17:36 <kostya> hi

17:37 <kostya> i experiencing strange perfomance: https://gist.github.com/kostya/c1748e5dde4aefa4cb14

17:37 <kostya> asterite, any ideas?

17:38 <asterite> Let me see...

17:43 <asterite> @kostya I have no idea… the generate ll code is very different if you add that third yield

17:44 <kostya> :)

17:44 <kostya> maybe excess limit for inline

17:44 <asterite> Maybe

17:44 <asterite> Probably some llvm optimization thing

17:45 <asterite> Blocks are always inlined (by us), I don't know if llvm is smart enough to extract that to a function

17:45 <asterite> but if that's the case, maybe that's why it becomes slow

17:56 <kostya> how to generate ll code?

18:00 <asterite> You can use the -ll flag

18:00 <asterite> it generates it inside a .crystal directory

18:01 <asterite> Normally you get one .ll for each class (and one for the main program), but if you compile with —release only one is generated, main.ll

18:15 <travis-ci> [travis-ci] Build details : http://travis-ci.org/manastech/crystal/builds/20356961

18:15 <travis-ci> [travis-ci] Change view : https://github.com/manastech/crystal/compare/203ad2bd301b...cad716fd46fc

18:15 <travis-ci> [travis-ci] manastech/crystal#946 (master - cad716f : Ary Borenszweig): The build passed.