[fpc-devel] constant propagation // kinda

Martin Frb lazarus at mfriebe.de
Wed Oct 1 22:37:26 CEST 2025


Just came arccos a case for optimisation. I was wondering if its worth a 
feature request.

Especially since I guess the compiler may already have most of the info 
needed.

var a: integer;
begin
   A := 42;
   // some code
   x := x + a;


In the entire procedure
- A is set before it is ever read
- A is set in an unconditional path
- A is set only/exactly once, and never changed.

So basically A is a constant.

Now given that the compiler can tell me
- if a variable is not used at all (gives a hint)
- if a variable is assigned but then never used (gives a hint)
It would seem the compiler should have (nearly?) all knowledge to see 
that A is a constant?

Of course at first code like this makes little sense. Isn't expected to 
exist. Except, if that for example happens in a generic.
And the value for A come from a generic param, e.g. a "record" passed 
in, that could have the value in a field, or alternatively have a 
declared const for the value.
Then the code written for the generic wouldn't know, until the compiler 
applies the param.


More information about the fpc-devel mailing list