fix(imgproc): CLAHE edge-pixel extrapolation and calc_back_project range validation

[kalwalt]

Summary

Fixes two real bugs found by Qodo's review on #121, verified against the actual
code before fixing (a third finding on that review — missing FromPrimitive
bound — is a false positive, the same recurring rule-2966076 pattern already
identified on #110: the functions only ever read pixels via
ToPrimitive::to_f32(), never construct T from a primitive).

CLAHE edge-pixel extrapolation. Interpolation clamped the tile index
before computing the blend fraction. At y=0 (tile_rows=4), tyf=-0.5 →
floor=-1 → clamped to 0 → ya = tyf - 0 = -0.5, ya1 = 1.5. Those aren't
interpolation weights (they don't lie in [0,1]) — they extrapolate beyond
the edge tile's LUT instead of blending within it, on the top row and left
column of both the u8 and u16 paths. Fixed by deriving the fraction from
the unclamped floor (always in [0,1) by construction) and clamping only
the LUT-lookup indices separately. Extracted the duplicated u8/u16 weight
math into a single tile_interp_weights() helper.

calc_back_project range validation. It checked only ranges.len() == dims, never the contents — unlike calc_hist, which already validates
Uniform(lo, hi) satisfies lo < hi (rejecting NaN) and NonUniform
boundaries have the right length and strict ordering. Extracted calc_hist's
validation into a shared validate_hist_ranges() helper so the two functions
can't drift apart again.

Test plan

• New unit test on tile_interp_weights checking weights stay in [0,1]
  and sum to 1 across the actual coordinate range (several tile counts),
  including the exact y=0 case from the bug report. Confirmed the old
  formula fails this assertion (ya=-0.5).
• New tests for calc_back_project rejecting NaN/inverted uniform ranges
  and malformed non-uniform boundaries.
• cargo fmt --check / clippy -D warnings clean under --all-features
  and --no-default-features.
• cargo test --workspace: 345 lib tests + 40 doc-tests passing under
  both default (parallel) and --no-default-features --features std
  (sequential) configs, identical results.

🤖 Generated with Claude Code

[qodo-code-review]

PR Summary by Qodo

Fix CLAHE edge interpolation and back-project range validation

🐞 Bug fix 🧪 Tests 🕐 20-40 Minutes

AI Description

• Keeps CLAHE edge interpolation weights bounded for both u8 and u16 paths.
• Shares strict histogram range validation across histogram and back-projection operations.
• Adds regressions for malformed ranges and edge-tile interpolation invariants.

Diagram

graph TD
  Input["Image Input"] --> Route{"Operation"} --> Hist["Histogram Path"] --> Validate["Range Validation"] --> Back["Back Projection"]
  Route --> CLAHE["CLAHE Path"] --> Weights["Tile Weights"] --> Blend["u8/u16 Blend"]

Loading

High-Level Assessment

The shared-helper approach is appropriate: it removes duplicated interpolation math across pixel depths and prevents histogram range validation from diverging between related APIs. Keeping separate implementations would increase regression risk without providing meaningful flexibility.

Files changed (2) +198 / -76

Bug fix (1) +100 / -76

histogram.rsCorrect CLAHE weights and centralize histogram range validation +100/-76

Correct CLAHE weights and centralize histogram range validation

• Extracts shared histogram range validation and applies it to both 'calc_hist' and 'calc_back_project', rejecting zero dimensions, invalid uniform ranges, and malformed non-uniform boundaries. Adds a shared CLAHE interpolation helper that derives blend fractions before clamping lookup indices, preventing edge extrapolation in both u8 and u16 paths.

src/imgproc/histogram.rs

Tests (1) +98 / -0

tests.rsAdd regressions for range validation and CLAHE edge weights +98/-0

Add regressions for range validation and CLAHE edge weights

• Verifies back projection rejects NaN, inverted, incorrectly sized, and unordered ranges. Exercises CLAHE interpolation across multiple tile counts and confirms indices remain valid while weights stay bounded and sum to one, including the reported edge coordinate.

src/imgproc/tests.rs

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📜 Skill insights (0)

1. Range validation can overflow ✓ Resolved 🐞 Bug ☼ Reliability

Description

For hist_size[d] == usize::MAX, validate_hist_ranges evaluates sz + 1, causing an overflow
panic before calc_back_project can return an input error. With overflow checks disabled and empty
boundaries, the wrapped expected length also reaches boundaries.len() - 1, which underflows and
ultimately panics.

Code

src/imgproc/histogram.rs[210]

+                if boundaries.len() != sz + 1 {

Evidence

The new validator directly computes sz + 1 without an overflow guard and then subtracts one from
the boundary length. calc_back_project now invokes this helper before computing total_bins or
checking the supplied histogram shape, so malformed input reaches the panic first.

src/imgproc/histogram.rs[209-223]
src/imgproc/histogram.rs[490-499]
src/imgproc/histogram.rs[519-535]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
`validate_hist_ranges` performs unchecked `sz + 1` arithmetic for non-uniform ranges. A `hist_size` entry of `usize::MAX` therefore panics rather than producing the function's expected `InvalidInput` result.

## Issue Context
The new `calc_back_project` validation call exposes this panic before its histogram-shape check. Compute the expected boundary count with `checked_add`, return a tagged `InvalidInput` error on overflow, and reuse the checked value in both the comparison and error message. Add a regression test using `usize::MAX` and an empty non-uniform boundary list.

## Fix Focus Areas
- src/imgproc/histogram.rs[209-223]
- src/imgproc/tests.rs[1591-1617]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

Context sources

✅ Compliance rules (platform): 27 rules

✅ Skills: license-header-adder

Review mode: ⚖️ Balanced: This changes image-processing interpolation behavior and shared histogram/back-projection validation across multiple code paths, so it carries meaningful correctness risk but is not broad or defect-dense enough to warrant redundant extended review.

Tip of the day

💡 Did you know, you can turn on the rule miner and Qodo learns your standards from review history

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

[kalwalt]

Confirmed real — reproduced it directly with std::panic::catch_unwind before fixing: hist_size[d] == usize::MAX panics with attempt to add with overflow at the sz + 1 in validate_hist_ranges, exactly as reported. The empty-boundaries underflow you also flagged is a consequence of the same root cause (only reachable once sz + 1 has already wrapped to 0 in a release build, matching an empty boundaries.len()), so guarding the addition closes both paths at once.

Fixed in f5da497: sz.checked_add(1) now returns a proper InvalidInput error on overflow instead of panicking, using cv_err! to stay consistent with the rest of the module's error handling. Added the regression test you suggested (hist_size = usize::MAX with empty non-uniform boundaries) — confirms calc_back_project now returns Err instead of panicking.

Verified: fmt/clippy clean under --all-features and --no-default-features; 346 lib tests + 40 doc-tests passing under both the default (parallel) and sequential configs.