Re: fortran 2003 compiler

Hello,

Richard Maine wrote:

<snip>

This came up when I groused about f2008 as being premature because it
was likely to be released before there were any f2003 compilers. I was
told that there probably would be by then. However, I still consider
f2008 premature. Even if one or two compilers are out by then, they
certainly won't have had much user experience on them. And even more so,
any such user experience will not be able to significantly influence the
content of f2008, since the content was allegedly set some time ago.

I think 'premature' is a good description of f08. That judgment
may be somewhat mitigated by the notion that f08 was supposed to be
a small revision of f03. But in fact, it is not. The main reason is,
of course, co-arrays. OTOH, co-arrays are, in the main, written
against f95, and are orthogonal to most, at least, of what's new
in f03. And there is at least one working implementation
of co-arrays today, so the end product is well-known.

But even removing co-arrays, f08 is still too big by far.

For starters, f08 already (before any further work) includes
the Enhanced Modules feature (a/k/a submodules).

The other two major features are the Macros and the Bits type.

I believe that the Macros (and their Blocks) should be removed,
and that Bits should be simplified to require
only the natural sizes supported by the hardware. This is
because Macros don't practically add any new functionality (anyone
who wants to have Macro-like functionality already has access
to several preprocessors) and I believe they will be intrinsically
very difficult to debug due to several design decisions taken together.

While the Bits type is intended to provide a type for BOZ constants
and to allow comparisons to work as expected without introducing
unsigned integers, I anticipate that the major actual use
of Bits will be to evade the type system when using
procedure arguments and type components. Bits of the sizes of other
intrinsic types is sufficient to support all of these uses
and greatly simplify the relationship of bits and kinds.

There are a few other small features which could be removed.

I expect two features to be requested during the Public Comment
period for f08. These are control of host association (a la control
of use association) and default values for optional arguments.
If a sufficiently broad request for these features is made,
either or both of them have a realistic chance of being included.

So ATC, I think Richard's comments above are right on the bull's eye.
The question is how to respond.

--
Cheers!

Dan Nagle
Purple Sage Computing Solutions, Inc.
.

Relevant Pages

  • Re: fortran 2003 compiler
    ... was likely to be released before there were any f2003 compilers. ... I think 'premature' is a good description of f08. ... OTOH, co-arrays are, in the main, written ... The other two major features are the Macros and the Bits type. ...
    (comp.lang.fortran)
  • Re: Another C99 challenge
    ... reason to believe that GCC is lying about their own work? ... C99-specific features anyway, at least in delivered code, until our ... I wouldn't mind if GCC did fully support C99, ... compilers using a "reasonable" variety of features new to C99. ...
    (comp.std.c)
  • Re: R6 Counterproposal
    ... Can't compilers which struggle enough still be able to theoretically ... The above statement contains a hidden assumption, which is that it applies to programs implemented in programming languages whose features are implemented efficiently. ... SCM, which works like this, is a fairly quick scheme implementation even though its just an interpreter. ...
    (comp.lang.scheme)
  • Re: Segmentation Fault....
    ... lcc-win32 doesn't support those features. ... Very few implementations support all of C99. ... diagnose *all* attempts to use C99-specific features. ... conforming C99 compilers, and none in widespread use, those of us who ...
    (comp.lang.c)
  • Re: Green Hills vs. Freescale CodeWarrior for ColdFire development
    ... intend to use those features to enhance code correctness. ... And we end up back at the original question: Which of the compilers would ... support the largest subset of the C++ specification? ...
    (comp.arch.embedded)