Project 45: Fortran as an assembler language

Madison 1982:
=============

Feldman summarized the characteristics of an assembly 
language as simplicity, approximately equal execution cost
per instruction, low level, simple data types, use of
addresses and registers. It is unsuitable for human users 
except as a last resort but is suitable as a target for 
computer generation. Fortran has most of these 
characteristics and the great advantage of wide 
availability. By using preprocessors, it is easy to provide 
more convenient syntax, general loop constructs and more 
general operators (e.g. COMPLEX, array operators); it is 
more difficult to provide new data types, inhomogeneous data 
structures, different i/o mechanisms, simple internal 
procedures (because of problems with array arguments); it 
is very difficult to provide recursion, pointer variables. 
Fortran 8x looks as if it will help by providing better 
floating-point facilities including the equivalent of DOUBLE 
COMPLEX, name hiding in libraries, humane calling sequences, 
data structures, bit strings, stack allocation of dynamic 
storage but does not look as if it will provide pointers.

Gentleman felt that it was helpful to know to what a 
pointer points. Feldman asked how a heap allocator could be 
implemented with typed pointers.

Feldman went on to discuss list processing, whose 
characteristics are frequent and dynamic storage allocation, 
objects of differing sizes, objects containing other 
objects, dynamic determination of sizes and structures and 
the need to manipulate references and atoms.

Rice remarked that the advantages of aiming only for a 
target language for preprocessors would be a small Fortran 
8x language and efficient compilation. Smith remarked that 
X3J3 has always been concerned about the efficiency of the 
object code. Feldman said that the size of the present 
language must mean some loss in execution speed (say by 5-
10%), a much more expensive compiler (perhaps by ten times) 
and a slower-running compiler (perhaps by three times). 
Gentleman saw a need for exception handling, concurrency, 
parameter passing and scooping rules. Smith saw the need for 
array processing. Dekker felt that some parallel mechanism 
is needed, e.g. the parallel do. Feldman said that this 
would be good for preprocessors.