Reject overflowing PyTorch uniform ranges

[FlorianPfaff]

Summary

• reject finite PyTorch random.uniform bounds when their promoted floating-point span is not representable
• compute the span after promoting integer/low-precision bounds to a safe arithmetic dtype
• preserve the existing random-stream dtype and scalar/array broadcasting behavior
• add a NumPy/PyTorch regression for the maximum-finite symmetric range

Bug

The PyTorch backend validated low and high independently as finite, then sampled with:

(high - low) * rand + low

For finite endpoints such as -np.finfo(np.float64).max and np.finfo(np.float64).max, high - low overflows to infinity. The sampler therefore returned non-finite values even though both inputs passed validation. NumPy rejects the same request with OverflowError.

Fix

Promote both validated bounds to the effective floating arithmetic dtype before subtraction, validate that the resulting span is finite, and raise OverflowError("high - low range exceeds valid bounds") when it is not. The random draw still uses the prior dtype path, so ordinary seeded sampling is unchanged.

Validation

• reproduced the pre-fix PyTorch result as all-inf samples for the maximum-finite symmetric range
• focused smoke test verifies the patched path raises OverflowError
• focused smoke test verifies ordinary array-valued bounds still return finite in-range samples
• seeded smoke test verifies the existing float64-bound RNG stream is preserved exactly
• python -m py_compile passes for both modified files
• branch is 3 commits ahead of and 0 behind main

The full repository test matrix is delegated to GitHub Actions.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	20.19s
✅ JSON	prettier	7	0	0	0	0.99s
✅ JSON	v8r	7		0	0	4.0s
✅ MARKDOWN	markdownlint	68	0	0	0	1.83s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	1.07s
✅ PYTHON	black	1489	182	0	0	85.95s
✅ PYTHON	isort	1489	327	0	0	2.55s
✅ REPOSITORY	betterleaks	yes		no	no	2.06s
✅ REPOSITORY	checkov	yes		no	no	53.88s
✅ REPOSITORY	gitleaks	yes		no	no	14.48s
✅ REPOSITORY	git_diff	yes		no	no	0.24s
✅ REPOSITORY	secretlint	yes		no	no	52.82s
✅ REPOSITORY	syft	yes		no	no	5.67s
✅ REPOSITORY	trivy-sbom	yes		no	no	6.22s
✅ REPOSITORY	trufflehog	yes		no	no	30.33s
✅ YAML	prettier	11	0	0	0	0.7s
✅ YAML	v8r	11		0	0	11.78s
✅ YAML	yamllint	11		0	0	0.6s

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