Re: Simple constant folding in bison parser

jeremy@sw.oz.au (Jeremy Fitzhardinge) writes:
      One thing that occured to me was that the grammar can be made to recognize
      simple constant expressions, and generate a parse tree node for the value
      of the expression rather than the nodes for the operator and constant
      arguments.


This is handled by duplicating all the nonterminals into a ".c"
(constant expression) and ".nc" (non-constant expression) version:


expression : expr.c
| expr.nc


expr.c : factor.c
| expr.c '+' factor.c


expr.nc : factor.nc
| expr.nc '+' factor.nc
| expr.nc '+' factor.c
| expr.c '+' factor.nc


factor.c : CONST
| '(' expr.c ')'


factor.nc : VAR
| '(' expr.nc ')'


Etc.


Really, what you're doing is constructing a synthesized (passed-upwards)
attribute (constant vs. non-constant), but most LALR parser generators
don't support that. When you're parsing C, you want a lot of these
attributes, and you want the attributes to control some aspects of the
parsing. The result is that some productions are cloned 6 or 8 times.
Ugh.


Dale Worley Dept. of Math., MIT drw@math.mit.edu
--
[This certainly works, but I'd think it'd be a lot easier to handle
synthesized attributes in the value cells, where they're easy to
construct and pass up the tree. -John]
--

Return to the comp.compilers page.
Search the comp.compilers archives again.