Stabilize ellipse covariance symmetrization

[FlorianPfaff]

Summary

• compute the symmetric part as the sum of two half-scaled operands
• avoid overflow when finite covariance entries are already near the floating-point limit
• add regressions for direct symmetrization and PSD covariance projection under strict NumPy error handling

Bug

tracking.ellipse_geometry.symmetrize() evaluated matrix + matrix.T before multiplying by 0.5.

For an already symmetric finite covariance such as diag([1e308, 2e307]), the diagonal entries are doubled first. That intermediate exceeds the float64 range, so the helper raises FloatingPointError under strict NumPy settings or returns infinities under ordinary settings, even though the requested symmetric result is exactly the original finite matrix.

The failure propagates into project_symmetric_covariance() and the shared ellipse canonicalization paths that rely on this helper.

Fix

Use the algebraically equivalent form

0.5 * matrix + 0.5 * matrix.T

Each finite operand is scaled before addition, so the representable average remains finite. Ordinary-scale behavior and backend-neutral operations are unchanged.

Validation

• reproduced the old overflow encountered in add failure with diag([1e308, 2e307])
• verified the scale-first expression preserves the matrix exactly under numpy.errstate(over="raise", invalid="raise")
• added focused tests for both symmetrize() and project_symmetric_covariance()
• branch comparison against current main: 4 commits ahead, 0 behind; final diff is limited to 2 files, 29 additions, and 1 deletion

Repository CI is the authoritative full backend, lint, packaging, documentation, dependency, security, and CodeQL validation.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	17.81s
✅ JSON	prettier	7	0	0	0	0.96s
✅ JSON	v8r	7		0	0	3.54s
✅ MARKDOWN	markdownlint	68	0	0	0	1.39s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	0.35s
✅ PYTHON	black	1859	84	0	0	68.74s
✅ PYTHON	isort	1859	136	0	0	2.04s
✅ REPOSITORY	betterleaks	yes		no	no	2.1s
✅ REPOSITORY	checkov	yes		no	no	53.57s
✅ REPOSITORY	gitleaks	yes		no	no	14.29s
✅ REPOSITORY	git_diff	yes		no	no	0.17s
✅ REPOSITORY	secretlint	yes		no	no	58.08s
✅ REPOSITORY	syft	yes		no	no	3.99s
✅ REPOSITORY	trivy-sbom	yes		no	no	4.74s
✅ REPOSITORY	trufflehog	yes		no	no	20.14s
✅ YAML	prettier	11	0	0	0	0.58s
✅ YAML	v8r	11		0	0	8.73s
✅ YAML	yamllint	11		0	0	0.38s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

• Documentation: Custom Flavors
• Command: npx mega-linter-runner@9.6.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_BETTERLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

Show us your support by starring ⭐ the repository

[FlorianPfaff]

Closing as superseded by #5178. Both PRs make the identical production change in tracking/ellipse_geometry.py (0.5 * matrix + 0.5 * matrix.T); #5178 is the newer current-main branch with focused overflow regression coverage.