[ANN] TCS 1.9.3p28 experimental build
Jon
Recall that Fenix optimizations (a.k.a use Windows API to make `File.expand_path` much faster) must be explicitly enabled similar to `set RUBYOPT=-rfenix/replace` or `ruby -rfenix/replace ...`
Notable updates from previous build:
* Yura's new Hash optimizations and updates to cached $LOAD_PATH and sorted $LOADED_FEATURES
* 1.9.3 backports as of ruby_1_9_3@34386
* OpenSSL 1.0.0g
* RubyGems 1.8.15
* Built with MinGW GCC 4.6.2 toolchain (untested on WinXP SP3)
Built and tested on:
* Windows 7 32bit/MinGW 4.6.2
* Arch 3.2.2 32bit/GCC 4.6.2
* FreeBSD 9.0 32bit/GCC 4.2.1 (build OK but false test failures; clang 3.0 build failures)
=== Build Summary ===
git co -b tcs-ruby_1_9_3 codeshop/ruby_1_9_3
git rebase winstat/ruby_1_9_3
git rebase winio/ruby_1_9_3
git rebase st_optimize/ruby_1_9_3
git rebase fileload-ok/ruby_1_9_3
git rebase cached-lp/ruby_1_9_3
git rebase sorted-lf/ruby_1_9_3
git rebase fenix/ruby_1_9_3
# fix `load.c:rb_get_expanded_load_path` merge conflict, `git rebase --continue`
# replace missed `rb_file_expand_path` in `load.c` and `ruby.c`
rake ruby19 local=C:\Users\Jon\Documents\RubyDev\ruby-git dkver=mingw-32-4.6.2
# add fenix artifacts (MinGW v4.6.2) to sandboxed `site-ruby/1.9.1/i386-msvcrt/`
RUBYOPT="-rfenix/replace" make test-all TESTS='openssl fiddle psych zlib io json mkmf pathname stringio erb fileutils ruby/test_io.rb ruby/test_io_m17n.rb ruby/test_file.rb'
RUBYOPT="-rfenix/replace" make test
=== Rails 3.2.1 (Win7 32bit) Startup Performance ===
# `rails new rails-empty`
C:\rails-empty>ruby --version
ruby 1.9.3p0 (2011-11-08 revision 33661) [i386-mingw32]
C:\rails-empty>timer ruby script\rails r "p $LOAD_PATH.size, $LOADED_FEATURES.size"
59
738
real 8.439
system 5.928
user 2.386
# `rails new rails-empty -d jdbcsqlite3`
C:\rails-empty>jruby --version
jruby 1.6.5.1 (ruby-1.8.7-p330) (2011-12-27 1bf37c2) (Java HotSpot(TM) Client VM 1.7.0_02) [Windows 7-x86-java]
C:\rails-empty>timer jruby script\rails r "p $LOAD_PATH.size, $LOADED_FEATURES.size"
60
738
real 7.332
system 1.778
user 6.318
# `rails new rails-empty`
C:\rails-empty>ruby --version
tcs-ruby 1.9.3p28 (2012-01-28, TCS patched 2012-01-30) [i386-mingw32]
C:\rails-empty>timer ruby script\rails r "p $LOAD_PATH.size, $LOADED_FEATURES.size"
59
738
real 3.775
system 2.652
user 1.107
C:\rails-empty>timer ruby -rfenix/replace script\rails r "p $LOAD_PATH.size, $LOADED_FEATURES.size"
59
740
real 2.121
system 0.967
user 1.107
=== Large File Read (Win7 32bit) Microbenchmark ===
C:\projects\measurements-git>pik run rci bench core_rd_filelines_lf
jruby 1.6.5.1 (ruby-1.8.7-p330) (2011-12-27 1bf37c2) (Java HotSpot(TM) Client VM 1.7.0_02) [Windows 7-x86-java]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 7.878000 0.000000 7.878000 ( 7.847000)
----------------------------------------------- total: 7.878000sec
user system total real
core_rd_filelines_lf 7.816000 0.000000 7.816000 ( 7.816000)
ruby 1.8.7 (2011-12-28 patchlevel 357) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 1.716000 0.203000 1.919000 ( 1.918804)
----------------------------------------------- total: 1.919000sec
user system total real
core_rd_filelines_lf 1.732000 0.171000 1.903000 ( 1.903203)
ruby 1.9.2p290 (2011-07-09) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 17.472000 0.063000 17.535000 ( 17.534431)
---------------------------------------------- total: 17.535000sec
user system total real
core_rd_filelines_lf 17.519000 0.078000 17.597000 ( 17.628031)
ruby 1.9.3p0 (2011-11-08 revision 33661) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 17.612000 0.062000 17.674000 ( 17.690431)
---------------------------------------------- total: 17.674000sec
user system total real
core_rd_filelines_lf 17.753000 0.016000 17.769000 ( 17.784032)
tcs-ruby 1.9.3p28 (2012-01-28, TCS patched 2012-01-30) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 1.155000 0.125000 1.280000 ( 1.279202)
----------------------------------------------- total: 1.280000sec
user system total real
core_rd_filelines_lf 1.170000 0.094000 1.264000 ( 1.263602)
ruby 2.0.0dev (2012-01-28 trunk 34390) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 1.185000 0.078000 1.263000 ( 1.263603)
----------------------------------------------- total: 1.263000sec
user system total real
core_rd_filelines_lf 1.155000 0.109000 1.264000 ( 1.263602)
Jon
---
Fail fast. Fail often. Fail publicly. Learn. Adapt. Repeat.
http://thecodeshop.github.com | http://jonforums.github.com/
twitter: @jonforums
Funny Falcon
Since my patches affects not only Windows, could we also pack source code? So that rvm and rbenv could include us as option.
Yura.
Jon
>
> Since my patches affects not only Windows, could we also pack source code?
> So that rvm and rbenv could include us as option.
I like it.
Since I rebased all your commits from `st_optimize/ruby_1_9_3`, `cached_lp/ruby_1_9_3`, and `sorted_lf/ruby_1_9_3` onto `tcs-ruby_1_9_3` can rvm and rbenv use the `tcs-ruby_1_9_3` branch as-is or do we need to do something else? (I'm a minimal rvm user and haven't yet looked at rbenv.)
My only concern is that since the `tcs-ruby_1_9_3` branch will likely be forced everytime I do a release, the rebasing could cause downstream problems.
What's the easiest and most maintainable way for us to work with rvm and rbenv for releases?
Jon
Jon
I use rvm on my Arch VM and will install rbenv on my FreeBSD VM a bit later...reading their doco...
Jon
> > > So that rvm and rbenv could include us as option.
> >
> > I like it.
> >
> > Since I rebased all your commits from `st_optimize/ruby_1_9_3`, `cached_lp/ruby_1_9_3`, and `sorted_lf/ruby_1_9_3` onto `tcs-ruby_1_9_3` can rvm and rbenv use the `tcs-ruby_1_9_3` branch as-is or do we need to do something else? (I'm a minimal rvm user and haven't yet looked at rbenv.)
> >
> > My only concern is that since the `tcs-ruby_1_9_3` branch will likely be forced everytime I do a release, the rebasing could cause downstream problems.
This seems to work for me on my FreeBSD 9.0 VM with rbenv. I will update the FAQ with rvm info after testing on my Arch VM.
https://github.com/thecodeshop/ruby/wiki/FAQ#wiki-rbenv
devil% uname -opr
FreeBSD 9.0-RELEASE i386
devil% rbenv global
system
devil% rbenv versions
tcs-ruby_1.9.3-p28
devil% rbenv which ruby
/usr/local/bin/ruby
devil% rbenv shell tcs-ruby_1.9.3-p28
devil% rbenv which ruby
/home/jon/.rbenv/versions/tcs-ruby_1.9.3-p28/bin/ruby
devil% ruby -ve 'puts "Hello from TCS Ruby!"'
tcs-ruby 1.9.3p28 (2012-01-28, TCS patched 2012-01-30) [i386-freebsd9.0]
Hello from TCS Ruby!
But I think the replacements for Fenix on Windows could degrade performance on *nix so I'll #ifndef the `tcs-ruby_1_9_3` and `fenix/ruby_1_9_3` branches unless I hear differently from anyone.
https://github.com/thecodeshop/ruby/commit/22277c0b5c524da4583895a3c9053914b889fd4c
Feel free to update the FAQ entry if needed.
Jon
Jon
> So that rvm and rbenv could include us as option.
>
Do these two solve the issue or do we need to do more?
https://github.com/thecodeshop/ruby/commit/8571639f894479378934996fb310c9205451d6f2
https://github.com/thecodeshop/ruby/wiki/FAQ#wiki-rbenv
It may take me some time to swing back and update the FAQ for rvm so
feel free to add a new entry for rvm if you've got a good solution.
Thanks,
Jon
Funny Falcon
I pushed couple of commits to tcs-ruby_1_9_3.
They:
- make forward order of packing keys and improves performance of GC (st_foreach_nocheck),
- makes pool-alloc portable to OpenBSD (copied from ruby-trunk),
- and backport of macros check in aligned_malloc/aligned_free
(also copied it from ruby-trunk, if it makes trouble, then we should make bug report)
They should be safe (I hope). I've pushed them cause they affects usability and will improve user attraction.
All of them are in st_optimize too.
With regards,
Yura.
Jon
Awesome...looking forward to testing in my Win7 (MinGW and Windows SDK), Arch, and FreeBSD environments.
When tweaking `tcs-ruby_1_9_3`, please update the TCS datestamps https://github.com/thecodeshop/ruby/commit/b29f5d410af004ccef475f91c0ef70ffd25c7707 just in case someone runs into issues and posts at the ML or repo.
What OS/toolchains did you test (`make test && make test-all` ??) the new mods on?
Thanks,
Jon
Funny Falcon
Not so. Only one or two percents on long running application.
> When tweaking `tcs-ruby_1_9_3`, please update the TCS datestamps
> What OS/toolchains did you test (`make test && make test-all` ??) the new mods on?
Best regards,
Yura.
Jon
> mods on?
>
> Only Ubuntu. But most OS dependent changes are copied from ruby-trunk - I
> hope they know what they do :)
:)
No build/test failures with...
== FreeBSD 9.0 VM/GCC 4.2.1 ==
../configure --enable-shared --disable-install-doc
make
make test
make test-all TESTS='openssl fiddle psych zlib io json mkmf pathname stringio erb fileutils ruby/test_io.rb ruby/test_io_m17n.rb ruby/test_file.rb'
== Win7/Windows SDK 7.1 (no deps) ==
..\win32\configure.bat --disable-install-doc --disable-win95
nmake
nmake test
nmake test-all TESTS="mkmf pathname stringio fileutils ruby/test_io.rb ruby/test_io_m17n.rb ruby/test_file.rb"