Make CI green and unblock the 0.1.9 release#4
Merged
Conversation
Packaging: - Six tracked examples (webcam_viewer, webcam_tuner, webcam_selector, render_braille, render_cuda_dojo, generate_watermark) had no [[example]] entry, so cargo built them with default features and they failed to compile. They ship inside the crate, so `cargo build --examples` was broken for consumers. Declare their required-features. - Gate tests/temporal_coherence_tests.rs on the `image` feature; it was the sole reason bare `cargo test` (what CI runs) failed to compile. - Exclude tests/visual_regression.rs from the package. tests/visual/ is already excluded, so the tarball shipped a test whose `mod visual;` could not resolve. MSRV: - wireframe.rs used `10.0_f32.to_radians()` in a const, but f32::to_radians is not const-stable until 1.85 while rust-version is 1.70. Replace with PI / 18.0 (bitwise-identical f32, no behavior change). Lint policy: - CI runs `cargo clippy --all-features -- -D warnings`, which promoted the advisory pedantic/nursery lints to blocking: 3,975 errors. Curate an allow-list so `clippy::all = deny` remains the real bar. Notably unnecessary_literal_bound (1,971) cannot be fixed without a breaking change to the public ProgressStyle trait, and suboptimal_flops (1,406) would fuse multiply-adds, altering float rounding and rendered output. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mechanical, semantics-preserving rewrites clippy marked MachineApplicable: unnecessary_min_or_max, bool_to_int_with_if, map_unwrap_or, int_plus_one, range_plus_one, collapsible_if, redundant_clone, and similar. Full suite (726 lib tests + visual regression) passes unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Vetted before applying: 48 sites are integer (usize) with correctly-ordered literal bounds, so clamp() cannot panic and NaN is not representable. The 2 float sites (floweroflife line_frac/arm_frac) clamp to literal 0.0..1.0 and their input derives from ctx.eased = ease(kind, progress), which is finite -- so the max().min() vs clamp() NaN divergence is unreachable. Clippy marks these MaybeIncorrect precisely because of that NaN case; it does not apply here. Full suite green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Deletes the vestigial 'outer: for pass in 0..2 loop in inject_electron. The body always terminated on the first iteration, so the TAIL branch was dead -- correctly so, since step() implements WireWorld (HEAD -> TAIL -> CONDUCTOR) and the automaton produces tails itself. The injector only ever needs to place a HEAD. Verified byte-identical rendering: all 14 cellular styles x 21 progress x 17 time x 4 grid sizes hash the same before and after. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CI runs RUSTDOCFLAGS="-D warnings" cargo doc, so these blocked the release. - medieval.rs: unnecessary parens around a function argument. - temporal.rs: [`process_dots`] is an associated item; needs Self:: to resolve. - image/mod.rs: ambient_seed's doc was not merely a broken link, it was wrong. It told callers to "pass None", but the method takes u64, not Option, and the unpin_ambient_seed function it linked does not exist anywhere in the crate. Replaced with the actual behavior: leave the seed unset for clock seeding. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…theory - needless_range_loop / explicit_counter_loop: real iterator rewrites; every rewritten bound was checked so no OOB-panic path became a silent truncation. - type_complexity: private MarkedTri alias in penrose. - too_many_arguments: bundle the five camera params into a private View struct in platonic and topology, using field-init shorthand so the same-typed ax/ay and cx/cy cannot be silently transposed. - approx_constant: platonic CUBOCTA_SCALE now uses f32::consts::SQRT_2. NOTE: this is a real 1-ULP change, not a no-op -- the old literal 1.414_213_6 was one ULP above SQRT_2. Verified output-neutral over 96M coordinate comparisons and a full render sweep. Rendering verified byte-identical: ~30k frames (goldenratio/platonic/topology) and 9,702 bars (penrose/numbertheory) hash the same before and after. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- needless_range_loop: iterate the fixed-size col_xs/row_ys arrays directly; buffer.iter_mut().enumerate() in the raytracer (buffer is built as vec![vec![0.0; width]; height], so the index sequence is identical). - too_many_arguments on render_edges_with_orientation: narrowly-scoped allow. It is a pub fn re-exported from raytracer/mod.rs, so folding the params into a struct would be a breaking public API change. - unnecessary_cast in geometry: n_max/n_total are only ever used as f32, so declare them f32 and drop the casts. Clippy's own suggestion here is broken (it retypes the binding but leaves the integer literal -> E0308) and was not applied. 400 and 361 are exactly representable in f32, so bit-identical. The separate n_total = 180 (cardioid) is genuinely usize and was left alone. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
d22cb8f used `git add -A ... examples/`, which pulled in four uncommitted scratch examples (mikey, raphe, truck, gif_to_pack) and six loose images under src/image/. None of them were ever meant to be in the repo: the examples do not compile (they reference dotmax::raytracer/image without required-features) and examples/ ships inside the published crate. Removed from the index only -- the files stay on disk, untracked, exactly as they were before this session. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…lle, obj_loader - needless_range_loop: iterator rewrites; each container's length was proven equal to the old range bound before rewriting. - fruits: collapsed an if/else whose branches were identical ((ry+ny)/2 both ways). - obj_loader: strip_prefix instead of starts_with + manual slice. - braille: fixed a broken intra-doc link ([0,1] was parsed as a link) that was failing the RUSTDOCFLAGS=-D warnings doc gate. Known deviation: raytracer/braille's .take().skip() rewrite silently visits fewer cells on a RAGGED buffer where indexing would previously have panicked. width is derived from buffer[0].len(), so every in-repo caller passes a rectangular buffer and output is bit-identical there. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- webcam: manual_let_else + match_wildcard_for_single_variants collapsed into one let-else (ffmpeg Context has exactly two variants, so the old _ arm and the new else-branch cover the same case); two needless_continue removed (both were the final statement on their path, so removal is a no-op). - dotmax_braille: explicit_iter_loop. - progress/mod: doc-only paragraph split for too_long_first_doc_paragraph. No code changed in the public progress API. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CI runs clippy with -D warnings, so these pedantic/nursery findings were blocking the release gate. - no_effect_underscore_binding: three dead let bindings with pure RHS, deleted. - comparison_chain: if/else-if ladders -> match on Ordering (tech, gameboy). - branches_sharing_code: hoisted the shared draw call in lasers and medieval. Draw order is preserved in both -- each branch already performed the shared draw first. - gadgets: inverted an if-not-else; NaN behavior unchanged since Range::contains returns false either way. Merged a match arm the wildcard already covered. - glitch: bool_to_int_with_if -> i32::from. - cosmos: needless_for_each -> for loop, same order, same in-place swap. Two narrowly-scoped allows, both deliberate: - chaos: #[inline(always)] on a per-dot hash is a real perf choice. - food: a while-loop on a float that provably runs at most twice (angle comes from atan2). A closed-form rewrite would change float rounding and therefore the rasterized wedge boundary. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Last missing_errors_doc site. The two real bail-outs are an unopenable file and an OBJ containing no geometry. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Clears the last of ci.yml's 'cargo clippy --all-targets --all-features -D warnings'. - examples: doc_lazy_continuation, needless_range_loop (real iterator rewrites), format_push_string -> write!/writeln!, match_same_arms, verbose_bit_mask, manual_rotate, bool_comparison, unchecked_time_subtraction -> saturating_sub. - benches: deprecated criterion::black_box -> std::hint::black_box; removed redundant clones in setup only (nothing inside b.iter changed, so no benchmark measures anything different). - prelude tests: six narrowly-scoped allows. The bindings and the Result-returning helpers ARE the assertions -- they prove the re-exported names resolve and type-check. Having no runtime effect is the point of the test. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
They were swept back in by another over-broad 'git add -A examples/'. Adding them to .gitignore so a future bulk add cannot pull them into the crate. Files remain on disk, untracked. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- easing tests: float_cmp allowed at the test module with a reason. Exact float equality is the property under test (curves must hit 0.0/1.0 exactly, lerp must be exact at the midpoint) and every compared value is exactly representable, so an epsilon comparison would weaken the assertions, not fix them. - quick.rs test: name the Animated variant instead of a wildcard, so a future MediaContent variant cannot silently fall into the panic arm. - image_rendering_tests: #[ignore] now carries its reason string. - property_tests: dropped a redundant clone; visual_regression / media_integration: dropped unused imports. - tests/visual/mod.rs: allow(dead_code) with a reason -- the baseline helpers are UPDATE_BASELINES=1 maintenance tooling, deliberately not called by the suite. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
test_extreme_wide/_tall_aspect_ratio and test_very_large_square_image loaded viper_ultra_wide.png, viper_ultra_tall.png and viper_4k.png -- ~31 MB of fixtures that are gitignored (.gitignore:51-54). So they only passed on a machine that already had the files. On a fresh clone, and in CI under 'cargo test --all-features' (which is exactly what release.yml runs), they failed on a missing file. They have never actually run in CI. Synthesize the pixels in memory instead: same dimensions, same pipeline, no assets. The per-pixel pattern varies so auto_threshold sees a real distribution. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
normalize_intensity must clamp to exactly 0.0/1.0 (including NaN and the infinities) and be the identity in range -- exact equality IS the assertion. Only visible under --all-targets, which builds the test cfg; the release gate does not. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
scheme_builder, animation::timing and raytracer::lighting tests assert exact stored/reset values (a 0.5 colour stop, 0.0 fps after reset, exactly-zero diffuse shading for a perpendicular light). Exact equality is the assertion. Only visible under --all-targets, which builds the test cfg. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
zone_stream_canvas_1/2/3.rs were exact copies of zone_stream.rs (same md5, 3,459 lines each). They carried no unique content, shipped inside the published crate, and accounted for three quarters of the example lint surface. -10,377 lines. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
CI has been red on
mainsince May, and the release pipeline could not publish. This makes both green.Why the release was blocked
release.ymlpublishes on av*tag, but only after avalidatejob (cargo test --all-features,cargo clippy --all-features -- -D warnings,cargo doc -D warnings,cargo test --doc). That job was failing, so tagging v0.1.9 would have published nothing.The clippy gate reported 3,975 errors. The cause was policy, not rot:
Cargo.tomlsetspedantic/nurserytowarn(advisory), but CI's-D warningspromoted them to blocking. Two advisory lints were 85% of the total, and both were wrong to "fix":unnecessary_literal_bound(1,971) — wantsProgressStyle::{name,theme,describe}to return&'static str. Those are public trait methods; narrowing them breaks every downstream implementor.suboptimal_flops(1,406) — wants fusedmul_add, which changes float rounding and so can change rendered output. Its siblingimprecise_flopswas already allowed.Both are now allowed with justification.
clippy::all = denyremains the real bar, and every one of those errors is fixed rather than suppressed.Real defects fixed (not just lints)
webcam_viewer,webcam_tuner,webcam_selector,render_braille,render_cuda_dojo,generate_watermarkhad no[[example]]entry, so cargo built them with default features and they failed to compile — inside the published crate.wireframe.rsusedf32::to_radians()in aconst, which is not const-stable until 1.85, whilerust-version = "1.70". Replaced withPI / 18.0(bitwise-identical f32).ImageRenderer::ambient_seedtold callers to "passNone" — but it takesu64, and theunpin_ambient_seedit linked does not exist.tests/visual_regression.rswithout its excludedtests/visual/module).Rendering is provably unchanged
All 664 styles hashed across 6 grid sizes × 21 progress × 17 time values (~1.4M frames), before vs after: byte-for-byte identical.
Known remaining
Cargo Deny/Security Auditmay still fail oncore2 v0.4.0, which is yanked and arrives transitively viaimage → ravif → rav1e → bitstream-io. The requirement is^0.4and the only published 0.4.x is the yanked one — there is nothing to move to. Not a release gate.MSRV Check(cargo check --all-featureson 1.70) is untested locally.Also note:
ci.ymlusesbranches: ["*"], which does not match branch names containing/— so push-CI never runs onfeature/...-style branches.🤖 Generated with Claude Code