The string budget's gate reads the measured mark share, whatever SPINEL_GC_OBJ_BUDGET pins

[namespaceMarcello]

What this changes

A follow-up to #8444: the two review findings it was merged with, both real, and a correction to one sentence of its description.

The string budget's gate read what SPINEL_GC_OBJ_BUDGET made of the mark share, not the share. sp_str_retune gated the share of the object old generation on sp_gc_obj_alpha1024. That is the object budget's decision: obj sets it to 0, walk to 1, and under fixed it stays at its initial 1, because the object retune returned before measuring anything. So pinning the object budget pinned the string budget too:

run	before	after
test/gc_str_budget_objects.rb, SPINEL_GC_OBJ_BUDGET=obj	224 collections, old obj in str trigger 0.00 MB	44 collections, 5.80 MB, as the default
test/gc_str_budget_still.rb, SPINEL_GC_OBJ_BUDGET=walk	33 collections, as SPINEL_GC_STR_BUDGET=walk	106, as the default
the same, SPINEL_GC_OBJ_BUDGET=fixed SPINEL_GC_THRESHOLD_OBJ_KB=65536	17 collections	86 (17 with SPINEL_GC_STR_BUDGET=walk beside it)

The measurement has its own variable now (sp_gc_mark_share1024, static), written before those knobs are applied, and fixed leaves sp_gc_retune_object after the measurement instead of at the top. Each knob pins its own budget. With neither set the two values are the same number, so nothing else moves: the default, walk and str arms of both programs print the same [gc] lines as on master.

old obj in str trigger kept a share the trigger no longer carried. A string sweep that reclaims under a quarter doubles the budget as it stands and puts no share in, but the reading stayed at what the last productive sweep had left. It is zero after such a sweep now. test/gc_str_budget_kept.rb (new) goes through that branch: 2.80 MB before, 0.00 after. Its kept Strings are 3 KB each because Strings the slab takes never reached the branch here: kept alive in ten different ways, every sweep of them read as 40% reclaimed. I have not chased why.

gc-str-budget-test holds both: four more runs under SPINEL_GC_OBJ_BUDGET and the new program, read against str (the share was carried) and against zero (the reading was cleared). With lib/sp_alloc.c put back as #8444 left it, four of the new checks fail; of 17 mutations of the final code 16 are red, and the green one is the share at 1/4, which the leg leaves free on purpose. The other six collector legs pass.

A correction to #8444's description. It said that a scheduled string major skipped when it has nothing to free "would also take away the gc_str_budget_still cost". Measured since, it does not: with those majors skipped the program still peaks at 56 MB. The peak is one wide cycle at the change of phase, when the first string sweep after the heap is built reads the mark share of the last collection of the building phase; it is not the cycles after the full marks. Wide cycles among the string retunes of that program: 13 of 86 with the schedule as master has it, 6 of 237 with the scheduled majors skipped, 2 of 323 with full collections kept out of the gate altogether, and 56 MB each time against 26 MB under SPINEL_GC_STR_BUDGET=str. So that cost stands as the description gave it, and nothing I have tried takes it away.

Counts and megabytes above are from one laptop under Docker (gcc 14.2, linux-x86_64); none of them is a time.

make gate (on this branch merged with current master)

The gate ran on this branch as it stands, on master ad681d651 (the Gate: trailer). Master is at 00802fd5e as this opens, 34 commits later; none of them touches lib/sp_alloc.c or lib/sp_gc.c, and the branch merges cleanly. On that merged tree I ran again the three string-budget tests and the seven collector legs (gc-str-budget-test, gc-obj-budget-test, gc-str-major-test, gc-threshold-test, gc-phases-test, gc-stress-test, gc-minor-test), all passing. I did not run the whole gate a second time.

Tests: 6831 pass, 0 fail, 0 error
gate-test-shared: 6829 pass, 2 known failures, 0 new failures
scale-test: boxed Hash store work at 2x the methods is 1.95x (limit 2.20)
scale-test: boxed-receiver alias work at 2x the writes is 1.86x (limit 2.20)
scale-test: instance_eval forwarding work at 2x the wrappers is 1.71x (limit 2.50)
scale-test: work at 4x the program is 5.00x (linear 4.00, limit 5.20)
scale-test: work at 4x the program, compiled to C, is 6.27x (limit 6.90)
scale-test: call-shape work at 4x the units, compiled to C, is 4.15x (linear 4.00, limit 4.50)
gate: ALL GREEN

• New tests have .expected files that match CRuby 4.0 run with --enable-frozen-string-literal
• Values past 2^31 are marked # spinel: int64 (none: the new program stays under 2^31)
• If optcarrot's generated C changed: callgrind numbers, checksum 59662 (the generated C does not change: the diff is in lib/)
• Depends on: #

Written by Claude Code (AI assistant) on behalf of @namespaceMarcello, who directs this work.

Summary by CodeRabbit

• Bug Fixes
  • String-budget behavior now uses the measured mark share independently of the object-budget setting. Unproductive string sweeps report a zero share.
• Tests
  • Added coverage for object-budget modes and fixed-budget behavior, including checks that these settings do not alter the string budget’s gate.
  • Added a workload to verify collection counts and measured shares with retained strings.
• Documentation
  • Clarified how string and object budgets use measured shares and operate independently.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: cf32c234-ffa8-4251-9b2a-25078f97bed6

---

📥 Commits

Reviewing files that changed from the base of the PR and between 00802fd and 51a3ee7.

---

📒 Files selected for processing (5)

• Makefile
• docs/thread.md
• lib/sp_alloc.c
• test/gc_str_budget_kept.rb
• test/gc_str_budget_kept.rb.expected

---

Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 7 remain after this review.

---

---

📝 Walkthrough

Walkthrough

The GC now retains the measured mark share separately from the object-budget mode. The string budget uses that share, and an unproductive sweep resets the reported object-share bytes. New workload and Makefile checks cover these cases.

Changes

GC string budget

Layer / File(s)	Summary
Retain measured mark share lib/sp_alloc.c	The object retune records the measured mark share before applying object-budget settings. A fixed object budget no longer skips that measurement.
Apply measured share to string budget lib/sp_alloc.c, docs/thread.md	The string-budget calculation uses the measured share regardless of the object-budget mode. An unproductive sweep resets the reported object-share bytes to zero. The documentation describes these behaviors.
Exercise budget modes and kept strings test/gc_str_budget_kept.rb, test/gc_str_budget_kept.rb.expected, Makefile	The new workload retains 8,000 strings of 3,000 characters. Makefile checks compare collection counts across budget modes and verify that the final measured share is zero.

Priority: ➖ Normal

Estimated code review effort: 3 (Moderate) | ~20 minutes

Change: Bug fix

---

Merge Risk: ⚪ Minimal · up to 51a3e

The string-budget gate now uses the measured mark share independently of the object-budget setting, and unproductive sweeps report zero share. Added regression cases cover the relevant budget modes and retained strings; no concrete blocker remains before normal checks.

Pre-merge checks | 4 | 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 75.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 4 functions across 2 files. (3 skipped: 3…	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check		Check skipped - CodeRabbit’s high-level summary is enabled.
Title check		The title clearly and specifically describes the main change: the string-budget gate now uses the measured mark share independently of object-budget pinning.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

---

Full details: Docstring Coverage

Explanation

Docstring coverage is 75.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 4 functions across 2 files. (3 skipped: 3 unsupported.)

---

• Fix all pre-merge checks with AI

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

---

_{Comment @coderabbitai help to get the list of available commands.}