PEG.buildParser efficiency about pegjs

[rymohr]

Running a quick profile gave me the following biggest offenders:

indentMultilineParts   42.26%   peg.js?body=1 (line 4429)
formatCode             35.47%   peg.js?body=1 (line 4403)
replace                 8.57%   xregexp.js?body=1 (line 336)
map                     6.61%   peg.js?body=1 (line 59)

[rymohr]

Just forked and noticed indentMultilineParts and formatCode are nowhere to be found. Was this issue already addressed and just hasn't been released yet?

[rymohr]

If anyone runs into the same issue using v0.6.2 here's the hot fix until the next version gets released (which appears to have a new emitter):

--- peg.js
+++ peg.js
@@ -4441,7 +4441,9 @@ PEG.compiler.emitter = function(ast) {
     var args = Array.prototype.slice.call(arguments);
     var vars = args[args.length - 1] instanceof Object ? args.pop() : {};

-    return indentMultilineParts(interpolateVariablesInParts(args)).join("\n");
+    // We want fast parsers, not pretty ones!
+    // return indentMultilineParts(interpolateVariablesInParts(args)).join("\n");
+    return interpolateVariablesInParts(args).join("\n");
   };

   /* Unique ID generator. */

[dmajda]

As you correctly discovered, the code emitter is the slow part here. Both the old and new emiter have performance problems with your example.

As a workaround, you may try to split the big choice into many smaller ones. For example, instead of

start = a1 / a2 / a3 / a4 / a5 / a6 / a7 / a8 / a9

you can use

start = a13 / a46 / a79
a13 = a1 / a2 / a3
a46 = a4 / a5 / a6
a79 = a7 / a8 / a9

This causes less passes through the templating engine in the emitter.

I'll keep this issue open because the emiter performance clearly is a problem, but it isn't a big priority for me (performance of the generated parser is).

[rymohr]

Thanks I'll keep in that mind. Right now I'm still using the old version
with the patch above and it's working fine for my needs.

[ewindso]

Hi there -- Yep I'm experiencing the same thing. I thought of saving the parser as a string, however it comes out to > 3MB! Any more solutions, by chance? I'm using it with Node.js, and PEG version 0.7.0.

[dmajda]

The original Gist is gone (so this is not reproducible anymore), the code generator was rewritten meanwhile, and there are no users complaining about this bug since 2012. Closing.