🚀 go-pugleaf

>Thanks Oliver - The very first thing I should have criticized in my post last week
>was Microsoft Fortran for the Mac.

>Well, I haven't had to touch Microsoft Fortran since then (I usually work in
>680x0 or C), but I did hear something very relevant and very curious.  It seems
>that Microsoft didn't actually do the compiler.  It was really a company called
>Absoft, which makes Fortran compilers.  Fine.  Well, they have THREE versions of
>a Fortran compiler for the Mac.  It is the low-end one that they sold to Micro-
>soft. 
>Heh - we still haven't heard from the Microsoft boys yet on this particular
>discussion.  I hope they're either planning a revolution or looking for new jobs.

Well, the Microsoft compiler has lots more wrong than that. First, all
the programs I wrote with it refused to run on Mac II's (their bugs).

Second, the code generation was abysmal. For example, a simple statement

     IF(i.lt.j)goto 100

should compile to a test and a branch on condition. How do they do it?
They test, they load the flag into a register as a byte, extend the
byte to a word, then the word to a long. Then they test the long,
and branch on that. That kind of code is INEXCUSABLY bad!

Third, and worst of all, this dumb compiler generates CODE for data
statements! Instead of just compiling the data into a static data section,
they store it as literals in the instruction stream. For example
       character*1 x(5)
       data x(1)/'a'/,x(2)/'a'/,x(3)/'r'/,x(4)/'g'/,x(5)/'h'/

generates 7 bytes of code for each character , 35 bytes total.

Interestingly enough, Microsoft's product for the IBM-PC is excellent.
Guess which computer I use now (hint: it has a very nice batch file
processing system right in the operating system).