It isn't MIG's job (or any other IDL compiler's job) to abstract messages. MIG's job is to *describe* a pre-existing abstraction layer. The interface *is* the abstraction layer.
On Tue, Oct 27, 2009 at 12:23 AM, <[email protected]> wrote: > Hi, > > On Thu, Sep 24, 2009 at 12:19:41PM +0300, Bahadir Balban wrote: > > > So why are you in favor of IDL compilers? To me that is also an > > abstraction that is a compromise. You could simply use a library with > > helper functions, and know what is going on, instead of design a whole > > system around compiler-generated communication idioms. > > I'm not familiar with other IDL compilers; but MiG at least doesn't > really do much in a way of abstractions. It rather describes what will > be sent in a message almost 1:1. It is a means to say, "create a pair of > stubs that will assemble the message in the following way". > > In as far as it is a compromise (though I'm not sure we are really > giving up anything relevant here), it's definitely a good one. It saves > the complexity of building messages per hand (and keeping client and > server stubs in sync), without really taking away understanding how the > messages look. > > -antrik- > > >
