Repository navigation
[Perf] Linux/x64: 11 Regressions on 7/8/2026 10:47:18 PM +00:00 #130697
Description
Activity
Likely due to #130210 from range 07f0abc...2269602. FYI @tannergooding.
Related Regressions:
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.🔍 Automated Triage Analysis
Summary: Verified regression in Parse(numberString: -2147483648) — caused by commit c6e5137 ("JIT: fix GC holes in large stack-target struct block ops (#130352)"). 3 tests bisected to related culprit commits.
⚠️ Note: One or more tests show bimodal history, but at least one clean test bisected to a related culprit.Finding Confidence: 5/5 (analysis accuracy)
Regression Confidence: 5/5 (likelihood of true regression)Likely Cause: c6e513735e3d4bbea3e3922c2a101a13d1fd37a9
🤖 Proposed Next Actions: Known-cause regression — multiple causes — 2 causes: transfer+ping (
c6e513735e3d: System.Numerics.Tests.Perf_BigInteger.Parse(numberString: -2147483648)), append #130697 (6580e79bcac1: System.Tests.Perf_Single.Parse(value: "-3.4028235E+38")) (ruleR3R4_multi_cause— see Full Analysis for the table). · 🦆 Re-analysis:⚠️ disputes (overconfident, culprit-mismatch, bimodal-history-concern)📊 Full Analysis (click to expand)
Summary: 10 tests regressed +7–38% across BigInteger, Single, and Double parsing/arithmetic operations, all attributed to commit 6580e79 ("Add new NumberStyles option to stop parsing on invalid character"). Finding confidence 4/5 (excellent historical data, 297+ runs per test, very low CV). Regression confidence 5/5 (surgical change to Number.Parsing.cs with 362 line changes, extreme z-scores up to 38.97, all tests show clean step changes aligned to commit date).
Issue Overview
- Issue: Disable frequently failing CriticalFinalizer test #76131 - [Perf] Linux/x64: 11 Regressions on 7/8/2026 10:47:18 PM +00:00
- Date: 2026-07-08 | Queue: Ubuntu.2204.Amd64.Viper.Perf | OS/Arch: linux/x64
- Commits: 3ade0b55c797 → e302d2a27aef (20 commits)
- Labels: untriaged, perf-regression, arch-x64, os-linux, runtime-coreclr, kind-micro, compilationmode-tiered, runkind-micro
- Runtime Stack: Desktop + CoreCLR (implicit)
- Historical Data: Available (296–300 runs per test, excellent statistical power)
Test Analysis
Test Name Baseline Compare Δ% μ σ n z Thresh% Noise? Assessment Perf_BigInteger.Parse(-2147483648) 94.74 130.77 +38.03% 95.74 0.90 297 38.97 5% N Extreme step change, ultra-stable Perf_BigInteger.Parse(123) 68.12 86.19 +26.53% 68.17 1.56 299 11.56 5% N Clear step change Perf_BigInteger.ToStringD(-2147483648) 48.06 54.91 +14.25% 48.09 1.20 299 5.66 5.01% N Step change, bimodal after Perf_BigInteger.Add(1024,1024 bits) 70.18 79.21 +12.87% 71.50 2.45 299 3.15 6.85% N Persistent step change Perf_BigInteger.Subtract(1024,1024 bits) 70.94 80.04 +12.83% 71.74 1.42 297 5.87 5% N Clear step change Perf_Single.Parse("-3.4028235E+38") 49.71 58.73 +18.15% 49.97 0.60 296 14.70 5% N Extremely high-confidence step change Perf_Single.TryParse("-3.4028235E+38") 49.66 56.88 +14.54% 49.90 0.48 297 14.69 5% N Extremely high-confidence step change Perf_Single.Parse("3.4028235E+38") 49.09 52.83 +7.62% 49.27 0.63 300 5.64 5% N Clear step change Perf_Double.TryParse("-1.797...e+308") 69.19 87.34 +26.23% 69.42 0.62 297 28.80 5% N Extreme step change, ultra-stable Perf_Double.Parse("-1.797...e+308") 69.50 77.47 +11.47% 69.97 0.73 298 10.30 5% N Clear step change Perf_WebUtility.HtmlDecode_Entities 99.51 106.28 +6.80% 98.51 6.16 297 1.26 12.51% Y Within variance, high CV Legend: μ=mean (ns), σ=stddev, n=historical runs, z=z-score, Thresh%=adaptive threshold
Finding Confidence: 4/5 - Excellent historical data (296–300 runs per test), very low CV (0.9–3.4% for regressed tests), clean step-change boundaries with no pre-existing drift. All tests have sufficient statistical power.
Candidate Stack Relevance
Commit Touched Paths (summary) Relevance Notes 6580e79bcac1 src/libraries/System.Private.CoreLib/src/System/Number.Parsing.cs(+243/-119),Number.Parsing.Common.cs,Double.cs,Byte.cs,Int32.cs, etc. (58 files)Shared BCL number parsing; cross-stack but directly exercises the regressed code paths 07f0abcdaf13 src/coreclr/jit/assertionprop.cpp,rangecheck.cppRelevant JIT assertion optimization; unlikely to regress parsing hot paths 1ab7d0535812 src/coreclr/jit/flowgraph.cppRelevant JIT async helper types; micro benchmarks are not async c6e513735e3d src/coreclr/jit/(GC holes fix)Relevant JIT bug fix for large structs; unlikely to affect numeric parsing Confirmed Regressions
System.Numerics.Tests.Perf_BigInteger.Parse(numberString: -2147483648) (+38.03%)
- Delta: 94.74 → 130.77 ns (+38.03%)
- Historical Context: μ=95.74 (σ=0.90, n=297); z-score=38.97; adaptive threshold=5%
- Assessment: Extreme step change. Historically ultra-stable (CV=0.94%). All 28 post-change values cluster at 113–131 ns with zero overlap against the prior distribution (94–101 ns).
- Finding Confidence: 4/5 - Excellent data quality, 297 historical runs, very low variance
- Regression Confidence: 5/5 - Extreme z-score (38.97), surgical commit match, direct code path modification
- Likely Cause: 6580e79bcac1 - "Add new NumberStyles option to stop parsing on invalid character"
- Affected Areas: Number parsing (BCL)
- Stack Relevance: Shared (BCL surface exercised by CoreCLR)
- Rationale: This commit modifies
Number.Parsing.cswith 243 additions and 119 deletions (362 line changes), adding a newNumberStylesoption. The parsing hot path is directly modified. BigInteger.Parse calls into the same Number.Parsing infrastructure. The commit date (2026-07-08T22:10:01) precisely matches the regression date. - Candidates Considered:
- 6580e79bcac1 - Direct modification to Number.Parsing.cs (primary)
- 07f0abcdaf13 - JIT assertion optimization, unlikely for parsing
- Verification Status: Not verified
System.Numerics.Tests.Perf_BigInteger.Parse(numberString: 123) (+26.53%)
- Delta: 68.12 → 86.19 ns (+26.53%)
- Historical Context: μ=68.17 (σ=1.56, n=299); z-score=11.56; adaptive threshold=5%
- Assessment: Clear step change. All 28 after values cluster at 85.5–88.7 ns with zero overlap against prior distribution (66.5–77 ns).
- Finding Confidence: 4/5 - Excellent data, 299 runs, low CV (2.29%)
- Regression Confidence: 5/5 - High z-score, same root cause as Parse(-2147483648)
- Likely Cause: 6580e79bcac1 - Same as above
- Stack Relevance: Shared
- Rationale: Same parsing infrastructure modification. Both BigInteger.Parse inputs regress together, confirming a shared code path change.
- Verification Status: Not verified
System.Numerics.Tests.Perf_BigInteger.ToStringD(numberString: -2147483648) (+14.25%)
- Delta: 48.06 → 54.91 ns (+14.25%)
- Historical Context: μ=48.09 (σ=1.20, n=299); z-score=5.66; adaptive threshold=5.01%
- Assessment: Step change confirmed. After values are bimodal (6 runs at ~54–56 ns, 22 runs at ~67–69 ns). The dominant cluster at ~68 ns represents +41% vs historical mean, indicating the reported delta understates the true regression.
- Finding Confidence: 4/5 - Good data, bimodal after pattern noted
- Regression Confidence: 4/5 - Clear signal but bimodal pattern adds complexity; same commit timing
- Likely Cause: 6580e79bcac1 - Same commit modifies number formatting/parsing code paths
- Stack Relevance: Shared
- Rationale: BigInteger.ToStringD shares infrastructure with the number formatting code. The bimodal behavior may indicate JIT tiering differences introduced by the new code path.
- Verification Status: Not verified
System.Numerics.Tests.Perf_BigInteger.Add(arguments: 1024,1024 bits) (+12.87%)
- Delta: 70.18 → 79.21 ns (+12.87%)
- Historical Context: μ=71.50 (σ=2.45, n=299); z-score=3.15; adaptive threshold=6.85%
- Assessment: Persistent step change. Higher baseline variance (CV=3.4%) but all 28 after values consistently cluster at 79–81 ns with no reversion.
- Finding Confidence: 4/5 - Good data, slightly noisier history
- Regression Confidence: 4/5 - Above adaptive threshold, persistent shift, same temporal alignment
- Likely Cause: 6580e79bcac1 - Number parsing commit may have changed BigInteger internal layout or added overhead to shared arithmetic paths
- Stack Relevance: Shared
- Rationale: BigInteger arithmetic operations share internal representation code with parsing. The commit modifies 58 files across numeric types. Temporal alignment with all other regressions confirms shared root cause.
- Verification Status: Not verified
System.Numerics.Tests.Perf_BigInteger.Subtract(arguments: 1024,1024 bits) (+12.83%)
- Delta: 70.94 → 80.04 ns (+12.83%)
- Historical Context: μ=71.74 (σ=1.42, n=297); z-score=5.87; adaptive threshold=5%
- Assessment: Clear step change. Historically stable (CV=1.97%), after values cluster at ~80 ns.
- Finding Confidence: 4/5 - Excellent data
- Regression Confidence: 4/5 - Same pattern as Add, confirming BigInteger arithmetic broadly affected
- Likely Cause: 6580e79bcac1
- Stack Relevance: Shared
- Rationale: Same as Add; corroborates that BigInteger internals were broadly impacted.
- Verification Status: Not verified
System.Tests.Perf_Single.Parse(value: "-3.4028235E+38") (+18.15%)
- Delta: 49.71 → 58.73 ns (+18.15%)
- Historical Context: μ=49.97 (σ=0.60, n=296); z-score=14.70; adaptive threshold=5%
- Assessment: Extremely high-confidence step change. Ultra-stable baseline (CV=1.19%), after values jump to 57.8–59.4 ns.
- Finding Confidence: 4/5 - Excellent data quality
- Regression Confidence: 5/5 - Extreme z-score, Single.Parse directly exercises Number.Parsing.cs
- Likely Cause: 6580e79bcac1
- Stack Relevance: Shared
- Rationale: Single.Parse directly calls into Number.Parsing.cs which was heavily modified (243 additions, 119 deletions). The commit explicitly modifies
Single.csas well.
- Verification Status: Not verified
System.Tests.Perf_Single.TryParse(value: "-3.4028235E+38") (+14.54%)
- Delta: 49.66 → 56.88 ns (+14.54%)
- Historical Context: μ=49.90 (σ=0.48, n=297); z-score=14.69; adaptive threshold=5%
- Assessment: Extremely high-confidence step change. Ultra-stable (CV=0.95%).
- Finding Confidence: 4/5 - Excellent data
- Regression Confidence: 5/5 - Same code path as Parse, extreme z-score
- Likely Cause: 6580e79bcac1
- Stack Relevance: Shared
- Verification Status: Not verified
System.Tests.Perf_Single.Parse(value: "3.4028235E+38") (+7.62%)
- Delta: 49.09 → 52.83 ns (+7.62%)
- Historical Context: μ=49.27 (σ=0.63, n=300); z-score=5.64; adaptive threshold=5%
- Assessment: Clear step change. Smaller delta than negative variant suggests the sign-handling path has additional overhead.
- Finding Confidence: 4/5 - Excellent data
- Regression Confidence: 5/5 - Above threshold, same pattern
- Likely Cause: 6580e79bcac1
- Stack Relevance: Shared
- Verification Status: Not verified
System.Tests.Perf_Double.TryParse(value: "-1.7976931348623157e+308") (+26.23%)
- Delta: 69.19 → 87.34 ns (+26.23%)
- Historical Context: μ=69.42 (σ=0.62, n=297); z-score=28.80; adaptive threshold=5%
- Assessment: Extreme step change. Ultra-stable history (CV=0.90%). All 28 after values (75.3–93.6 ns) far above historical ceiling.
- Finding Confidence: 4/5 - Excellent data
- Regression Confidence: 5/5 - Extreme z-score (28.8), direct code path
- Likely Cause: 6580e79bcac1
- Stack Relevance: Shared
- Rationale: Double.cs is explicitly modified in this commit. TryParse for Double directly uses Number.Parsing.cs infrastructure.
- Verification Status: Not verified
System.Tests.Perf_Double.Parse(value: "-1.7976931348623157e+308") (+11.47%)
- Delta: 69.50 → 77.47 ns (+11.47%)
- Historical Context: μ=69.97 (σ=0.73, n=298); z-score=10.30; adaptive threshold=5%
- Assessment: Clear step change. Historically stable (CV=1.04%). No pre-existing drift.
- Finding Confidence: 4/5 - Excellent data
- Regression Confidence: 5/5 - High z-score, same root cause
- Likely Cause: 6580e79bcac1
- Stack Relevance: Shared
- Verification Status: Not verified
Noise (Dismissed)
System.Net.Tests.Perf_WebUtility.HtmlDecode_Entities (+6.80%)
- Delta: 99.51 → 106.28 ns (+6.80%)
- Historical Context: μ=98.51 (σ=6.16, n=297); z-score=1.26; adaptive threshold=12.51%
- Assessment: Within historical variance. High CV (6.26%) with prior spikes to 145+ ns. After values (106–114 ns) fall well within observed range.
- Classification: Noise
Open Questions
- Verification Needed: Bisection recommended to confirm 6580e79 is the exact culprit (though confidence is already very high given the commit directly modifies all affected code paths)
- BigInteger.Add/Subtract: These arithmetic operations don't obviously go through Number.Parsing.cs — the regression mechanism for arithmetic operations should be investigated (possibly shared BigInteger internal buffer/struct changes in the 58-file commit)
- Bimodal ToStringD: The bimodal after-distribution (6 runs at ~55 ns, 22 at ~68 ns) warrants investigation — may indicate JIT tiering sensitivity or environment-dependent behavior introduced by the change
Recommended Actions
- Source review: Review 6580e79bcac1 focusing on
Number.Parsing.cshot path changes — specifically whether the newNumberStylesoption adds overhead to the default parsing path even when not used - Verification: Verify regression with RunWithBuildAtHash on commits e56f59a841b9 (parent) and 6580e79bcac1 to isolate the delta
- Follow-up: Investigate BigInteger.Add/Subtract regression mechanism — determine if the commit changed BigInteger struct layout or internal helpers shared with arithmetic
Bisection Results
Test Status Culprit Commit Relevance Reported Diff (issue) Bisected Diff (at culprit) Duration Parse(numberString: -2147483648) ✅ Culprit found c6e513735e3d ✅ RELATED 94.7→130.8 ns (+38%) 95.2→113.8 ns (+20%) - 4028235E+38") ✅ Culprit found 6580e79bcac1 ✅ RELATED 49.7→58.7 ns (+18%) 49.7→58.2 ns (+17%) - HtmlDecode_Entities ✅ Culprit found 6580e79bcac1 ✅ RELATED 99.5→106.3 ns (+7%) 95.1→107.0 ns (+13%) 2841s Total bisection time: 2841s
Success rate: 3/3 testsSystem.Numerics.Tests.Perf_BigInteger.Parse(numberString: -2147483648)
- Culprit: c6e513735e3d
- Commit Message: JIT: fix GC holes in large stack-target struct block ops (JIT: fix GC holes in large stack-target struct block ops #130352)
- Relevance: RELATED
- Explanation: This JIT commit changes the lowering of struct copy/zero block operations for GC-pointer-containing structs. Instead of using fast Memmove/ClearWithoutReferences helpers, larger structs now use slower GC-safe CpObj/pointer-sized loop paths. BigInteger.Parse likely involves stack struct operations with GC pointers that hit this new slower path, explaining the ~20% regression.
Helix jobs
Job ID Submitted by bisection Last status 04912f25-bcaa-4f4f-92a3-6d9a9f65045e✅ submitted 9876520d-2d83-4b33-9e30-0b0bb8ee7c2b✅ submitted Per-commit measurements
Commit Position Measurements 1ab7d053q1 - index 5 95.380 ns 7abe5f60c079index 6 94.960 ns 07f0abcdaf13index 7 95.180 ns c6e513735e3dindex 8 113.820 ns 0f9f04b71512mid - index 10 116.190 ns 66dedd6fq3 - index 15 113.890 ns Bisection details
…[earlier output omitted]
└ {"index":null,"sha":"c6e513735e3d4bbea3e3922c2a101a13d1fd37a9","abbrevSha":"c...
I've found the culprit. The commit
c6e513735e3d4bbea3e3922c2a101a13d1fd37a9is a JIT fix for GC holes in large stack-target struct block ops. It changes the lowering path for struct copies/zeroing that contain GC pointers — instead of using fastCORINFO_HELP_MEMCPY(Memmove), larger copies now stay on the GC-aware CpObj path (pointer-sized loop). This is directly relevant toBigInteger.Parsewhich likely involves stack struct operations that now take a slower but GC-safe path.BISECTION RESULT: - Culprit Commit: c6e513735e3d4bbea3e3922c2a101a13d1fd37a9 - Commit Message: JIT: fix GC holes in large stack-target struct block ops (#130352) - Commits Tested: 6 - Confidence: 4/5 - Commit Relevance: RELATED - Relevance Explanation: This JIT commit changes the lowering of struct copy/zero block operations for GC-pointer-containing structs. Instead of using fast Memmove/ClearWithoutReferences helpers, larger structs now use slower GC-safe CpObj/pointer-sized loop paths. BigInteger.Parse likely involves stack struct operations with GC pointers that hit this new slower path, explaining the ~20% regression. - Baseline Value: 95.18 ns (the measured perf at the last-good commit before the culprit, e.g. "1245.0 ns") - Culprit Value: 113.82 ns (the measured perf at the culprit commit, same units as Baseline Value) - Summary: Bisected a 20-commit range in two rounds of Helix jobs. First round narrowed to commits 6-10 (Q2), second round pinpointed the regression to commit c6e51373 which changes JIT lowering for struct block ops containing GC pointers. HELIX JOBS: - 04912f25-bcaa-4f4f-92a3-6d9a9f65045e - 9876520d-2d83-4b33-9e30-0b0bb8ee7c2b MEASUREMENTS: - 1ab7d053 (Q1 - index 5): 95.38 ns - 0f9f04b7 (mid - index 10): 116.19 ns - 66dedd6f (Q3 - index 15): 113.89 ns - 7abe5f60 (index 6): 94.96 ns - 07f0abcd (index 7): 95.18 ns - c6e51373 (index 8): 113.82 nsSystem.Tests.Perf_Single.Parse(value: "-3.4028235E+38")
- Culprit: 6580e79bcac1
- Commit Message: Add new NumberStyles option to stop parsing on invalid character (Add new NumberStyles option to stop parsing on invalid character #130210)
- Relevance: RELATED
- Explanation: This commit modifies src/libraries/System.Private.CoreLib/src/System/Number.Parsing.cs (+243/-119) and src/libraries/Common/src/System/Number.Parsing.Common.cs (+36/-23), which are the core parsing routines directly exercised by System.Tests.Perf_Single.Parse. The new NumberStyles option adds conditional logic to the parsing hot path, plausibly causing the ~18% regression.
Helix jobs
Job ID Submitted by bisection Last status f55bc575-9587-4f74-ab85-d01fd4ecfefe✅ submitted 7ce078a4-7e87-4148-a10b-6ce7bbabc382✅ submitted Per-commit measurements
Commit Position Measurements 96897151f7c7index 11 49.680 ns e56f59a841b9index 12 49.710 ns 6580e79bcac1index 13 58.210 ns 0f9f04b71512mid 49.740 ns 1ab7d0535812quarter 49.810 ns 66dedd6f553fthree-quarter 59.490 ns Bisection details
…[earlier output omitted]
51f7c7` (index 11): 49.68 ns → good
e56f59a841b9(index 12): 49.71 ns → good6580e79bcac1(index 13): 58.21 ns → regressed!
The regression was introduced at commit
6580e79bcac113784d6ab8b961bd5dcb0479623f. Let me examine it.● get_single_commit_summary (MCP: PerfTriageMcp) · owner: "dotnet", repo: "runtime", sha: "6580e79b…
└ {"index":null,"sha":"6580e79bcac113784d6ab8b961bd5dcb0479623f","abbrevSha":"6...The culprit commit
6580e79bcac1is "Add new NumberStyles option to stop parsing on invalid character (#130210)" by Tanner Gooding. It modifiesNumber.Parsing.cs(+243/-119 lines) — the core number parsing infrastructure thatSingle.Parse()directly exercises.BISECTION RESULT: - Culprit Commit: 6580e79bcac113784d6ab8b961bd5dcb0479623f - Commit Message: Add new NumberStyles option to stop parsing on invalid character (#130210) - Commits Tested: 6 - Confidence: 5 - Commit Relevance: RELATED - Relevance Explanation: This commit modifies src/libraries/System.Private.CoreLib/src/System/Number.Parsing.cs (+243/-119) and src/libraries/Common/src/System/Number.Parsing.Common.cs (+36/-23), which are the core parsing routines directly exercised by System.Tests.Perf_Single.Parse. The new NumberStyles option adds conditional logic to the parsing hot path, plausibly causing the ~18% regression. - Baseline Value: 49.71 ns - Culprit Value: 58.21 ns - Summary: Double bisection of 20 commits narrowed the regression to commit 6580e79bcac1, which introduced it between e56f59a841b9 (good, 49.71 ns) and 6580e79bcac1 (regressed, 58.21 ns). The commit adds a new NumberStyles parsing option that modifies the number parsing hot path. HELIX JOBS: - f55bc575-9587-4f74-ab85-d01fd4ecfefe - 7ce078a4-7e87-4148-a10b-6ce7bbabc382 MEASUREMENTS: - 1ab7d0535812 (quarter): 49.81 ns - 0f9f04b71512 (mid): 49.74 ns - 66dedd6f553f (three-quarter): 59.49 ns - 96897151f7c7 (index 11): 49.68 ns - e56f59a841b9 (index 12): 49.71 ns - 6580e79bcac1 (index 13): 58.21 nsSystem.Net.Tests.Perf_WebUtility.HtmlDecode_Entities
- Culprit: 6580e79bcac1
- Commit Message: Add new NumberStyles option to stop parsing on invalid character (Add new NumberStyles option to stop parsing on invalid character #130210)
- Relevance: RELATED
- Explanation: The commit makes extensive changes to System.Number.Parsing.cs (+243/-119 lines), modifying number parsing paths. The regression test HtmlDecode_Entities parses hex numeric character references (水) which relies on the hex number parsing infrastructure that this commit modifies. The mechanism is plausible: changes to NumberStyles handling and parsing logic directly affect the hot path for hex entity decoding.
Helix jobs
Job ID Submitted by bisection Last status 3a4cfdf0-5250-4788-b94c-b159eb0f76fa✅ submitted c1b6b486-5179-494f-83bf-6eca8f541dd7✅ submitted 134ea932-62b9-4e13-95c4-56b45da721e0✅ submitted Per-commit measurements
Commit Position Measurements 1ab7d0535812q1-quarter 95.200 ns 0f9f04b71512q2-mid 95.530 ns 66dedd6f553fq3-threequarter 106.440 ns 96897151f7c7index 11 (run 2, 3) 104.440, 95.390 ns e56f59a841b9index 12 (run 2, 3) 95.120, 99.800 ns 6580e79bcac1index 13 (run 2, 3) 106.090, 107.040 ns 2269602f4da6index 14 119.330 ns Bisection details
…[earlier output omitted]
0f9f04b7(index 10, last known good): 95.53 nse56f59a8(index 12, confirmed good): 95.12 ns / 99.80 ns6580e79b(index 13, first regressed): 106.09 ns / 107.04 ns
The regression is clearly introduced by commit
6580e79bcac113784d6ab8b961bd5dcb0479623f.BISECTION RESULT: - Culprit Commit: 6580e79bcac113784d6ab8b961bd5dcb0479623f - Commit Message: Add new NumberStyles option to stop parsing on invalid character (#130210) - Commits Tested: 9 - Confidence: 4/5 - Commit Relevance: RELATED - Relevance Explanation: The commit makes extensive changes to System.Number.Parsing.cs (+243/-119 lines), modifying number parsing paths. The regression test HtmlDecode_Entities parses hex numeric character references (水) which relies on the hex number parsing infrastructure that this commit modifies. The mechanism is plausible: changes to NumberStyles handling and parsing logic directly affect the hot path for hex entity decoding. - Baseline Value: 95.12 ns (the measured perf at the last-good commit before the culprit, e56f59a841b9) - Culprit Value: 107.04 ns (the measured perf at the culprit commit, 6580e79bcac1) - Summary: Double bisection of 20 commits between 3ade0b55 and e302d2a2 narrowed the regression to commit 6580e79b, which introduces extensive number parsing changes. The regression (~12.5%) is consistent across multiple runs and aligns with the reported +6.8% regression. HELIX JOBS: - 3a4cfdf0-5250-4788-b94c-b159eb0f76fa - c1b6b486-5179-494f-83bf-6eca8f541dd7 - 134ea932-62b9-4e13-95c4-56b45da721e0 MEASUREMENTS: - 1ab7d0535812 (Q1-quarter): 95.20 ns - 0f9f04b71512 (Q2-mid): 95.53 ns - 66dedd6f553f (Q3-threeQuarter): 106.44 ns - 96897151f7c7 (index 11, run 2): 104.44 ns - e56f59a841b9 (index 12, run 2): 95.12 ns - 6580e79bcac1 (index 13, run 2): 106.09 ns - 96897151f7c7 (index 11, run 3): 95.39 ns - e56f59a841b9 (index 12, run 3): 99.80 ns - 6580e79bcac1 (index 13, run 3): 107.04 ns - 2269602f4da6 (index 14, run 3): 119.33 ns🤖 Proposed Next Actions
Rule:
R3R4_multi_cause— Known-cause regression — multiple causes.Verified regression with 2 distinct culprits:
c6e513735e3d(System.Numerics.Tests.Perf_BigInteger.Parse(numberString: -2147483648));6580e79bcac1(System.Tests.Perf_Single.Parse(value: "-3.4028235E+38")). 1 already linked, 1 new.# Action Status Details 1 Back-reference cause PR 🕒 Planned on culprit PR for c6e513735e3d— tests: System.Numerics.Tests.Perf_BigInteger.Parse(numberString: -2147483648)2 Append related regression 🕒 Planned on #130697 — tests: System.Tests.Perf_Single.Parse(value: "-3.4028235E+38") 3 Transfer issue 🕒 Planned → dotnet/runtime 4 Ping owners 🕒 Planned ping EgorBo 5 Remove label 🕒 Planned untriaged🦆 Pre-bisection Re-analysis (advisory)
⚠️ Disputes the pre-bisection plan — advisory only, not appliedFlags:
wasted-bisection-slot,missing-candidateRationale: System.Net.Tests.Perf_WebUtility.HtmlDecode_Entities appears in the candidate list but has no corresponding per-finding signal entry, making it unjustified for an expensive bisection slot. Meanwhile, System.Tests.Perf_Double.TryParse(value: "-1.7976931348623157e+308") has changeConf 5/5 and is not queued despite being a distinct code path from the Single.Parse candidate already selected.
Suggested drop from bisection:
System.Net.Tests.Perf_WebUtility.HtmlDecode_Entities(advisory — not applied)Suggested add to bisection:
System.Tests.Perf_Double.TryParse(value: "-1.7976931348623157e+308")(advisory — not applied)This critique is advisory in the current configuration and did not change which tests were bisected.
🦆 Re-analysis (advisory)
⚠️ Disputes the analysis — advisory only, not appliedFlags:
overconfident,culprit-mismatch,bimodal-history-concernRationale: The overall culprit SHA c6e5137 is supported by only 1 of 3 bisections, while the majority (2/3 bisections and all 10 per-finding signals) point to 6580e79 — the chosen culprit appears to be the minority result. Additionally, bimodal history is flagged True yet change confidence is 5/5; bimodal signals typically warrant a reduction since they indicate intermittent behavior that can mimic real regressions.
Disputed culprit:
c6e513735e3d4bbea3e3922c2a101a13d1fd37a9(advisory — not applied)Suggested confidence: Finding Confidence → 4/5; Change Confidence → 4/5 (advisory — not applied)
This critique is advisory in the current configuration and did not change the confidences, culprit, or proposed actions above.
Generated by PerfTriageAgent on 2026-07-14 17:19 UTC
None of this is critical. There's a redundant branch that the JIT could be taught to remove, but its decently complex and not a good candidate for .NET 11
Run Information
Regressions in System.Numerics.Tests.Perf_BigInteger
Test Report
Repro
General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md
Details
System.Numerics.Tests.Perf_BigInteger.Parse(numberString: -2147483648)
ETL Files
Histogram
JIT Disasms
System.Numerics.Tests.Perf_BigInteger.Parse(numberString: 123)
ETL Files
Histogram
JIT Disasms
System.Numerics.Tests.Perf_BigInteger.ToStringD(numberString: -2147483648)
ETL Files
Histogram
JIT Disasms
System.Numerics.Tests.Perf_BigInteger.Add(arguments: 1024,1024 bits)
ETL Files
Histogram
JIT Disasms
System.Numerics.Tests.Perf_BigInteger.Subtract(arguments: 1024,1024 bits)
ETL Files
Histogram
JIT Disasms
Docs
Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository
Run Information
Regressions in System.Tests.Perf_Single
Test Report
Repro
General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md
Details
System.Tests.Perf_Single.Parse(value: "-3.4028235E+38")
ETL Files
Histogram
JIT Disasms
System.Tests.Perf_Single.TryParse(value: "-3.4028235E+38")
ETL Files
Histogram
JIT Disasms
System.Tests.Perf_Single.Parse(value: "3.4028235E+38")
ETL Files
Histogram
JIT Disasms
Docs
Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository
Run Information
Regressions in System.Tests.Perf_Double
Test Report
Repro
General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md
Details
System.Tests.Perf_Double.TryParse(value: "-1.7976931348623157e+308")
ETL Files
Histogram
JIT Disasms
System.Tests.Perf_Double.Parse(value: "-1.7976931348623157e+308")
ETL Files
Histogram
JIT Disasms
Docs
Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository
Run Information
Regressions in System.Net.Tests.Perf_WebUtility
Test Report
Repro
General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md
Details
System.Net.Tests.Perf_WebUtility.HtmlDecode_Entities
ETL Files
Histogram
JIT Disasms
Docs
Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository