the C++ implementation (meh), or static
> errors of the IPDL spec (yay!).

That really depends on the error handling for dynamic protocol errors. If
the only recourse is to abort the child process, then that's a very serious
problem. In e.g. a channel protocol:

parent: channel.close() and child: channel.dataAvailable() are both async
messages. They race, with .close() winning somehow. That really shouldn't be
a fatal error, it should just cause .dataAvailable to fail gracefully or be
thrown away.

--BDS