Re: LL(1) vs LALR(1) parsers

Bill Leonard <Bill.Leonard@mail.hcsc.com> wrote:
>If I were in charge of a compiler project (and I have been), I'd much
>rather devote engineering resources to making the generated code better or
>the compiler faster or any of a number of other improvements than in coding
  ^^^^^^^^^^^^^^^^^^^
>a parser.


Bingo! You've just answered why recursive descent parsing is making a
comeback. I don't have the papers handy, but if you look at Ken
Thompson's paper on the Plan 9 C compiler (New C), he says (now that
the language has stablized) he'd write a recursive descent parser,
because the compiler was spending 35% - 40% of its time in the yacc
generated parser. Niklaus Wirth's Oberon compilers (and single pass
Modula-2, for that matter) are quite fast (on slow hardware, by
contemporary standards) partly because they parse quickly. (We're
talking speeds of 5 KLOC/second if I remember properly. On 1-2 MIPS
CPUs...) You may object that Wirth's compilers are not
highly-optimizing, but non-optimized compile/link times are on just
about every programmer's critical path.


--
Drew Dean
<ddean@cs.princeton.edu> http://www.cs.princeton.edu/~ddean PSTN: 609-258-1797
--

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