August 2026 - ruby-core - ml.ruby-lang.org

[ruby-core:126246] [Ruby Bug#22228] Do a major GC in YJIT test helper `assert_compiles` before running the test code
by jprokop (Jarek Prokop) 04 Aug '26

04 Aug '26

Issue #22228 has been reported by jprokop (Jarek Prokop). ---------------------------------------- Bug #22228: Do a major GC in YJIT test helper `assert_compiles` before running the test code https://bugs.ruby-lang.org/issues/22228 * Author: jprokop (Jarek Prokop) * Status: Open * ruby -v: ruby 4.1.0dev (2026-08-04T15:30:26Z master 036159508e) +PRISM [x86_64-linux] * Backport: 3.3: UNKNOWN, 3.4: UNKNOWN, 4.0: UNKNOWN ---------------------------------------- I was debugging why a YJIT test added from backport to 3.3.11 with GC.stress fails downstream. This revealed an assumption in the assert_compiles, that the environment is "clean", or as default as possible. `assert_compiles` already sanitizes environment with `--disable-gems` and `reset_stats!` call. I'd like to propose the `assert_compiles` method to trigger a GC.start before executing the test proc, and before the stats reset. While the issue described below does not happen on default Rubies 3.4 and 4.0, it could enhance the YJIT test environment. I was investigating downstream test failure on Ruby 3.3.11 (and later 3.3.12) [0] that appeared after backport of: https://bugs.ruby-lang.org/issues/21266 Where the tests in YJIT expects no exits by default: Provided as a default here: https://github.com/ruby/ruby/blob/3e646aee/test/ruby/test_yjit.rb#L2015 Expected as part of condition here: https://github.com/ruby/ruby/blob/3e646aee/test/ruby/test_yjit.rb#L2080 Downstream we insert hook for optional ABRT error reporting for Ruby: https://src.fedoraproject.org/rpms/ruby/blob/11eb00c7bf5c87033a73301ca8dad5… Relevant patchfile: https://src.fedoraproject.org/rpms/ruby/blob/11eb00c7bf5c87033a73301ca8dad5… We create empty `abrt.rb` to test with later on: https://src.fedoraproject.org/rpms/ruby/blob/11eb00c7bf5c87033a73301ca8dad5… To reproduce the failure in 3.3, both the patch and the file are required to exist for the failure at [0] to appear. Later I found that parse.y opens a file to load the abrt.rb and registers it with Ruby GC, which causes zombie and interrupt on `GC.stress=true`, since prism parser became default, these failures no longer appear as the parser does not copy parse.y's approach of registering the opened file with Ruby GC. The note on #19921 https://bugs.ruby-lang.org/issues/19921#note-20 regarding this was correct and it's also the earliest known appearance of the issue. --- [0] Test failure result: TestYJIT#test_proc_block_with_kwrest = 0.08 s 1) Failure: TestYJIT#test_proc_block_with_kwrest [/builddir/build/BUILD/ruby-3.3.11/test/ruby/test_yjit.rb:1531]: Expected no exits, but got: {:send=>1} Reasons: method call fallback reasons: iseq_has_kwrest: 1 (100.0%) invokeblock fallback reasons: proc: 2 (100.0%) invokesuper fallback reasons: (all relevant counters are zero) method call exit reasons: interrupted: 1 (100.0%) invokeblock exit reasons: (all relevant counters are zero) invokesuper exit reasons: (all relevant counters are zero) getblockparamproxy exit reasons: (all relevant counters are zero) getinstancevariable exit reasons: (all relevant counters are zero) setinstancevariable exit reasons: (all relevant counters are zero) leave exit reasons: interp_return: 6 (100.0%) left shift (ltlt) exit reasons: (all relevant counters are zero) invalidation reasons: constant_ic_fill: 2 (100.0%) -- https://bugs.ruby-lang.org/

1 1

[ruby-core:126237] [Ruby Bug#18947] Unexpected Errno::ENAMETOOLONG on Windows
by hsbt (Hiroshi SHIBATA) 04 Aug '26

04 Aug '26

Issue #18947 has been updated by hsbt (Hiroshi SHIBATA). Two questions before this goes further. * The bit is set in `rb_w32_sysinit`, so only hosts that call `ruby_sysinit` get it, which effectively means `ruby.exe`. Embedders that do their own initialization keep the 260 character limit. Should I move it somewhere every embedder reaches? * `IsLongPathAwareProcess` is undocumented. golang has set it since 1.23, but Microsoft's Go maintainers proposed removing it. The alternative is `\\?\` prefixing at about 15 call sites, which still cannot fix `Dir.chdir` or `require`. ---------------------------------------- Bug #18947: Unexpected Errno::ENAMETOOLONG on Windows https://bugs.ruby-lang.org/issues/18947#change-118323 * Author: inversion (Yura Babak) * Status: Open * Assignee: windows * ruby -v: ruby 3.1.2p20 (2022-04-12 revision 4491bb740a) [x64-mingw-ucrt] * Backport: 2.7: UNKNOWN, 3.0: UNKNOWN, 3.1: UNKNOWN ---------------------------------------- On Windows 10, I am working on a script to copy a complex folder structure. Pathname and FileUtils work fine for me until there is a folder with a **very long path** (>260 chars). Normally you cannot access such a folder with Ruby. The next operations will raise `Errno::ENOENT` ``` ruby Pathname.new(300_chars_path).children FileUtils.mkpath(300_chars_path) ``` But there is a way in Windows to remove the MAX_PATH limitation. You can find a small .reg file in this article: https://docs.microsoft.com/en-us/windows/win32/fileio/maximum-file-path-lim… After changing this system option, things start to work strangely in Ruby. This will now raise `Errno::ENAMETOOLONG`: ``` ruby Pathname.new(300_chars_path).children ``` But at the same time, you can create a folder with such a long path and write-read a file in it ``` ruby FileUtils.mkpath(300_chars_path) file = Pathname.new(300_chars_path+'/file.txt') file.write 'oooooooooo' puts Pathname.new(300_chars_path+'/file.txt').read ``` So you can work with individual items but attempts to list such folders' content fail (`.children`, `.glob`, `.copy`, etc). In my case, deep `.glob` is broken for all the parent folders of that deep long-path folder (( The only way I found for listing is ``` ruby require 'win32ole' fso = WIN32OLE.new 'Scripting.FileSystemObject' for file in fso.GetFolder(300_chars_path).files file.name file.path.length end ``` But using this workaround breaks all my code workflow built on top of Pathname and FileUtils ((. So for me, it looks like some operations with long-path folders are not working just because in Ruby there is a check for the path length and not a real operation problem. And in some places (see .mkpath) there is no such check and all works fine. Also notice that other applications on Windows have no problems with long-path folders (like Total Commander). Please consider reviewing if we really need to raise `Errno::ENAMETOOLONG` if the `LongPathsEnabled` option is enabled in the Windows registry. -- https://bugs.ruby-lang.org/

1 0

[ruby-core:126226] [Ruby Bug#22224] YJIT: rb_yjit_invalidate_ep_is_bp takes the VM lock stopping Ractors on every Proc materialization; multi-Ractor throughput collapses (up to ~150x)
by yaroslavmarkin (Yaroslav Markin) 03 Aug '26

03 Aug '26

Issue #22224 has been reported by yaroslavmarkin (Yaroslav Markin). ---------------------------------------- Bug #22224: YJIT: rb_yjit_invalidate_ep_is_bp takes the VM lock stopping Ractors on every Proc materialization; multi-Ractor throughput collapses (up to ~150x) https://bugs.ruby-lang.org/issues/22224 * Author: yaroslavmarkin (Yaroslav Markin) * Status: Open * ruby -v: 4.0.6 * Backport: 3.3: UNKNOWN, 3.4: UNKNOWN, 4.0: UNKNOWN ---------------------------------------- Hi! Fair warning: a significant portion of this bugreport and experimentation/benchmarking was done by an agent. Sorry for the slop, but it was so much faster to benchmark and find a root issue. ## Summary With YJIT enabled, `vm_make_env_each` calls `rb_yjit_invalidate_ep_is_bp` on every Proc/lambda environment materialization. That function enters `with_vm_lock` unconditionally (the only early return covers boot, before `INVARIANTS` is initialized). In multi-Ractor mode this lock acquisition is `rb_jit_vm_lock_then_barrier`: the VM lock plus a stop-all-Ractors barrier. The result: any workload that creates Procs loses parallel scalability under Ractors, and beyond ~2 Ractors adding workers makes the whole process slower in absolute terms. In the repro below, 8 Ractors with YJIT run ~150x slower than 1 Ractor with YJIT, and ~195x slower than 8 Ractors without YJIT, on the same code. Note the invalidation itself is not the cost. After the first escape of a given iseq, its `no_ep_escape_iseqs` entry is an empty set forever (`ep_is_bp()` returns false for it, so no new blocks are ever registered), yet every subsequent materialization still pays lock + global barrier to look up the entry and iterate nothing. ## Standalone repro (no gems) ```ruby # Usage: ruby [--yjit] yjit_ractor_repro.rb [proc|calc] [n_ractors] [seconds] MODE = (ARGV[0] || "proc").to_sym N = (ARGV[1] || 1).to_i DUR = (ARGV[2] || 3).to_f def make_proc(i) x = i -> { x + 1 } # captures x: the frame env is materialized on every call end def work_proc(n) s = 0 n.times { |i| s += make_proc(i).call } s end def work_calc(n) s = 0 n.times { |i| s += (i * i) % 7 } # same shape, no Proc escapes s end def bench_loop(mode, dur) count = 0 deadline = Process.clock_gettime(Process::CLOCK_MONOTONIC) + dur while Process.clock_gettime(Process::CLOCK_MONOTONIC) < deadline mode == :proc ? work_proc(1000) : work_calc(1000) count += 1000 end count end bench_loop(MODE, 0.5) # warm up: get the methods YJIT-compiled before measuring total = if N == 1 bench_loop(MODE, DUR) else N.times.map { Ractor.new(MODE, DUR) { |m, d| bench_loop(m, d) } }.sum(&:value) end puts format("mode=%s yjit=%s ractors=%d throughput=%.2fM iters/s", MODE, RubyVM::YJIT.enabled?, N, total / DUR / 1_000_000.0) ``` Results on Apple M1 Pro (10 cores), macOS, `ruby 4.0.6 (2026-07-14 revision 03b6d3f889) +PRISM [arm64-darwin25]`, 3 s per cell: | workload | ractors | interpreter (M iters/s) | --yjit (M iters/s) | |----------|---------|-------------------------|--------------------| | calc (control, no Proc) | 1 | 21.3 | 129.4 | | calc | 4 | 83.0 | 495.3 | | calc | 8 | 162.2 | **969.1** | | proc | 1 | 4.5 | 6.0 | | proc | 4 | 7.7 | **0.14** | | proc | 8 | 7.8 | **0.04** | The control shows this is not a general YJIT-vs-Ractor problem: on the Proc-free workload YJIT scales superbly (969M iters/s at 8 Ractors). Only the environment-materializing workload collapses, and only with YJIT on. ## Where the time goes Native sampling (macOS `sample`) of a loaded multi-Ractor process shows almost all threads parked in `__psynch_cvwait`, with the barrier initiated from: ``` vm_make_env_each -> rb_yjit_invalidate_ep_is_bp (yjit/src/invariants.rs) -> with_vm_lock -> rb_jit_vm_lock_then_barrier -> rb_ractor_sched_barrier_start (victims: rb_ractor_sched_barrier_join / ractor_sched_barrier_join_wait_locked) ``` In a 6 s sample of a 5-Ractor process, barrier-related frames appear ~9,000 times vs ~50 in the single-Ractor run of the same workload. Ruby-side profilers cannot see this: the wait time is attributed as diffuse "self time" across whatever frames are on top, which is presumably why it has gone unreported. Current code (`yjit/src/invariants.rs`, same on master as of 2026-08-03): ```rust pub extern "C" fn rb_yjit_invalidate_ep_is_bp(iseq: IseqPtr) { // Skip tracking EP escapes on boot. We don't need to invalidate anything during boot. if unsafe { INVARIANTS.is_none() } { return; } with_vm_lock(src_loc!(), || { let no_ep_escape_iseqs = &mut Invariants::get_instance().no_ep_escape_iseqs; match no_ep_escape_iseqs.get_mut(&iseq) { Some(blocks) => { for block in mem::take(blocks) { invalidate_block_version(&block); incr_counter!(invalidate_ep_escape); } } None => { no_ep_escape_iseqs.insert(iseq, HashSet::new()); } } }); } ``` ## Real-world impact Rails 8.1 enables YJIT by default, and a Rails request materializes many Proc environments (middleware blocks, route handling, view rendering), so any Rails app served by a multi-Ractor server hits this out of the box. Found while investigating https://github.com/yaroslav/kino/issues/6, where a stock Rails 8.1 app showed worker scaling inverting: more Ractors, less total throughput. Rails 8.1.3 health-check endpoint (`/up`, no database), `ab -c 64 -k`, kino `:ractor` mode, N workers x 1 thread, requests/sec: | workers | YJIT on (Rails default) | YJIT off | |---------|-------------------------|----------| | 1 | 6,416 | 3,964 | | 2 | 5,036 | 6,999 | | 5 | 2,800 | 13,269 | | 8 | 2,003 | 12,596 | Fully disabling GC changed the YJIT-on numbers by only ~7%, ruling out GC barriers as the driver; the native profile above identifies the initiator. -- https://bugs.ruby-lang.org/

2 1

[ruby-core:126201] [Ruby Bug#22220] Performance regression when requiring `aws-sdk-ec2` in Ruby 4.0.6
by emanuelevans (Emanuel Evans) 03 Aug '26

03 Aug '26

Issue #22220 has been reported by emanuelevans (Emanuel Evans). ---------------------------------------- Bug #22220: Performance regression when requiring `aws-sdk-ec2` in Ruby 4.0.6 https://bugs.ruby-lang.org/issues/22220 * Author: emanuelevans (Emanuel Evans) * Status: Open * ruby -v: 4.0.6 * Backport: 3.3: UNKNOWN, 3.4: UNKNOWN, 4.0: UNKNOWN ---------------------------------------- Ruby 4.0.6 introduced a large performance regression when loading `aws-sdk-ec2` (~30s on my machine, compared to <1s for ruby 4.0.5). Script to reproduce: ``` for v in 4.0.5 4.0.6; do docker run --rm ruby:$v-slim bash -c ' gem install --no-document aws-sdk-ec2 -v 1.402.0 >/dev/null 2>&1 ruby -e "t = Process.clock_gettime(Process::CLOCK_MONOTONIC) require \"aws-sdk-ec2\" printf(\"ruby %s: %.2fs\n\", RUBY_VERSION, Process.clock_gettime(Process::CLOCK_MONOTONIC) - t)"' done ``` It appears to be related to `Shape.new` and/or `Class.new` compile times; minimal reproduction: ``` for v in 4.0.5 4.0.6; do docker run --rm ruby:$v-slim ruby -e ' src = "class Shape\n def self.new(**) = nil\nend\nmodule M\n" + (1..16_000).map { |i| %( K_#{i} = Shape.new(name: "k#{i}")) }.join("\n") + "\nend\n" File.write("/tmp/bench.rb", src) t = Process.clock_gettime(Process::CLOCK_MONOTONIC) RubyVM::InstructionSequence.compile_file("/tmp/bench.rb") printf("ruby %s: %.2fs\n", RUBY_VERSION, Process.clock_gettime(Process::CLOCK_MONOTONIC) - t)' done ``` -- https://bugs.ruby-lang.org/

3 3

[ruby-core:115016] [Ruby master Bug#19921] TestYJIT#test_bug_19316 test failure
by vo.x (Vit Ondruch) 03 Aug '26

03 Aug '26

Issue #19921 has been reported by vo.x (Vit Ondruch). ---------------------------------------- Bug #19921: TestYJIT#test_bug_19316 test failure https://bugs.ruby-lang.org/issues/19921 * Author: vo.x (Vit Ondruch) * Status: Open * Priority: Normal * ruby -v: ruby 3.3.0dev (2023-10-12 master 52709a4862) [x86_64-linux] * Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN ---------------------------------------- Testing with 52709a4862 on Fedora Rawhide, I observe following failure: ~~~ 94) Failure: TestYJIT#test_bug_19316 [/builddir/build/BUILD/ruby-3.3.0-52709a4862/test/ruby/test_yjit.rb:1126]: Expected no exits, but got {:opt_and=>1} ~~~ -- https://bugs.ruby-lang.org/

6 21

[ruby-core:126223] [Ruby Bug#15598] Deadlock on mutual reference of autoloaded constants
by wanabe (_ wanabe) 02 Aug '26

02 Aug '26

Issue #15598 has been updated by wanabe (_ wanabe). This issue appears to have been corrected at https://github.com/ruby/ruby/commit/6fbc32b5d0da31535cccc0eca1853273313a0b52. ``` $ (git checkout 6fbc32b5d0da31535cccc0eca1853273313a0b52~; make -j miniruby) >/dev/null && ./miniruby -v base_thread_const.rb ; echo $? HEAD is now at 30e5e7c005f Revert "Fix jump buffer leak in setjmp handler in WASI builds" ruby 3.5.0dev (2025-04-01T16:11:01Z v3_5_0_preview1~146 30e5e7c005) +PRISM [x86_64-linux] base_thread_const.rb:5:in 'Thread#join': No live threads left. Deadlock? (fatal) 3 threads, 3 sleeps current:0x000064627f26fd10 main thread:0x000064627f200ad0 * #<Thread:0x00007c9089d2a668 sleep_forever> rb_thread_t:0x000064627f200ad0 native:0x00007c908a2a07c0 int:0 base_thread_const.rb:5:in 'Thread#join' base_thread_const.rb:5:in '<main>' * #<Thread:0x00007c9089d12ae0 base_thread_const.rb:3 sleep_forever> rb_thread_t:0x000064627f269ea0 native:0x00007c906ebbe6c0 int:0 mutex:0x000064627f26fef0 cond:1 depended by: tb_thread_id:0x000064627f200ad0 /home/wanabe/work/prog/ruby/ruby/tmp/master/a.rb:3:in '<class:A>' /home/wanabe/work/prog/ruby/ruby/tmp/master/a.rb:1:in '<top (required)>' base_thread_const.rb:3:in 'Kernel#require' base_thread_const.rb:3:in 'block in <main>' * #<Thread:0x00007c9089d129c8 base_thread_const.rb:4 sleep_forever> rb_thread_t:0x000064627f26fd10 native:0x00007c906e9bc6c0 int:0 /home/wanabe/work/prog/ruby/ruby/tmp/master/b.rb:3:in '<class:B>' /home/wanabe/work/prog/ruby/ruby/tmp/master/b.rb:1:in '<top (required)>' base_thread_const.rb:4:in 'Kernel#require' base_thread_const.rb:4:in 'block in <main>' from base_thread_const.rb:5:in '<main>' 1 $ (git checkout 6fbc32b5d0da31535cccc0eca1853273313a0b52; make -j miniruby) >/dev/null && ./miniruby -v base_thread_const.rb ; echo $? Previous HEAD position was 30e5e7c005f Revert "Fix jump buffer leak in setjmp handler in WASI builds" HEAD is now at 6fbc32b5d0d GCC defines __linux__, not __LINUX__ ruby 3.5.0dev (2025-04-01T16:11:01Z v3_5_0_preview1~146 30e5e7c005) +PRISM [x86_64-linux] ["/home/wanabe/work/prog/ruby/ruby/tmp/master/b.rb", 3, [:a1]] ["/home/wanabe/work/prog/ruby/ruby/tmp/master/a.rb", 3, [:b1, :b2]] 0 ``` And for reasons unknown, it depends on the parser. This problem still occurs when using parse.y. ``` $ for parser in prism parse.y; do ./miniruby -v --parser=$parser; for i in $(seq 1 1 1000); do ./miniruby -v --parser=$parser base_thread_const.rb || break; done > /dev/null; done ruby 4.1.0dev (2026-08-02T09:00:37Z :detached: 63c7beb2ef) +PRISM [x86_64-linux] ruby 4.1.0dev (2026-08-02T09:00:37Z :detached: 63c7beb2ef) [x86_64-linux] base_thread_const.rb:5:in 'Thread#join': No live threads left. Deadlock? (fatal) 3 threads, 3 sleeps current:0x00005f6d13cab8c0 main thread:0x00005f6cf8483400 * #<Thread:0x00007dea51d58310 sleep_forever> rb_thread_t:0x00005f6cf8483400 native:0x00007dea522c97c0 int:0 base_thread_const.rb:5:in 'Thread#join' base_thread_const.rb:5:in '<main>' * #<Thread:0x00007dea51c9fe00 base_thread_const.rb:3 sleep_forever> rb_thread_t:0x00005f6d13cab8c0 native:0x00007dea36fbe6c0 int:0 depended by: tb_thread_id:0x00005f6cf8483400 /home/wanabe/work/prog/ruby/ruby/tmp/master/a.rb:3:in '<class:A>' /home/wanabe/work/prog/ruby/ruby/tmp/master/a.rb:1:in '<top (required)>' base_thread_const.rb:3:in 'Kernel#require' base_thread_const.rb:3:in 'block in <main>' * #<Thread:0x00007dea51c9ff18 base_thread_const.rb:4 sleep_forever> rb_thread_t:0x00005f6d13cabfc0 native:0x00007dea36ebd6c0 int:0 mutex:2 cond:1 /home/wanabe/work/prog/ruby/ruby/tmp/master/b.rb:3:in '<class:B>' /home/wanabe/work/prog/ruby/ruby/tmp/master/b.rb:1:in '<top (required)>' base_thread_const.rb:4:in 'Kernel#require' base_thread_const.rb:4:in 'block in <main>' from base_thread_const.rb:5:in '<main>' ``` ---------------------------------------- Bug #15598: Deadlock on mutual reference of autoloaded constants https://bugs.ruby-lang.org/issues/15598#change-118305 * Author: akr (Akira Tanaka) * Status: Open * ruby -v: ruby 2.7.0dev (2019-02-11 trunk 67049) [x86_64-linux] * Backport: 2.4: UNKNOWN, 2.5: UNKNOWN, 2.6: UNKNOWN ---------------------------------------- Mutual reference of autoloaded constants can cause deadlock sporadically. Assume A is defined in a.rb and it uses B at loading time. Also, B is defined in b.rb and it uses A at loading time. ``` % cat a.rb class A def a1() end p [__FILE__, __LINE__, B.instance_methods(false)] def a2() end end % cat b.rb class B def b1() end p [__FILE__, __LINE__, A.instance_methods(false)] def b2() end end ``` If they are loaded via autoload and constants are referenced sequentially, it works (no error, at least). However, incomplete A (which a2 is not defined) is appear in b.rb, though. ``` % cat base_seq.rb autoload :A, "./a" autoload :B, "./b" A B % ruby base_seq.rb ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1, :b2]] ``` However, the constants are referenced in multi threads, deadlock can occur, or works like sequential version, sporadically. ``` % cat base_thread_const.rb autoload :A, "./a" autoload :B, "./b" t1 = Thread.new { A } t2 = Thread.new { B } t1.join t2.join % ruby base_thread_const.rb Traceback (most recent call last): 1: from base_thread_const.rb:5:in `<main>' base_thread_const.rb:5:in `join': No live threads left. Deadlock? (fatal) 3 threads, 3 sleeps current:0x000055f9e2fa1b00 main thread:0x000055f9e2ec14b0 * #<Thread:0x000055f9e2eef188 sleep_forever> rb_thread_t:0x000055f9e2ec14b0 native:0x00007f259bc54b40 int:0 base_thread_const.rb:5:in `join' base_thread_const.rb:5:in `<main>' * #<Thread:0x000055f9e31ece30@base_thread_const.rb:3 sleep_forever> rb_thread_t:0x000055f9e31403c0 native:0x00007f2597e99700 int:0 depended by: tb_thread_id:0x000055f9e2ec14b0 /tmp/h/a.rb:3:in `<class:A>' /tmp/h/a.rb:1:in `<top (required)>' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' base_thread_const.rb:3:in `block in <main>' * #<Thread:0x000055f9e31ecbb0@base_thread_const.rb:4 sleep_forever> rb_thread_t:0x000055f9e2fa1b00 native:0x00007f258ffff700 int:0 /tmp/h/b.rb:3:in `<class:B>' /tmp/h/b.rb:1:in `<top (required)>' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' base_thread_const.rb:4:in `block in <main>' % ruby base_thread_const.rb ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1, :b2]] ``` Also, if "require" is used instead of constant references in the threads, deadlock can occur (sporadically) too. Note that incomplete A can appear in b.rb and incomplete B can appear in a.rb. The incompleteness vary. ``` % cat base_thread_require.rb autoload :A, "./a" autoload :B, "./b" t1 = Thread.new { require './a' } t2 = Thread.new { require './b' } t1.join t2.join % ruby base_thread_require.rb Traceback (most recent call last): 1: from base_thread_require.rb:5:in `<main>' base_thread_require.rb:5:in `join': No live threads left. Deadlock? (fatal) 3 threads, 3 sleeps current:0x00005591a27f5190 main thread:0x00005591a24264b0 * #<Thread:0x00005591a24531a0 sleep_forever> rb_thread_t:0x00005591a24264b0 native:0x00007feced36ab40 int:0 base_thread_require.rb:5:in `join' base_thread_require.rb:5:in `<main>' * #<Thread:0x00005591a2754cc8@base_thread_require.rb:3 sleep_forever> rb_thread_t:0x00005591a27f5190 native:0x00007fece95af700 int:0 depended by: tb_thread_id:0x00005591a24264b0 /tmp/h/a.rb:1:in `<top (required)>' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' base_thread_require.rb:3:in `block in <main>' * #<Thread:0x00005591a2754a98@base_thread_require.rb:4 sleep_forever> rb_thread_t:0x00005591a2506b00 native:0x00007fece13ad700 int:0 mutex:0x00005591a27f5190 cond:1 /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' /tmp/h/b.rb:3:in `<class:B>' /tmp/h/b.rb:1:in `<top (required)>' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' /home/akr/ruby/o0/lib/ruby/2.7.0/rubygems/core_ext/kernel_require.rb:54:in `require' base_thread_require.rb:4:in `block in <main>' % ruby base_thread_require.rb ["/tmp/h/b.rb", 3, []] ["/tmp/h/a.rb", 3, [:b1]] % repeat 100 (ruby base_thread_require.rb >& /tmp/z && cat /tmp/z) ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/b.rb", 3, []] ["/tmp/h/a.rb", 3, [:b1]] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1, :b2]] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1, :b2]] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1, :b2]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1, :b2]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1]] ["/tmp/h/b.rb", 3, [:a1, :a2]] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, [:b1, :b2]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ["/tmp/h/a.rb", 3, []] ["/tmp/h/b.rb", 3, [:a1]] ``` I think there are several ways to solve this issue. - Prohibit mutual reference. I.e. raise an error at autoload constant reference currently loading. Since mutual reference causes incomplete definition, it is dangerous even with single thread. However, if real application uses such code, this is incompatible. - More coarse locking. Since the deadlock is caused because two threads lock the constants in different order: A to B and B to A. I think it is possible to fix this issue by locking whole autoloading procedure by single lock, namely "global autoload lock". Note that it should also be locked by "require" method if it load a file for autoload. -- https://bugs.ruby-lang.org/

1 0