We started our RubyGems Hackfest Saturday, with a turnout of 6 people. Five were regulars from Seattle.rb, and we even got someone up from Portland! We started Saturday by brainstorming things we'd like, writing up story cards, estimating and risking them. Then we paired up and got testing!
By the end of day 1, we'd finished cards 1-3 and gotten pretty good testing infrastructure in place for stuff we needed. I spent some extra time making it so that all unit tests generate their gems on the fly, weren't dependent on the location from which they were run, and didn't use or create any files in the test directory. We went over our estimates by 3.5 hours. Much of that was my fault, it was too hot for me to think.
On day 2, we were down to 4 people, but had much better testing capabilities. Aaron and Evan finished up the package listing command extension and gave .gemrc the ability to have .cvsrc-style command option specifications. Eric and I worked on the hardest bug we had, which was to get rid of the use of Date. We succeeded on both counts! We went over by 1 hour on a task, and under by 1 hour on the other. YAY! I've been wanting to conquer this issue for quite some time!
Below you can see a the fruits of our labors:
Old and Bad:
% time ruby -rprofile -rrubygems -e 'require_gem "RubyInline"' 2>&1 |head % cumulative self self total time seconds seconds calls ms/call ms/call name 19.79 0.95 0.95 132 7.20 10.38 Integer#gcd 8.96 1.38 0.43 572 0.75 1.47 Gem::Specification#copy_of 6.46 1.69 0.31 68 4.56 100.88 Array#each 5.21 1.94 0.25 3315 0.08 0.08 Fixnum#== 3.96 2.13 0.19 220 0.86 1.23 Rational#initialize 3.54 2.30 0.17 1628 0.10 0.16 Kernel.=== 3.12 2.45 0.15 132 1.14 13.48 Rational#reduce 2.71 2.58 0.13 463 0.28 0.37 Kernel.dup real 0m5.530s user 0m4.856s sys 0m0.323s
New and Good:
% time ruby -Ilib -rprofile -rrubygems -e 'require_gem "RubyInline"' 2>&1 |head % cumulative self self total time seconds seconds calls ms/call ms/call name 24.63 0.50 0.50 572 0.87 1.50 Gem::Specification#copy_of 8.87 0.68 0.18 37 4.86 82.43 Array#each 7.39 0.83 0.15 44 3.41 4.32 Date#_parse 6.90 0.97 0.14 1628 0.09 0.14 Kernel.=== 4.43 1.06 0.09 1386 0.06 0.06 Kernel.== 4.43 1.15 0.09 1211 0.07 0.07 Module#=== 3.94 1.23 0.08 5 16.00 22.00 Hash#each 3.94 1.31 0.08 20 4.00 29.00 Kernel.require real 0m2.415s user 0m2.070s sys 0m0.183s
sorry: did'nt you get rid of Date? then why there is Date#_parse in the latter profile trace?
We got rid of Date and switched it to Time, yes. What time does under the covers we don't care as long as it is faster and passes all its tests.
It turns out that Time.parse calls ParseDate.parse which calls Date._parse (which isn't in date.rb, but is in date/format.rb). It looks like _parse does all the work, but without creating a Date instance and so avoids the heavy cost of Rational#gcd.
I still think you should simply have reimplemented Rational#gcd. :)