🚀 go-pugleaf

I am writing a PL/M frontend to GCC 2.1.  I have now come to realize
that PL/M needs a two-pass compiler because you can use
procedures/functions before their declaration (and it's a strongly
typed language with parameter type and number checking).

(Actually, the Intel PL/M manual seems to say that you cannot use
procedures before their declaration, but their compiler works
otherwise.)

I thought of implementing the two passes like this:

In the compiler proper, the first pass scans the source with a grammar
that just recognizes the block structure and procedure/function
declarations, and inserts "forward" declarations at the beginning of
the appropriate block.  This pass stores the pre-lexed source in a
internal data structure (basically, a linked list of obstacks or
something like that).

The first pass also handles the peculiar "declare literally" concept
in PL/M -- a kind of textual macro (without parameters) that has
block-structured scope.

The "real" 2nd pass parser gets tokens from this pre-lexed structure.
Because of the inserted forward declarations, there is no problem with
using procedures/functions before declaration.

The 1st pass is automatically run throught the whole source the first
time (the 2nd pass) yylex is called.

Does this sound like a reasonable idea?

(Actually, at the very first there is a preprocessor, like cpp, that
handles conditional compilation etc. This is easy.)
--
Tor Lillqvist,
working, but not speaking, for the Technical Research Centre of Finland