Re: Publishing binary interfaces [was Re: [PATCH] Move "types deeply equal" code from py-type.c to gdbtypes.c]

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Doug" == Doug Evans <dje@sebabeach.org> writes:

Doug> Some people IIRC were *profoundly* against publishing binary interfaces.

Not sure if you are referring to me.

If so, then I think you've misunderstood me.  This is understandable
given the deficiencies of irc.

If not, I suppose whoever it is can speak for themselves.

Doug> Is the community changing it's mind on binary interfaces?

I'm ok with it in principle, but the details matter.  It's hard to
discuss this in the abstract.

For example, there are many parts of gdb that I think are plainly
unsuitable for ABI promises.  So if your plan involves fixing the ABI,
then I will be against it.

I'm also skeptical of this on doability grounds.  It's harder to publish
a library than a program; and my view is we barely have enough developer
time to address the glaring internal deficiencies in gdb, let alone make
it all of "publishable API" quality.

That said, I'm happy to note that my skepticism isn't a determinant of
whether it will succeed or fail.  I'll help if I can.

I think if you really want to pursue this, you ought to come up with a
plan covering what you intend to do, how you intend to do it, what the
results will look like, how the API will be managed, what additional
constraints this will place on gdb development, etc.  This will clarify
whether it is something I would support.

Doug> Do people actually envision dlopen'ing GDB's Python extension?

If there is some concrete benefit.

Tom