[pull] master from ruby:master #722
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Though very unlikely, it could potentially allocate a large array of whitespace. ruby/prism@3389947819
timer_thread_check_exceed() was returning true when the remaining time was less than 1ms, treating it as "too short time". This caused sub-millisecond sleeps (like sleep(0.0001)) to return immediately instead of actually sleeping. The fix removes this optimization that was incorrectly short-circuiting short sleep durations. Now the timeout is only considered exceeded when the actual deadline has passed. Note: There's still a separate performance issue where MN_THREADS mode is slower for sub-millisecond sleeps due to the timer thread using millisecond-resolution polling. This will require a separate fix to use sub-millisecond timeouts in kqueue/epoll. [Bug #21836]
[Feature #21846] There is a single path through our GC Sweeping code, and we always call rb_gc_obj_free_vm_weak_references and rb_gc_obj_free before adding the object back to the freelist. We do this even when the object has no external resources that require being free'd and has no weak references pointing to it. This commit introduces a conservative fast path through gc_sweep_plane that uses the object flags to identify certain cases where these calls can be skipped - for these objects we just add them straight back on the freelist. Any object for which gc_sweep_fast_path_p returns false will use the current full sweep code (referred to here as the slow path). Currently there are 2 checks that will _always_ require an object to go down the slow path: 1. Has it's object_id been observed and stored in the id2ref_table 2. Has it got generic ivars in the gen_fields table If neither of these are true, then we run some flag checks on the object and send the following cases down the fast path: - Objects that are not heap allocated - Embedded strings that aren't in the fstring table - Embedded Arrays - Embedded Hashes - Embedded Bignums - Embedded Strings - Floats, Rationals and Complex - Various IMEMO subtypes that do no allocation We've benchmarked this code using ruby-bench as well as the gcbench benchmarks inside Ruby (benchmarks/gc) and this patch results in a modest speed improvement on almost all of the headline benchmarks (2% in railsbench with YJIT enabled), and an observable 30% improvement in time spent sweeping during the GC benchmarks: ``` master: ruby 4.1.0dev (2026-01-19T12:03:33Z master 859920d) +YJIT +PRISM [x86_64-linux] experiment: ruby 4.1.0dev (2026-01-16T21:36:46Z mvh-sweep-fast-pat.. c3ffe37) +YJIT +PRISM [x86_64-linux] -------------- ----------- ---------- --------------- ---------- ------------------ ----------------- bench master (ms) stddev (%) experiment (ms) stddev (%) experiment 1st itr master/experiment lobsters N/A N/A N/A N/A N/A N/A activerecord 132.5 0.9 132.5 1.0 1.056 1.001 chunky-png 577.2 0.4 580.1 0.4 0.994 0.995 erubi-rails 902.9 0.2 894.3 0.2 1.040 1.010 hexapdf 1763.9 3.3 1760.6 3.7 1.027 1.002 liquid-c 56.9 0.6 56.7 1.4 1.004 1.003 liquid-compile 46.3 2.1 46.1 2.1 1.005 1.004 liquid-render 77.8 0.8 75.1 0.9 1.023 1.036 mail 114.7 0.4 113.0 1.4 1.054 1.015 psych-load 1635.4 1.4 1625.9 0.5 0.988 1.006 railsbench 1685.4 2.4 1650.1 2.0 0.989 1.021 rubocop 133.5 8.1 130.3 7.8 1.002 1.024 ruby-lsp 140.3 1.9 137.5 1.8 1.007 1.020 sequel 64.6 0.7 63.9 0.7 1.003 1.011 shipit 1196.2 4.3 1181.5 4.2 1.003 1.012 -------------- ----------- ---------- --------------- ---------- ------------------ ----------------- Legend: - experiment 1st itr: ratio of master/experiment time for the first benchmarking iteration. - master/experiment: ratio of master/experiment time. Higher is better for experiment. Above 1 represents a speedup. ``` ``` Benchmark │ Wall(B) Sweep(B) Mark(B) │ Wall(E) Sweep(E) Mark(E) │ Wall Δ Sweep Δ ───────────────┼─────────────────────────────────┼─────────────────────────────────┼────────────────── null │ 0.000s 1ms 4ms │ 0.000s 1ms 4ms │ 0% 0% hash1 │ 4.330s 875ms 46ms │ 3.960s 531ms 44ms │ +8.6% +39.3% hash2 │ 6.356s 243ms 988ms │ 6.298s 176ms 1.03s │ +0.9% +27.6% rdoc │ 37.337s 2.42s 1.09s │ 36.678s 2.11s 1.20s │ +1.8% +13.1% binary_trees │ 3.366s 426ms 252ms │ 3.082s 275ms 239ms │ +8.4% +35.4% ring │ 5.252s 14ms 2.47s │ 5.327s 12ms 2.43s │ -1.4% +14.3% redblack │ 2.966s 28ms 41ms │ 2.940s 21ms 38ms │ +0.9% +25.0% ───────────────┼─────────────────────────────────┼─────────────────────────────────┼────────────────── Legend: (B) = Baseline, (E) = Experiment, Δ = improvement (positive = faster) Wall = total wallclock, Sweep = GC sweeping time, Mark = GC marking time Times are median of 3 runs ``` These results are also borne out when YJIT is disabled: ``` master: ruby 4.1.0dev (2026-01-19T12:03:33Z master 859920d) +PRISM [x86_64-linux] experiment: ruby 4.1.0dev (2026-01-16T21:36:46Z mvh-sweep-fast-pat.. c3ffe37) +PRISM [x86_64-linux] -------------- ----------- ---------- --------------- ---------- ------------------ ----------------- bench master (ms) stddev (%) experiment (ms) stddev (%) experiment 1st itr master/experiment lobsters N/A N/A N/A N/A N/A N/A activerecord 389.6 0.3 377.5 0.3 1.032 1.032 chunky-png 1123.4 0.2 1109.2 0.2 1.013 1.013 erubi-rails 1754.3 0.1 1725.7 0.1 1.035 1.017 hexapdf 3346.5 0.9 3326.9 0.7 1.003 1.006 liquid-c 84.0 0.5 83.5 0.5 0.992 1.006 liquid-compile 74.0 1.5 73.5 1.4 1.011 1.008 liquid-render 199.9 0.4 199.6 0.4 1.000 1.002 mail 177.8 0.4 176.4 0.4 1.069 1.008 psych-load 2749.6 0.7 2777.0 0.0 0.980 0.990 railsbench 2983.0 1.0 2965.5 0.8 1.041 1.006 rubocop 228.8 1.0 227.5 1.2 1.015 1.005 ruby-lsp 221.8 0.9 216.1 0.8 1.011 1.026 sequel 89.1 0.5 89.1 1.8 1.005 1.000 shipit 2385.6 1.6 2371.8 1.0 1.002 1.006 -------------- ----------- ---------- --------------- ---------- ------------------ ----------------- Legend: - experiment 1st itr: ratio of master/experiment time for the first benchmarking iteration. - master/experiment: ratio of master/experiment time. Higher is better for experiment. Above 1 represents a speedup. ``` ``` Benchmark │ Wall(B) Sweep(B) Mark(B) │ Wall(E) Sweep(E) Mark(E) │ Wall Δ Sweep Δ ───────────────┼─────────────────────────────────┼─────────────────────────────────┼────────────────── null │ 0.000s 1ms 4ms │ 0.000s 1ms 3ms │ 0% 0% hash1 │ 4.349s 877ms 45ms │ 4.045s 532ms 44ms │ +7.0% +39.3% hash2 │ 6.575s 235ms 967ms │ 6.540s 181ms 1.04s │ +0.5% +23.0% rdoc │ 45.782s 2.23s 1.14s │ 44.925s 1.90s 1.01s │ +1.9% +15.0% binary_trees │ 6.433s 426ms 252ms │ 6.268s 278ms 240ms │ +2.6% +34.7% ring │ 6.584s 17ms 2.33s │ 6.738s 13ms 2.33s │ -2.3% +30.8% redblack │ 13.334s 31ms 42ms │ 13.296s 24ms 107ms │ +0.3% +22.6% ───────────────┼─────────────────────────────────┼─────────────────────────────────┼────────────────── Legend: (B) = Baseline, (E) = Experiment, Δ = improvement (positive = faster) Wall = total wallclock, Sweep = GC sweeping time, Mark = GC marking time Times are median of 3 runs ```
It relies too much on VM level concerns, such that it can't be built with modular GC enabled. We'll move it into the VM, and then expose it to the GC implementations so they can use it.
Most compilers will optimise this anyway
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )