Re: [Gcl-devel] Interpretation of random tester output

[Camm Maguire]

Sorry, one last bit -- do you think it is valuable to put the bugs I've
found and closed into misc.lsp?  I won't have time to do further hand
pruning, but the auto-pruner left me with small forms.

Take care,

"Paul F. Dietz" <address@hidden> writes:

> Camm Maguire wrote:
> > Greetings!  OK I guess your machine doesn't speak gzip.
> > I've placed the uncompressed file at
> > http://people.debian.org/~camm/e.out
> > Thanks again!
> 
> The long printout is the pruner successively simplifying
> the form.
> 
> The final part of that is the reduced failing test
> case.  The fields mean:
> 
> :vars       Names of variables bound in the lambda forms
> :var-types  Their types
> :vals       The values that were to be used as arguments to the
>              lambda forms.
> :decls1
> :decls2     The OPTIMIZE decls for the optimized/unoptimized forms.
> :form       The body of the lambda forms.  This is what is printed
>              by prune-results as it does its work.
> 
> :optimized-form
> :unoptimized-form  The two lambda forms built from those fields.
> 
> In gcl, the unoptimized form is not compiled, but instead a form
> like (funcall (function ,unoptimized-lambda-form) ,@vals) is constructed
> and eval-ed.
> 
> The test fails if the compile of optimized-form aborts, if either evaluation
> aborts, or if the return results are not the same.
> 
> In this case, the compile aborted.  I've reduced the form a bit
> more and pruned off unnecessary lambda parameters and type declarations:
> 
> ;;; gcl (found by Camm)
> ;;; Error in COMPILER::CMP-ANON [or a callee]: The function NIL is undefined.
> (deftest misc.286
>    (funcall
>     (compile
>      nil
>      '(lambda (e)
>         (declare (optimize (speed 1) (space 3) (safety 3) (debug 3)
>                            (compilation-speed 1)))
>         (flet ((%f11 (f11-2) 0))
>           (%f11 (unwind-protect
>                     e
>                (tagbody
>                    (let* ((v4 (unwind-protect (go 0)))) 0)
>                    0)
>                   (logand (handler-bind () 0)))))))
>     10)
>    0)
> 
> BTW, once LOOP-RANDOM-INT-FORMS finishes reducing test cases, you can abort
> it and get at the test cases in the list $y.  The function (f i) will compile
> and evaluate the optimized form of (elt $y i) on the apppropriate vals; (g i)
> will do the same on the unoptimized form (f, g, and $y are all in the CL-TEST
> package.)
> 
>       Paul
> 
> 
> _______________________________________________
> Gcl-devel mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/gcl-devel
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah