On 7/6/07, Robert Dodier <robert.dodier at gmail.com> wrote:
>
> On 7/5/07, Stavros Macrakis <macrakis at alum.mit.edu> wrote:
>
> > comm.lisp works properly when compiled, but not when interpreted.
> This appears to be a GCL-specific bug:
>
It does seem likely that the problem is with GCL if not only all other Lisp
interpreters and compilers, but even GCL's own compiler, agree. Then again,
I have seen stranger things. So let's find the bug and report it to the GCL
maintainers.
> Shouldn't we be testing the interpretive version of Maxima at least
> > occasionally?
>
> Certainly anyone is free to do so if they wish, but I'll
> recommend that it not become part of the
> standard release testing. I don't see the point of it.
There are a couple of reasons:
1) If there is a difference in behavior between the interpretive and
compiled version, that points to some underlying problem. Either the Maxima
code is buggy or the Lisp implementation is buggy. These bugs may affect the
behavior of the code in cases where compilation/interpretation is not the
issue and which the test suite doesn't cover (which is a lot). After all,
compilation and interpretation should be semantically indistinguishable,
compilation simply being an optimization.
2) Debugging and experimenting with code are easier with interpreted code.
If we can't count on interpreted code to work the same way as compiled,
we're in trouble.
I have no opinion on GCL vs. SBCL. I'm sure others have informed opinions
on this. But as long as we *are* using GCL as our standard implementation
on our most popular platform, I am surprised that you are willing to ignore
bugs in it.
-s