Re: [j-devel] ANSI test failures RESTART-CASE.{29, 30, 31} [interpreted mode]

[Erik H. <eh...@gm...>]

Since I'm answering myself anyways, I might as well top post:

The fix was much easier than the bits mentioned below: restart-case is
a macro macro-expanding subforms, which means it should itself take an
environment parameter and use that when expanding the subforms.
RESTART-CASE didn't do that. Now it does :-)

This takes us to only 43 failures in the ansi test suite (interpreted
mode) on my system.

Bye,

Erik.

On Fri, Apr 3, 2009 at 9:16 AM, Erik Huelsmann <eh...@gm...> wrote:
> On Thu, Apr 2, 2009 at 11:54 PM, Erik Huelsmann <eh...@gm...> wrote:
>> Ok. After studying both our code (using TRACE) and the CLHS, I found
>> the issue with the 3 RESTART-CASE failures. Unfortunately, the cause
>> is not what I had hoped for (which is that we found issues with more
>> special variable bindings) but it's not. The good thing however, is
>> that the failure is really related only to macroexpansion.
>>
>> This is what happens:
>>
>> MACROEXPAND doesn't expand subforms established by MACROLET
>>
>> When expanding RESTART-CASE, the macro inspects the first atom in the
>> restart form, in order to see if the restart needs to be associated
>> with some condition or another. It establishes that by looking for the
>> ERROR, CERROR, SIGNAL or WARN symbols (it may use others; see CLHS
>> RESTART-CASE).
>
> The passage I'm referring to is (in RESTART-CASE definition):
>
> "
> If the restartable-form is a list whose car is any of the symbols
> signal, error, cerror, or warn (or is a macro form which macroexpands
> into such a list), then with-condition-restarts is used implicitly to
> associate the indicated restarts with the condition to be signaled.
> "
>
>> When RESTART-CASE looks at the form in casu, it's finding %m or %s
>> instead of (ERROR ...) to which the %m and %s expand.
>>
>> It looks our interpreted MACROEXPAND needs to be a bit smarter...
>
> I've been thinking about this; our precompiler does macro-expansion.
> How about using the precompiler to expand the forms? For example by
> renaming the built-in MACROEXPAND -> %MACROEXPAND. Then, in boot.lisp
> one of the first things to do is to define MACROEXPAND as follows:
>
> (defun macroexpand (&rest args)
>  (apply #'%macroexpand args))
>
> Which would get redefined in precompile.lisp as (something like):
>
> (defun macroexpand (form &optional environment)
>   (let ((*compile-file-environment* environment))
>     (precompile-form form)))
>
>
> Then, after the precompiler gets loaded, full macro expansion becomes
> available as per the requirements of the test.
>
>
> Comments?
>
>
> Bye,
>
> Erik.
>