Keeping Ruby Running on Cygwin

https://rubykaigi.org/2026/presentations/fd0.html

Transcript

1. Keeping Ruby Running on Cygwin Daisuke Fujimura 2026-04-23 1
2. ˒ Daisuke Fujimura ˒ Maintainer of the ruby binary package on cygwin (Not a Ruby commi tt er) ˒ Interested in software packaging, build systems, and platform engineering About Me 2
3. ˒ Background ˒ Packaging Journey ˒ Upstream Contributions ˒ Looking Back / Looking Forward Outline 3
4. What is Cygwin? 4
5. ˒ A large collection of GNU and Open Source tools which provide functionality similar to a Linux distribution on Windows ˒ A DLL (cygwin1.dll) which provides substantial POSIX API functionality Cygwin is... 5 h tt ps://cygwin.com/
6. ˒ WSL has taken over many use cases ˒ But Cygwin still has demand ˒ Native Windows interop ˒ Legacy work fl ows Cygwin in the WSL Era 6
7. ˒ O ff icial pre-compiled binary packages ˒ Distributed like rpm or deb ˒ Via the Cygwin installer; ruby.exe ready ˒ Maintained by me Ruby on Cygwin 7
8. Cygwin Installer 8
9. ruby.exe (x86_64-cygwin) ˒ x86_64-cygwin: identifying this platform 9
10. ˒ Works like Linux ˒ Unix-style paths (e.g. /home, not C:\Users) ˒ Many POSIX APIs work (e.g. Process.fork) ˒ ...and more Di ff erent from Windows Ruby 10
11. ˒ Each package has a volunteer maintainer ˒ Step away → package becomes orphaned ˒ Orphaned = no active maintainer ˒ Ruby went orphaned after 2.6 ˒ Ruby commi tt ers had no Cygwin maintainer either [1] The Orphaned Package 11 [1] h tt ps://github.com/ruby/ruby/blob/master/doc/maintainers.md
12. No one to ask for help 12
13. Wait — am I possibly the most active Ruby-on-Cygwin person in the world? 13
14. ˒ My interests all converge here ˒ Long-time Cygwin user ˒ Into language runtimes and libraries ˒ Love Linux packaging (e.g. rpm, deb,...) Motivation: Why I started maintaining Ruby on Cygwin (1) 14
15. ˒ The outdated Ruby was a real problem ˒ Security risk ˒ Missing new features Motivation: Why I started maintaining Ruby on Cygwin (2) 15
16. ˒ KaigiE ff ect ˒ First a tt ended RubyKaigi 2018 ˒ Blew my mind; never missed one since Motivation: Why I started maintaining Ruby on Cygwin (3) 16
17. Release History 17 Version Original Release Cygwin Package Release EOL 2.6.4 2019-08-28 2019-09-16 2022-04-12 3.2.2 2023-03-30 2023-04-28 2026-03-31 3.4.7 2025-10-07 2025-10-17 TBD h tt ps://cygwin.com/packages/summary/ruby.html
18. Packaging Journey 18
19. ˒ No handover docs: the repository was the only guide ˒ Let's look at what's in there Climbing from 2.6 to 3.2 19
20. ˒ Cygport: Cygwin's build system ˒ Rules wri tt en in bash syntax in a .cygport fi le Repository Tree (1) 20 h tt ps://cygwin.com/cgit/cygwin-packages/ruby/
21. ˒ Some parts of Ruby assume only major platforms; patches make them work correctly on Cygwin Repository Tree (2) 21 h tt ps://cygwin.com/cgit/cygwin-packages/ruby/
22. ˒ Some patches are sourced from Fedora for reliability and license compatibility Repository Tree (3) 22 h tt ps://cygwin.com/cgit/cygwin-packages/ruby/
23. ˒ Opt out of arch-speci fi c bindir (e.g. /usr/libexec/x86_64-linux/ bin) Patch Example 23 h tt ps://cygwin.com/cgit/cygwin-packages/ruby/tree/3.4.2-rbinstall-archbindir.patch
24. ˒ $ cygport *.cygport <command> Building with cygport 24 fetch prep compile install package
25. ruby.cygport — Metadata 25 ˒ Declared as bash variables: name, version, source URLs, patch locations, etc.
26. ˒ Build steps (compile, install, etc.) are de fi ned by overriding bash functions ˒ Ruby uses GNU Autotools; con fi gure options are speci fi ed in src_compile() ruby.cygport — Build Functions 26
27. Package Layout (1) 📃 ruby.cygport (3.2.2) 📦 ruby (3.2.2) 📦 ruby-devel (3.2.2) 📦 rubygems (3.4.10) 📦 ruby-rdoc (6.5.0) $ cygport ruby.cygport package 27 X X (3.2.2) (3.2.2)
28. Package Layout (2) 📃 rubygems.cygport (3.4.10) 📦 rubygems (3.4.10) 📦 ruby-rdoc (6.5.0) 📃 ruby-rdoc.cygport (6.5.0) $ cygport rubygems.cygport package $ cygport ruby-rdoc.cygport package 28 📃 ruby.cygport (3.2.2) 📦 ruby (3.2.2) $ cygport ruby.cygport package
29. Package Layout (3) 📦 ruby (3.2.2) /usr/bin/ruby.exe /usr/bin/gem /usr/bin/rdoc : 📦 rubygems (3.4.10) /usr/bin/gem : 📦 ruby-rdoc (6.5.0) /usr/bin/rdoc : Con fl ict Con fl ict 29
30. ruby.cygport — install 📃 ruby.cygport (3.2.2) 📦 ruby (3.2.2) 30 Drop gem, rdoc, ...
31. ABI Info Missing (1) 📦 ruby-bcrypt (3.1.20) 📦 ruby (2.6.4) 📦 ruby (3.2.2) Wants to install Which ABI? ˒ New Ruby → ABI changes → gems must be rebuilt ˒ The installer had no way to know which Ruby ABI each gem was built for ˒ → Unable to resolve dependencies 31
32. ˒ Hint fi les (generated per package build) → aggregated into setup.ini ABI Info Missing (2) 📃 ruby-bcrypt.hint 📃 setup.ini 📃 ruby-devel.hint : 📃 ruby.hint $ mksetupini 32
33. setup.ini 33 ˒ depends2 includes ruby_xy, the Ruby x.y ABI marker
34. ˒ O ff icial CI/CD takes it from here ˒ GitHub Actions: build, upload, update setup.ini, announce ˒ Users get the latest Ruby. happy ending! Ge tt ing It Out the Door 34
35. ˒ 3.3 never shipped: a SEGV with unknown cause blocked the release ˒ gdb investigation, but hard to use e ff ectively ˒ With AI's step-by-step guidance, narrowed down the cause Climbing Again: 3.2 to 3.4 — Debugging with AI 35
36. ˒ Root cause: SEGV inside pthread, triggered by libssl 3 ˒ 3.2 was fi ne: linked against libssl 1.1 ˒ Shipped with 3.4.7, after libssl resolved the issue ˒Lesson: sometimes the bug is outside Ruby Climbing Again: 3.2 to 3.4 — Root Cause & Resolution 36
37. Upstream Contributions 37
38. ˒ My own (uno ff icial) CI pipeline ˒ Tracks Ruby HEAD ˒ Fix issues upstream, reduce local patches ˒ Led to several upstream PRs ˒ #9269 (#9357), #11952, #14870 Running CI Myself — and What I Found 38
39. #9269: Use native_thread_init_stack 39
40. ˒ test_ractor.rb, test_thread.rb → SEGV on Cygwin ˒ Grepping #if de fi ned(__CYGWIN__) ˒ native_thread_init_stack() was skipped ˒ Tried removing the exclusion → SEGV was gone ˒ But why? 🤔 ˒ Kept digging to fi nd the reason #9269: Use native_thread_init_stack — Problem 40
41. ˒ Found commit in 2010 [1] ˒ Incomplete signal stack → exclusion was justi fi ed ˒ Cygwin fi xed signal stack in 2015 [2] ˒ Code outlived the reason #9269: Use native_thread_init_stack — Investigation 41 [1] h tt ps://github.com/ruby/ruby/commit/6957c6d9cf0b6a1e211c4b0175bc5474296f2031 [2] h tt ps://github.com/cygwin/cygwin/commit/c8432a01c8401c121940c806a9d868c4adc4cefd
42. ˒ Removing the macro and branch #9269: Use native_thread_init_stack — Fix 42
43. ˒ make test on Cygwin passing again after ~10 years ˒ That's win! #9357: Use ubf list 43
44. #11952: Avoid dangling pointer 44
45. ˒ A dangling pointer is a pointer to memory that is no longer valid ˒ GCC's -Wdangling-pointer caught one in dln.c ˒ But only on Cygwin #11952: Avoid dangling pointer — Problem: Warning 45
46. ˒ dladdr(): looks up which library a given address belongs to ˒ Fills a Dl_info struct with the result ˒ dli_fname member holds the library fi le path ˒ libname holds dli_fname... #11952: Avoid dangling pointer — Problem: The Code 46
47. #11952: Avoid dangling pointer — Investigation: Type Di ff erence 47 Linux [1] Cygwin [2] [1] h tt ps://sourceware.org/git/?p=glibc.git;a=blob;f=dlfcn/dlfcn.h [2] h tt ps://github.com/cygwin/cygwin/blob/main/winsup/cygwin/include/dlfcn.h
48. ˒ Why the type di ff erence? ˒ dladdr() standardized in POSIX Issue 8 (2024) ˒ Cygwin: 2017 [1] ˒ Each platform de fi ned types independently #11952: Avoid dangling pointer — Investigation: Background 48 [1] h tt ps://github.com/cygwin/cygwin/commit/c8432a01c8401c121940c806a9d868c4adc4cefd
49. ˒ libname is only used in error messages; skipping is su ff icient #11952: Avoid dangling pointer — Fix 49
50. #14870: Fix extension fi le permissions in namespace feature 50
51. ˒ Tried namespace (box) locally ˒ Native extension .so → LoadError #14870: Fix extension fi le permissions in namespace feature — Problem 51
52. ˒ fwrite copy → dlopen #14870: Fix extension fi le permissions in namespace feature — Investigation: Copy 52
53. ˒ Cygwin: copied .so permissions come from Windows ACL ˒ The executable bit was not set on the copy #14870: Fix extension fi le permissions in namespace feature — Investigation: Permission 53 -rwxr-xr-x openssl.so -rw-r--r-- openssl.so Copy by fwrite
54. ˒ dlopen internally calls Windows LoadLibraryW ˒ No Executable in ACL → ERROR_ACCESS_DENIED ˒ Not o ff icially documented, but widely known behavior #14870: Fix extension fi le permissions in namespace feature — Investigation: Implementation 54
55. ˒ Preserve permissions after fwrite copy #14870: Fix extension fi le permissions in namespace feature — Fix 55
56. #12051: Try to build with Cygwin 56
57. ˒ My CI caught a compile failure on Cygwin ˒ -municode (MinGW-w64-only) accidentally applied to Cygwin ˒ Fixed by #12015 Cygwin Joins the O ff icial CI Matrix (1) 57
58. ˒ To prevent this class of breakage from landing again ˒ Now every commit is veri fi ed to build on Cygwin Cygwin Joins the O ff icial CI Matrix (2) 58
59. Looking Back / Looking Forward 59
60. ˒ Upgraded Ruby on Cygwin from 2.6 to 3.4 ˒ Deepened understanding of the packaging system ˒ Encountered a new debugging experience ˒ A dependency-blocked release tackled with AI Looking Back — Packaging Journey 60
61. ˒ Niche bugs that only surface on Cygwin ˒ Hard to see without downstream involvement ˒ Cygwin is now in the o ff icial CI matrix Looking Back — Upstream Contributions 61
62. ˒ All of this contributes to Ruby's "Highly Portable" goal ˒ Covering underserved platforms helps grow the Ruby community Looking Forward 62
63. Ruby is still running on Cygwin 63
64. O ff icial Cygwin package for Ruby 4.0? 64
65. 🙇 Sorry! 65