asciidoc calls frequently re.compile. In case of cpython re.compile uses cache, where ironpython performs compile unconditionally. https://ironpython.codeplex.com/workitem/35146 --pawel
On Mon, Apr 21, 2014 at 3:09 PM, Pawel Jasinski <pawel.jasin...@gmail.com>wrote: > > > > On Fri, Apr 18, 2014 at 12:33 PM, Jeff Hardy <jdha...@gmail.com> wrote: > >> On Thu, Apr 17, 2014 at 11:32 PM, Pawel Jasinski >> <pawel.jasin...@gmail.com> wrote: >> > I think it is simple .net regex which is slower. >> > I did the following: >> > 1. Proved that asciidoc run under python (cygwin) evaluates the same >> regular >> > expression as under ipy - there are no bugs which make ipy evaluate >> more REs >> > 2. Captured the evaluated regular expressions in a form suitable for >> .net >> > and python >> > 3. Write trivial program in c# to evaluate captured REs. (6 seconds) >> > 4. Write trivial program in python to evaluate captured REs (0.4s for >> > python3.2 and 0.65s for python 2.7.5) >> > I also got rid off ipy/c# quirks (extra named groups), to get functional >> > equivalent and no conversion overhead - no change >> > Number of captured REs is 46253. >> > For this particular sample, cpython implementation if regular expression >> > appears to be 12 times faster than .net implementation >> >> Can you put the sample code in a gist? Maybe it's hitting a >> pathological case in the .NET regex engine or something, because >> that's a crazy difference. >> >> - Jeff >> > > The re evaluated by ipy and cpython are the same and the end result of > ascii doc is the same but, the number of matches reported is different. > I will keep digging. > --pawel > >
_______________________________________________ Ironpython-users mailing list Ironpython-users@python.org https://mail.python.org/mailman/listinfo/ironpython-users
