On Jun 8, 2007, at 7:36 AM, Roman Redziejowski wrote:
On May 20 15:02:40 EDT 2007 Terrence Parr wrote:
"I typed in a Java grammar from spec, cleaning up issues. It never
terminates on some JDK 1.4 input files w/o memoization."
Hi Roman,
Interesting thoughts...thanks.
My parser was pretty fast with memoization ;) No attempt at
optimization.
A recursive-descent parser obtained by transcribing this PEG into
executable code accepts, without memoization,
all source files from J2SE 5.0 JDK, and a number of files from
other projects, alltogether almost 12000 files.
Of course, the performance is not brilliant, but what one can
expect if you need a whole pile of procedure calls to get to a
single letter of an identifier at the bottom of an expression, and
a lot of calls and backtracking to distinguish an identifier from
a keyword. However, the parsing time, measured in number of
procedure calls, is linear in the file size across the whole sample.
Despite the disadvantages, I still favor the separate lexer as it's
faster.
Ter
_______________________________________________
PEG mailing list
PEG@lists.csail.mit.edu
https://lists.csail.mit.edu/mailman/listinfo/peg
