[erlang-questions] rpc is bad? (was Re: facebook chat server)

Steve Vinoski vinoski@REDACTED
Mon May 26 03:07:37 CEST 2008


On 5/24/08, Ben Hood <0x6e6562@REDACTED> wrote:
>  If understand you correctly though, you are saying that the RPC style
>  of defining a endpoint to dispatch data to is limiting because if you
>  change the *server* side semantics or syntax, this means that you have
>  to change the client at the same time, right?

I think that's part of the problem, yes. Basically, there's that pesky
IDL specification shared between clients and the server; changes to it
are difficult to deal with because 1) it changes the generated stubs
and skeletons that are compiled into the client and server, 2) changes
to types and methods can easily ripple through the application code
that uses the stubs and skeletons, necessitating source changes and
recompilation, and 3) it affects not only the distributed interface
shared across the network between client and server but for certain
programming languages might also create binary compatibility issues as
well. This can happen when a client calls a "distributed" object or
service that is actually loaded locally in the form of a shared
library or DLL, on demand, by the RPC or distributed object
infrastructure.

BTW, is this topic general enough and relevant enough that we should
continue it here? I feel like I'm spamming the list!

--steve



More information about the erlang-questions mailing list