Re: [MIT-Scheme-devel] compiler bug

[Chris Hanson]

OK, I've tracked it down to my last commit, which isn't yet pushed.
It looks like my change tickled some bug in the compiler.  I'll chase
it down and fix it before I push.

On Sat, Sep 19, 2009 at 3:42 PM, Chris Hanson <address@hidden> wrote:
> Thanks.  So far it's looking like my last commit, but I'll know in an
> hour or so.
>
> On Sat, Sep 19, 2009 at 3:31 PM, Joe Marshall <address@hidden> wrote:
>> If there's anything I can help with, let me know.
>>
>> On Sat, Sep 19, 2009 at 12:58 PM, Chris Hanson <address@hidden> wrote:
>>> Not yet.  I've had trouble because some of the commits are broken and
>>> require patching.  Also, I got sloppy late last night so I have to
>>> re-do some of the tests.  I just semi-automated the testing process
>>> and made a more careful log of the known results.  I should have
>>> something in a few hours.
>>>
>>> On Sat, Sep 19, 2009 at 10:28 AM, Joe Marshall <address@hidden> wrote:
>>>> Any luck?
>>>>
>>>> On Fri, Sep 18, 2009 at 8:48 PM, Chris Hanson <address@hidden> wrote:
>>>>> Not sure about that.  I'm doing the bisection now.  The bug isn't
>>>>> present in commit 745a16218a43692d2c9ecdad72d1bab73fab0522 from Sept.
>>>>> 7, so it's pretty recent.
>>>>>
>>>>> On Fri, Sep 18, 2009 at 8:35 PM, Joe Marshall <address@hidden> wrote:
>>>>>> On Fri, Sep 18, 2009 at 1:57 PM, Chris Hanson <address@hidden> wrote:
>>>>>>> FYI, while testing my most recent changes, I ran across a recent
>>>>>>> compiler bug.  To reproduce: using the current compiler, do a fresh
>>>>>>> compile from sources.  Then use that compiler to do another fresh
>>>>>>> compile.  The second compilation hangs during the compilation of
>>>>>>> "cref/anfile" (the very first file to compile).  The problem appears
>>>>>>> to be that the compiler is running map over a list of items, but that
>>>>>>> the list is circular.
>>>>>>
>>>>>> Any indication of what sort of items are in the list?
>>>>>> Any idea of which procedure is being compiled?
>>>>>> What compiler phase?
>>>>>>
>>>>>> I fixed a bug in SF that was preventing things like
>>>>>> ((access vector-ref system-global-environment) ....)
>>>>>> from being optimized.  I tested it pretty well on a second and
>>>>>> third generation syntaxing of itself, but maybe it created some
>>>>>> scode that works for the interpreter, but isn't quite kosher
>>>>>> for the compiler.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> ~jrm
>>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> ~jrm
>>>>
>>>
>>
>>
>>
>> --
>> ~jrm
>>
>