All bugfixes are incompatibilities
0 likes6,076 views
This document summarizes lessons learned from failures in backporting bug fixes to Ruby stable branches: - Don't backport performance improvements or fixes for imaginary use cases as they can introduce regressions. - Be careful backporting fixes related to parsing, constants/method search, and refinements as they are complex and prone to causing new bugs. - Some long-standing bugs may not need fixing if no real applications are affected. It's better to avoid regressions. - Consider an application's needs before backporting - don't backport fixes if no one requested or needs them. Be practical.
![% ruby -ve 'puts File.read("|echo foo")'
ruby 2.4.2p198 (2017-09-14 revision 59899) [x86_64-darwin15]
foo
% ruby -ve 'puts File.read("|echo foo")'
ruby 2.5.4p155 (2019-03-13 revision 67245) [x86_64-darwin15]
-e:1: warning: IO.read called on File to invoke external command
foo
% ruby -ve 'puts File.read("|echo foo")'
ruby 2.6.2p47 (2019-03-13 revision 67232) [x86_64-darwin15]
Traceback (most recent call last):
1: from -e:1:in `<main>'
-e:1:in `read': No such file or directory @ rb_sysopen - |echo foo (Errno::ENOENT)
IO.read and File.read](https://image.slidesharecdn.com/allbugfixesareincompatibilities-190421000620/85/All-bugfixes-are-incompatibilities-22-320.jpg)
All bugfixes are incompatibilities
- 1. 2019-04-19 All bug-fixes are incompatibilities nagachika @ RubyKaigi 2019
- 2. Important Notification about RubyGems.org The attackers try to hijack your account with compromised password list. Example: bootstrap-sass was compromised by this attack. It was injected the vulnerable RCE code. We should prepare the multi-factor authentication on RubyGems.org for protecting from them.
- 3. ✤ @nagachika ✤ CRuby committer ✤ CRuby stable branch maintainer (2.5 → 2.6) ✤ ruby trunk changes ✤ Fukuoka.rb/TFUG Fukuoka WHO AM I
- 4. ✤ @nagachika ✤ CRuby committer ✤ CRuby stable branch maintainer (2.5 → 2.6) ✤ ruby trunk changes ✤ Fukuoka.rb/TFUG Fukuoka WHO AM I
- 5. “What’s the Ruby’s stable branch maintenance?”
- 6. ✤ We backport fixes into stable branches from trunk Ruby’s stable branch model EoL ruby_2_2 ruby_2_5 ruby_2_6 backport trunk bugfix
- 7. ✤ We backport fixes into stable branches from trunk Ruby’s stable branch model nurse unak Release Manager Stable Branch Maintainer EoL ruby_2_2 ruby_2_5 ruby_2_6 backport trunk bugfix nagachika
- 8. ✤ Backport bug-fixes from trunk into stable branches ✤ Release stable packages. ✤ Handle security issues in concert with stakeholders. Stable Branch Maintainer
- 9. ✤ Backport bug-fixes from trunk into stable branches ✤ Release stable packages ✤ Handle security issues in concert with stakeholders Stable Branch Maintainer
- 10. ✤ Watch commits on trunk. ✤ Backport bug-fixes. Stable Branch Maintainer’s daily work
- 11. ✤ Watch commits on trunk → Watch backport requests on Redmine. ✤ Backport bug-fixes. Stable Branch Maintainer’s daily work
- 12. ✤ Redmine: bugs.ruby-lang.org Ruby’s stable maintenance policy
- 13. ✤ File tickets with rubyʼs version information Ruby’s stable maintenance policy
- 14. ✤ You can maintain only one stable branch at once. Ruby’s stable branch model nurse nagachika unak Release Manager Stable Branch Maintainer EoL ruby_2_2 ruby_2_5 ruby_2_6 backport trunk bugfix
- 15. “How can I become a Ruby stable maintainer?”
- 16. Requirements for Stable Branch Maintainers
- 17. ✤ Wide understandings for the Ruby implementation ✤ Sustainability to some extent ✤ Understanding maintenance policy Requirements for Stable Branch Maintainers
- 18. ✤ Backport bug-fixes ✤ Donʼt backport new features/spec changes Ruby’s maintenance policy stable trunk bug-fix new feature
- 19. IO.read could execute a command IO.read(“|cmd…”) executes the command (same as Kernel#open)
- 22. % ruby -ve 'puts File.read("|echo foo")' ruby 2.4.2p198 (2017-09-14 revision 59899) [x86_64-darwin15] foo % ruby -ve 'puts File.read("|echo foo")' ruby 2.5.4p155 (2019-03-13 revision 67245) [x86_64-darwin15] -e:1: warning: IO.read called on File to invoke external command foo % ruby -ve 'puts File.read("|echo foo")' ruby 2.6.2p47 (2019-03-13 revision 67232) [x86_64-darwin15] Traceback (most recent call last): 1: from -e:1:in `<main>' -e:1:in `read': No such file or directory @ rb_sysopen - |echo foo (Errno::ENOENT) IO.read and File.read
- 23. ✤ 2.0.0: 469 ✤ 2.1: 250 ✤ 2.2: 205 ✤ 2.3: 217 ✤ 2.4: 190 ✤ 2.5: 101 Backport commits by nagachika
- 24. ✤ 2.0.0: 469 ✤ 2.1: 250 ✤ 2.2: 205 ✤ 2.3: 217 ✤ 2.4: 190 ✤ 2.5: 101 Backport commits by nagachika include Documentation, performance improvement, memory usage improvement…
- 28. ✤ The change broke CI on mswin Backport performance improvements
- 29. ✤ The lesson learned ✤ Donʼt backport performance improvements Backport performance improvements
- 30. Backport introduced memory violation
- 31. Backport introduced memory violation
- 32. Backport introduced memory violation
- 33. ✤ The lesson learned ✤ Bug-fixes for Method/Constants search in edge cases sometimes introduce another memory violation. ✤ Fixing a minor bug can introduce severe bugs Backport introduced memory violation
- 34. Refinements
- 35. Refinements
- 36. ✤ The lesson learned ✤ “Refinements” is not a matured feature and continue expanding its scope of adaptation. ✤ Deal it carefully. Refinements
- 37. Fixing SyntaxError caused another SyntaxError
- 38. Fixing SyntaxError caused another SyntaxError
- 39. Fixing SyntaxError caused another SyntaxError
- 40. cond ? expr1 :expr2 Fixing SyntaxError caused another SyntaxError { key:(expr rescue nil) } { key:<<EOF } EOF
- 41. ✤ The lesson learned ✤ parse.y is “魔境” (mass of legendary complexity) ✤ SyntaxError is the one of the most severe issues ✤ Monkey patching cannot get rid of SyntaxError Fixing SyntaxError caused another SyntaxError
- 42. Fixing SyntaxError caused another SyntaxError(2)
- 43. Fixing SyntaxError caused another SyntaxError(2)
- 44. Fixing SyntaxError caused another SyntaxError(2)
- 45. def foo key1: … end Fixing SyntaxError caused another SyntaxError(2) { key:if foo then 1 else 0 }
- 46. ✤ The lesson learned(?) ✤ Fixing the SyntaxError introduced another SyntaxError in the more rare case ✤ But the prior SyntaxError was in the new syntax (Mandatory keyword arguments) ✤ There was maybe no real application affected by the prior SyntaxError. Fixing SyntaxError caused another SyntaxError(2)
- 47. Fixing SyntaxError caused another SyntaxError(3)
- 48. Fixing SyntaxError caused another SyntaxError(3)
- 49. Fixing SyntaxError caused another SyntaxError(3)
- 50. ✤ The lesson learned ✤ parse.y is “魔境” ✤ Long-standing bugs have less priority → “Long-standing” means thereʼs no real application affected ✤ “Known bug” can be better than “Regression” Fixing SyntaxError caused another SyntaxError(3)
- 51. Load FIFO
- 52. ✤ Release GVL during open fd for `load` Load FIFO
- 53. ✤ Performance degradation of YAML.parse on multi-threads environments Load FIFO
- 54. ✤ The lesson learned ✤ Donʼt backport bug-fixes for imaginary use case. No one wanted to load from FIFO! Load FIFO
- 55. ✤ Donʼt backport performance improvements ✤ Donʼt backport fix for the long-standing bugs ✤ Donʼt jump at the fix in parse.y, constants/ methods search and refinements ✤ Donʼt backport fixes on one wanted to backport The Lessons Learned from failures
- 56. ✤ Be-Practical! The Lessons Learned from failures