[Issue Backlog] Backlog Analysis - March 9, 2026

[github-actions[bot]]

Issue Backlog Analysis - March 9, 2026

Executive Summary

• Total open issues analyzed: 143
• New issues since last run (Feb 24): 7
• Critical soundness bugs: 23
• Performance regressions: 11
• Issues with suggested fixes/analysis: 3
• Potential duplicate clusters: 1

---

Issues Requiring Immediate Attention

Critical Soundness Bugs

Z3 currently has 23 open soundness issues. These produce incorrect results and undermine trust in formal verification workflows. Priority should be given to:

Issue	Title	Status	Notes
#8185	Incorrect model in mixed FP/Real logic triggered by extra string constraint	Active	Related to #8194; proof mode + string bug
#8194	Incorrect model with :produce-proofs true involving String Theory and UF	Active	Related to #8185; proof mode + string bug
#8023	Soundness issue in string logic	Active (8 comments)	Active investigation
#7841	Incorrect model for QF_SLIA	Needs attention	1 comment, may be stale
#7990	Wrong verdict with (set-option :sat.smt true)	No activity	0 comments since Oct 2025
#7842	Incorrect model	No activity	0 comments since Sep 2025
#7611	Spacer unsoundness regarding CHCs involving inductively defined datatypes	Limited activity	1 comment

Pattern identified: Issues #8185 and #8194 both involve incorrect models when :produce-proofs true is enabled with String theory. This suggests a systematic bug in proof generation mode's interaction with the string solver.

---

New Issues Since Last Run (Feb 24, 2026)

Performance Regressions

#8893 - String solver performance regression in 4.16

• Severity: High
• Component: String solver (QF_SLIA)
• Impact: 25x slowdown (2.2s → 51s) on regex matching benchmarks
• Platforms: Both macOS arm64 and Linux amd64
• Status: No comments yet, needs investigation
• Suggested action: Bisect commits between 4.15.2 and 4.16.0 to identify regression

Multi-Objective Optimization

#8890 - z3-solver==4.16.0 finds less solutions for BNH problem than z3-solver==4.8.12

• Severity: Medium
• Component: Python bindings / core solver
• Impact: Fewer solutions found in optimization problems (7 solutions → 5 solutions)
• Status: No comments, needs clarification whether this is soundness or completeness issue
• Suggested action: Need to determine if this is expected behavior change or regression

Build System

#8885 - Regression in packaging: JNI bindings use wrong architecture in macOS cross-compilation

• Severity: High (blocking Java users on macOS)
• Component: Build system (mk_make.py)
• Impact: JNI bindings built for wrong architecture (arm64 instead of x86_64)
• Affected versions: 4.15.5 through 4.16.0
• Status: Active discussion, PR Fix arm64 libz3.dylib being packaged in x86_64 macOS wheels #8696 fixed main lib but missed JNI
• Relevant files: scripts/mk_make.py, JNI build configuration

---

Performance Regression Cluster

Z3 has accumulated 11 open performance regression issues across multiple versions:

String Solver Regressions

• String solver performance regression in 4.16 #8893 (4.16) - 25x slowdown on regex benchmarks
• Performance regression issue on string #6668 - General string performance regression
• Performance regression from 4.8.7.0 to 4.8.8.0 and all later versions. #5763 (4.8.8.0) - String solver regression
• Performance regression issue on QF-S logic #4770 - QF-S logic performance

Arithmetic Solver Regressions

• Performance regression from arith.solver=2 to arith.solver=6 #5614, [Consolidated] Performance regressions of arith.solver=6 vs. arith.solver=2 #4934 - arith.solver=6 regressions (previously tracked as duplicate cluster)

Other

• Massive performance regression in 4.15.4.0 apparently due to lock contention #8019 (4.15.4.0) - Lock contention causing performance issues
• Infinite loop/performance regression in Z3 4.15.3 #7991 (4.15.3) - Infinite loop/performance regression

Recommendation: Many of these regressions span multiple versions and theories. A systematic performance testing infrastructure would help catch regressions before release.

---

Duplicate / Mergeable Issues

Proof Mode + String Theory Soundness Bugs

Canonical issue: #8194
Related issues: #8185

Both issues describe incorrect models when :produce-proofs true is enabled along with String theory constraints. The workaround for both is to disable proof production. These should be tracked together as they likely share the same root cause in the proof generation infrastructure's interaction with theory_str.

Relevant code areas:

• src/smt/theory_str.cpp - String theory solver
• src/smt/smt_context.cpp - Proof generation coordination
• src/smt/theory_proof.cpp - Theory-specific proof generation

---

Issues Needing More Information

Issue	Title	Missing Information
#8890	z3-solver==4.16.0 finds less solutions	Need to clarify: Is this completeness regression or expected behavior? Reproducible SMT-LIB2 test case would help
#8841	behavior difference windows/linux	Needs reproducible test case and specification of expected behavior

---

Notable Long-Standing Issues

Issues Over 1 Year Old With Activity

Issue	Title	Age (days)	Last Activity	Notes
#6512	NodeJs process won't stop after using 'solve'	~1170	Feb 24, 2026	WASM threading issue; previously commented
#5920	Issues with models of Horn clauses	~1750	May 2022	Meta-issue tracking Horn clause model problems

---

Recommended Actions

Immediate Priority

1. Investigate proof mode + string bug cluster (Incorrect model in mixed FP/Real logic triggered by extra string constraint #8185, Incorrect model with :produce-proofs true involving String Theory and UF #8194) - systematic soundness issue
2. Bisect string performance regression (String solver performance regression in 4.16 #8893) - blocking users on 4.16
3. Fix JNI cross-compilation (Regression in packaging: JNI bindings use wrong architecture in macOS cross-compilation (arm64 to x64) #8885) - blocking Java users on macOS

Short Term

4. Review and triage stale soundness bugs (Incorrect model for QF_SLIA #7841, Incorrect model #7842, Wrong verdict with (set-option :sat.smt true) #7990, Spacer unsoundness regarding CHCs involving inductively defined datatypes #7611)
5. Clarify z3-solver==4.16.0 finds less solutions for BNH problem than z3-solver==4.8.12 #8890 behavior change (completeness vs. expected)
6. Establish performance regression testing to catch issues before release

Long Term

7. Address accumulation of performance regression issues across versions
8. Consider closing or consolidating duplicate performance regression reports

---

Automated by Issue Backlog Processor - Next run: March 11, 2026

AI generated by Issue Backlog Processor

• expires on Mar 16, 2026, 12:18 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-03-16T00:18:53.697Z.

Closed by Workflow