Vanilla, Chocolate or Strawberry?
One of the issues we are facing in writing this standard is that there are
different constituencies. Trying to satisfy everybody leads to compromises
that may annoy everybody, but providing a clean and consistent specification
that satisfies one constituency may leave another with no choice but to avoid
the standard completely.
The differences of point of view are usually restricted to just a few issues,
with general agreement on everything else. Is there a way to take advantage
of this?
I suggest that there is: Define a small number of Flavours of the interval
environment, and in the standard. Most of the standard would apply to all
flavours, but some specific issues -- e.g. unbounded intervals -- could be
handled in a flavour-dependent manner.
At the implementation level one or more flavours could be supported;
if more than one, a static attribute would be defined that selects the
flavour. This notion is used already (typically by means of compiler
flags or source-code pragmas) for things like floating-point flavour
(for IBM Z there is a choice of HFP vs BFP), character set flavour
(Ascii vs Ebcdic vs Unicode), pointer size etc. This setting would
influence which libraries a program is bound with, and what kind of
code is to be generated. It would definitely be a compile-time choice.
For interpreted languages it could be an irrevocable choice made by
an early statement (before any flavour-dependent facilities are used).
There could be some flexibility of scope -- we have interchange formats
to support communications between different-flavoured components.
Am I completely off my rocker?
Michel.
---Sent: 2012-05-06 18:10:47 UTC