Q4071@pucc.Princeton.EDU (Interface Associates) (07/14/89)
Thank you all for your many replies to my request for information
on FORTRANs for the PC. To refresh everyone's memory, I was looking
for a fast compiler which can use extended memory from DOS, although
I was willing to learn UNIX if that was required to get a good compiler.
I also emphasized the need for a usable debugger, or at least good
error tracing because of my horrible experiences with MicroSoft FORTRAN.
I placed FORTRAN'77 far down my list as a way of indicating that if
someone had a '66-based compiler out there which was otherwise perfect,
then I would re-convert. Interestingly, some responders indicated that
'of course' anything reasonable would be F'77. In point of fact, my
experience on the mainframe indicates that FORTRAN HX ('66 extended)
code runs at least a factor of two faster than VSFORTRAN ('77 extended)
code, so had there been a fast non-77 compiler, I would have looked at
it. None mentioned were cited to be faster than their '66 counterparts,
so '77 it is, it seems.
Fifteen readers responded with helpful comments beyond quoting
advertisements. Many of these offered a wide range of commentary,
while others identified one product the person considered superior.
There was no consensus. I received the following general advice.
1. Don't use any Microsoft product.
2. Use UNIX.
3. Don't even THINK about UNIX, none of the compilers work.
4. Don't use Microsoft/XENIX.
5. Don't use any MS-DOS oriented compiler.
6. Don't use Microsoft for OS/2.
Actually, there was something approaching an anti-Microsoft consensus,
if on no other point.
Eight FORTRANs, one other compiler, and one alternative approach were
suggested. I tabulated votes into first choice, second choice, and
honorable mention. Since Microsoft's two products were only mentioned
in negative or neutral terms, I exclude them from the survey. One
correspondent indicated that there was no good choice, and got a
category of his own.
Name First Second Honorable mention
MS-DOS compilers...
Microway NDP 3(20%) 2(33.3%) 1(33.3%)
Lahey 6(40%) 2(33.3%) 1(33.3%)
Prospero 1(6.67%) 2(33.3%) 0
Salford 1(6.67%) 0 0
RM FORTRAN 0 0 1(33.3%)
UNIX compilers...
XENIX/LPI 1(6.67%) 0 0
other compilers and approaches...
Gauss/386 (not FORTRAN) 1(6.67%) 0 0
use a remapping product 1(6.67%) 0 0
A pox on all their houses 1(6.67%) 0 0
----------------------------------------------------------------
Total 15(100%) 6(40% of totl) 3(20% of totl)
Virtually everybody agreed that Lahey FORTRAN is a good choice. I was
warned about limitations on equivalenced arrays, something I have had
to do without before and can do without again, and lack of 1-byte
integers, which DOES concern me. I talked to the people at Lahey, and
they indicated that they have a new version which does not have any
restrictions on number or size of equivalenced arrays. (v2.0 of F77EM32)
It is also supposed to support most VSFORTRAN and VAX extensions. I have
used both, and the extensions Lahey does not support (other than Logical*2
and Integer*1) I have never used.
The conclusion is that I will probably get Lahey (although I am still
looking into MicroWay.) As someone who is used to doing serious
computing only on a mainframe, I thank everyone for the assistance.
=========================================================================
Robert A. West c/o Interface Associates, Inc. (Q4071@PUCC)
US Mail: 666 Plainsboro Rd. Office Commons, Suite 1A, Plainsboro NJ 08536
Voice : (609) 275-5711