Hi Kaspar, Thanks for your reply.
In fact, the Ruby's csv parser uses some quite advanced regular expressions, optimized for speed: https://github.com/circle/fastercsv/blob/master/lib/faster_csv.rb#L1775 However, although this parser is fast, it is very difficult to maintain due its heavy use of regular expressions. Parslet, on the other side, reads almost like normal bnf, which I like a lot. Keep up the good work! Stefan On Mon, Sep 5, 2011 at 15:53, Kaspar Schiess <[email protected]> wrote: > Hi Stefan, > > >> As I plan to use my parser in a real application, but cannot do so if it > is > >> too slow, I took a look at the code by could not find anything that > might > >> explain the long execution time. > >> > >> Therefore I would be glad if someone could help me find this bottleneck. > > The csv library uses a hand-written parser; such a parser (suitably > constructed) will always be faster than a parser generated using a > general purpose parser generation lib. (such as parslet, treetop, ...) > > I've taken the liberty of incorporating your csv parser in my benchmark > suite. Here are the results of comparing treetop with parslet on that > parser: > > CSV parsing > size parslet treetop > 253: 0.020 0.010 > 5293: 0.410 0.120 > 10207: 0.910 0.310 > 15247: 1.430 0.430 > 20161: 1.940 0.610 > 25201: 2.420 0.900 > 30115: 3.180 1.060 > 35155: 3.900 1.100 > 40069: 4.130 1.360 > 45109: 5.080 1.680 > 50023: 5.840 1.900 > > Nothing new here. I would like parslet to perform the same as treetop - > I am working on that. But I will never achieve performance of the csv > stdlib. > > I guess the reason for this is that parslet&treetop do a lot more work > than csv does: Overhead ;) > > regards, > kaspar > > >
