[Axiom-developer] Re: defintrf bug

[Stephen Wilson]

*,

Stephen Wilson <address@hidden> writes:
> I will try again with my original idea to make compWithMappingMode
> smarter, to lift only those identifiers which it can prove are bound.

OK.  I believe we can bury this bug finally.

Attached is a patch against
compiler.boot.pamphlet:compWithMappingMode.  Im not including a
documentation change as I think a clean patch addressing this bug in
isolation is desirable. I will follow up with a patch which documents
the affected function within the next few days (as well as salvaging
the documentation from my previous patch).

Locally, this change appears to affect the generated lisp for DFINTTLS
only.  The change appears to pass all tests.  I am confident enough at
this point to `sign-off' this change.

The principle modification is basically one line.  Call getmode on an
identifier to see if it has a type in the current environment.  I was
reluctant to pursue this path initially because I could not convince
myself that this was a `correct' usage of the function.  This is due
to my lack of knowledge of the compiler as a whole.

Therefore, for those interested, find attached the file cwmm-test.spad
which basically stress tests the patch.  This gives me confidence that
the change is working as I expect it to for general use cases --
esp. when examining the generated Lisp.  I would like to see a
facility in place where spad code such as this could be tested as part
of the regression suit.  Something like AldorUnit for spad would
certainly be of great help.  Perhaps I can find the time to get
something like that in place.


Take care,
Steve

changelog.patch
Description: Text Data

compiler.boot.pamphlet.patch
Description: Text Data

cwmm-test.spad
Description: Binary data