Hi, On 10/24/2015 09:52 AM, Timo Paulssen wrote: > On 24/10/15 09:40, Moritz Lenz wrote: >> Hi Gabor, >> >> On 10/24/2015 09:26 AM, Gabor Szabo wrote: >>> The Devel::NYTProf helped me a lot locating the source of slowness on >>> the Perl Maven site. >>> Is there something similar for Perl 6 so I can try to improve the speed >>> of the Perl 6 Maven site too? >> Rakudo has a --profile command line option, which generates a HTML file >> containing a single-page JS app that presents the profiling data. >> >> But you should take care to only profile short program executions that >> way (a few seconds at most), otherwise the result will be unusable in >> the browser. >> >> Cheers, >> Moritz > > It's a common tip to "only profile short program executions", but that > misses the actual point by a few centimeters. The things that makes the > profiler super slow are:
The profiler, or the web app generated by the profiler? Curently I mostly care about the latter. > 1) having a gigantic amount of different subs and methods that get > touched by the program at some point > 2) having very, very deep call chains > 3) having the GC run thousands of collections > > if I understand correctly. > > > Recently, Coke++ switched the profiler JS app from a kind of old version > of angular.js to a newer version that has had lots of performance > improvements done to it by the google devs; i expect that helps > tremendously, though i have not actually measured anything beyond a new > feeling of "snappyness". I've tried to profile htmlify.p6 in the perl6/doc repo; it has a --sparse option to make it process fewer files and thus generate less profiler output. With --sparse=200, it takes about 22s to run on hack.p6c.org. I managed to run it with --profile, but the generated profile still doesn't work in any browser I tried. See http://hack.p6c.org/~moritz/profile-1445674278.30788.html for the output. If anybody manages to get it to render, please send me some screenshot :-) I don't know if htmlify.p6 has deep call chains; it does some recursion along the Pod tree, but I'd be surprised if that's more than 20 levels deep. Cheers, Moritz
