[m-rev.] For review: Ignore dummy types in the IL backend

Jonathan Morgan jonmmorgan at gmail.com
Tue Sep 26 16:17:36 AEST 2006


On 26/09/06, Julien Fischer <juliensf at csse.unimelb.edu.au> wrote:
>
> On Mon, 18 Sep 2006, Jonathan Morgan wrote:
>
> > The below patch fixes most of the problems with the IL backend and
> > dummy types, by simply not recognising them.  I suspect that a better
> > fix would be to modify type_needs_lowlevel_rep to force dummy
> > variables to use a low-level representation, but that would require
> > more information than is currently passed into type_needs_lowlevel_rep
> > (eg. a module_info), and I'm not convinced that adding that
> > information is the right step (or even possible in all the places it
> > is called from).  Alternatively, it would be better to only disable
> > dummy var recognition in places where it causes type conflicts, rather
> > than everywhere.
> >
> > Note that this patch does not seem to have completely disabled the
> > recognition of dummy types - io.write() fails because it tries to call
> > an op_table method with an op_table, when the method has optimised the
> > op_table away.
>
> This is not the correct fix as it completely disables the optimization
> of dummy types (which is not what we want.) and AFAIK is actually working
> correctly in most instances on the IL backend.  The problem looks like
> it is mlds_to_il.m's handling of dummy types - I suspect that mlds_to_il.m
> wasn't updated correctly when the last lot of changes to dummy types were
> made.

I have a different fix which changes mlds_to_il.m, and does work.  I
just haven't got around to submitting it yet.

Jon
--------------------------------------------------------------------------
mercury-reviews mailing list
Post messages to:       mercury-reviews at csse.unimelb.edu.au
Administrative Queries: owner-mercury-reviews at csse.unimelb.edu.au
Subscriptions:          mercury-reviews-request at csse.unimelb.edu.au
--------------------------------------------------------------------------



More information about the reviews mailing list