Subject: Re: gcc optimizer bug in netbsd-1-6 on alpha (gcc 2.95.3 20010315 (release) (NetBSD nb3))
To: Jonathan Stone <jonathan@DSG.Stanford.EDU>
From: Nathan J. Williams <firstname.lastname@example.org>
Date: 08/15/2003 17:31:09
Jonathan Stone <jonathan@DSG.Stanford.EDU> writes:
> Nate, I wonder, did you never take a networking class?
Nathan, please. I took 6.033 at MIT, which is a survey class that
covers a good deal of networking, but nothing specifically labeled as
a networking class. There's clearly an implicit "If you had, you would
know <xxx>" in there. I don't like guessing games; want to spell it
> >A good compiler should be able to see a copy to an intermediate
> >structure and a subsequent use of that structure and elide the copy
> >entirely, accomplishing the type-punning in the implementation rather
> >than in the expression. This seems like a good goal to work towards.
> No, that's a bad idea, that way lieth ML. A _good_ compiler for
> systems work should let well enough alone and access the data inplace,
> as typical C compilers have done for the past 20 years and more.
The fact that "that way lieth ML" is exactly what makes it a good
idea. We should be working on abolishing C, not justifying
it. In-place access to data should be a compiler-level optimization.