[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
Re: mapping 754R to specific languages
What distinguishes reproducibility from other modes such as rounding and
even alternate exception handling (to some extent) is that it does not
map to hardware structures. You do however have a point with:
The forerunner of reproducibility is the extended rounding precision mode
of 754. It wasn't enough to do the job however.
Reproducibility, as understood by the application programmer, might be
implemented by exploiting various hardware features behind the scenes.
What remains to be done is to describe the limitations that the user
who is concerned about reproducibility has to be aware of. It has
occurred to me that most of my objections could be handled by documented
restrictions -- my concern was that the restrictions might leave no room
for writing the intended programs.
In my current proposal for "clause 10" I have tried to lay out
restricted definitions of reproducibility,
according to my notions (within a language)
and according to Bob Davis (across languages).
The latter constrains the programmer quite a bit more than the former.
http://754r.ucbtest.org/msc-ballots/ee.pdf
As for PL/I and widenTo, I have to confess that I have not used PL/I in
almost 30 years
As a general issue, 754R assumes that languages that aren't evolving in a
standards sense will be extended, if at all, by ad-hoc library/class
definitions. It would be a pious hope rather than a realistic expectation
that if two vendors were motivated to create such extensions, they would
be compatible with each other.
My point was that if a language does not leave any
choice to the implementation, then widenTo should not be forced upon it.
Since my "clause 10" requires that a universal expression evaluation mode
be available within a language, I revised it slightly to make all the
widenTo features SHOULD again rather than SHALL.