[plt-scheme] Error reporting and batch compilation (was: fluid-let-syntax may get flushed)

Lo, on Wednesday, August 11, Matthias Felleisen did write:

> 
> On Aug 11, 2004, at 10:39 AM, Joe Marshall wrote:
> 
> >   For list-related administrative tasks:
> >   http://list.cs.brown.edu/mailman/listinfo/plt-scheme
> >
> > "Richard C. Cobbe" <cobbe at ccs.neu.edu> writes:
> >
> >> In general, my tools should give me as many error messages per run
> >> as possible; it wastes less of my time.  (This is one thing the C
> >> folks got right; why has the Scheme community forgotten this?)
> >
> > As many *legitimate* error message, please.

Ideally, yes.  But I'll gladly accept several spurious error messages in
order to get more legitimate ones.  It's a question of balance.

> The C people didn't get this any "rigther" than the Scheme people.
> They were forced to report as many type and syntax errors in one
> pass as possible because they were and are batch people, who just
> don't understand how incremental work helps people.

Right, but as execution time increases, the distinction between batch
and interactive development decreases.  It's really not very hard to
write a Scheme program whose execution time (by which I mean time
between hitting the `execute' button and getting a prompt back) is
comparable to running make.

Case in point: the test cases for my PLT redex implementation of Jacques
take forever to execute, largely because they pull in SchemeUnit, which
in turn pulls in all of the framework stuff.  Compiling to .zos helps
somewhat, but then you're back to the batch model.

If someone can suggest a testing strategy that doesn't require hitting
execute after every change (or even most of them), then I'd love to hear
about it.  For full credit, the testing system must also work for
changes in a different module than the one that contains the test cases.

Richard