Issues

@expect tol means absolute when want == 0 and relative otherwise: 'tol = 0.5' at want = -2 accepted a deviation of 0.91

QAtlasHub/Pinax.jl#160 · opened

on Oct 3, 2026

@caption after @table overwrites the caption of the figure above it, and the table gets none

QAtlasHub/Pinax.jl#159 · opened

on Oct 3, 2026

register_theme! accepts a theme that cannot render, and hasmethod cannot be the guard that rejects it

QAtlasHub/Pinax.jl#142 · opened

on Sep 7, 2026

serve: every failure inside _serve_handle is dropped with no record, so a server bug is wire-identical to a client disconnect

QAtlasHub/Pinax.jl#141 · opened

on Sep 7, 2026

Four smaller defects in the test-report path: magic-triple indicator, in-place Check mutation, silent depth flattening, last-writer-wins sweep band

QAtlasHub/Pinax.jl#130 · opened

on Sep 1, 2026

@table params= is accepted and stored but emitted by neither face

QAtlasHub/Pinax.jl#128 · opened

on Sep 1, 2026

One public name has no docstring (Theme); ~22 more are absent from CLAUDE.md / README / docs

QAtlasHub/Pinax.jl#127 · opened

on Sep 1, 2026

· v0.1.0

2

The shard merge drops a root testset desc, tables, codes and panels after dump_test_report correctly serialised them

QAtlasHub/Pinax.jl#126 · opened

on Sep 1, 2026

_FILE_AST is an unsynchronised process-global mutated inside Test.record, so a threaded testset races only when the report is on

QAtlasHub/Pinax.jl#125 · opened

on Sep 1, 2026

@pinaxignore is skipped by the per-file table but counted in the page header, so the two disagree with no note

QAtlasHub/Pinax.jl#124 · opened

on Sep 1, 2026

Check, code-block and page ids move between runs, so agent.json ids and gallery deep links are not stable

QAtlasHub/Pinax.jl#123 · opened

on Sep 1, 2026

The documented dogfooding does not run: no workflow produces pinax-report-delegation, and runtests.jl never installs the capture

QAtlasHub/Pinax.jl#120 · opened

on Sep 1, 2026

· v0.1.0

1

The test capture reads Test.Pass.data and Test.Fail.data, so a Julia upgrade degrades every check to a 1/0 indicator silently

QAtlasHub/Pinax.jl#119 · opened

on Sep 1, 2026

A shard dump with no PINAX_TEST_MATRIX is filtered out of a matrix report without a diagnostic, and the summary counts only the survivors

QAtlasHub/Pinax.jl#118 · opened

on Sep 1, 2026

The render cache manifest is theme-blind, so rendering two themes into one out= deletes the other theme assets and leaves agent.json pointing at nothing

QAtlasHub/Pinax.jl#117 · opened

on Sep 1, 2026

The @test bridge has none of the @expect guards: delta can be the JSON string Inf, and a NaN got is silently replaced with fabricated 1/0 numbers

QAtlasHub/Pinax.jl#116 · opened

on Sep 1, 2026

The @test isapprox bridge picks the wrong tolerance: atol never governs while want is nonzero, and delta divides by |want| instead of max(|got|,|want|)

QAtlasHub/Pinax.jl#115 · opened

on Sep 1, 2026

_benchmark_verdict returns PASS for a page with zero checks, so a benchmark whose checks never ran is green in all three backends

QAtlasHub/Pinax.jl#114 · opened

on Sep 1, 2026

agent.json and agent.md drop every diagnostic, so a crashed figure is indistinguishable from an empty one -- only in the LLM face

QAtlasHub/Pinax.jl#113 · opened

on Sep 1, 2026

sweep_mean discards n and the spread it already computed: a 1-sample point and a 10-sample point are the same value, and a single NaN silently makes a 5-sample group NaN — against CONVENTIONS.md §2

QAtlasHub/Pinax.jl#110 · opened

on Aug 16, 2026

_manifest_key collapses every non-[A-Za-z0-9_-] character to '_', so 'results/fig1' and 'results_fig1' share one manifest entry — the comment claims the key is unique per figure

QAtlasHub/Pinax.jl#109 · opened

on Aug 16, 2026

serve: a suffix Range (bytes=-N) is read as bytes=0-N — 'Range: bytes=-10' returns the FIRST 11 bytes with Content-Range 0-10/1024, and a range request on an empty file drops the connection

QAtlasHub/Pinax.jl#108 · opened

on Aug 16, 2026

1

read_comments swallows a TOML parse error with a bare catch, so a corrupt comments.toml renders identically to no comments file — while add_comment on the same file refuses by name

QAtlasHub/Pinax.jl#106 · opened

on Aug 16, 2026

bib_link double-prefixes a DOI that is already a URL: doi = {https://doi.org/10.1000/xyz123} renders as https://doi.org/https://doi.org/10.1000/xyz123 and does not resolve

QAtlasHub/Pinax.jl#105 · opened

on Aug 16, 2026

cite.jl renders authors from first+last only: 'van der Vorst' becomes 'Vorst' and 'King, Jr., Martin Luther' becomes 'Martin King' — the parser supplies particle/middle/junior and the adapter discards them

QAtlasHub/Pinax.jl#104 · opened

on Aug 16, 2026

Navigation Menu

Repository navigation

Issues

@expect tol means absolute when want == 0 and relative otherwise: 'tol = 0.5' at want = -2 accepted a deviation of 0.91

@caption after @table overwrites the caption of the figure above it, and the table gets none

register_theme! accepts a theme that cannot render, and hasmethod cannot be the guard that rejects it

serve: every failure inside _serve_handle is dropped with no record, so a server bug is wire-identical to a client disconnect

Four smaller defects in the test-report path: magic-triple indicator, in-place Check mutation, silent depth flattening, last-writer-wins sweep band

@table params= is accepted and stored but emitted by neither face

One public name has no docstring (Theme); ~22 more are absent from CLAUDE.md / README / docs

The shard merge drops a root testset desc, tables, codes and panels after dump_test_report correctly serialised them

_FILE_AST is an unsynchronised process-global mutated inside Test.record, so a threaded testset races only when the report is on

@pinaxignore is skipped by the per-file table but counted in the page header, so the two disagree with no note

Check, code-block and page ids move between runs, so agent.json ids and gallery deep links are not stable

The documented dogfooding does not run: no workflow produces pinax-report-delegation, and runtests.jl never installs the capture

The test capture reads Test.Pass.data and Test.Fail.data, so a Julia upgrade degrades every check to a 1/0 indicator silently

A shard dump with no PINAX_TEST_MATRIX is filtered out of a matrix report without a diagnostic, and the summary counts only the survivors

The render cache manifest is theme-blind, so rendering two themes into one out= deletes the other theme assets and leaves agent.json pointing at nothing

The @test bridge has none of the @expect guards: delta can be the JSON string Inf, and a NaN got is silently replaced with fabricated 1/0 numbers

The @test isapprox bridge picks the wrong tolerance: atol never governs while want is nonzero, and delta divides by |want| instead of max(|got|,|want|)

_benchmark_verdict returns PASS for a page with zero checks, so a benchmark whose checks never ran is green in all three backends

agent.json and agent.md drop every diagnostic, so a crashed figure is indistinguishable from an empty one -- only in the LLM face

sweep_mean discards n and the spread it already computed: a 1-sample point and a 10-sample point are the same value, and a single NaN silently makes a 5-sample group NaN — against CONVENTIONS.md §2

_manifest_key collapses every non-[A-Za-z0-9_-] character to '_', so 'results/fig1' and 'results_fig1' share one manifest entry — the comment claims the key is unique per figure

serve: a suffix Range (bytes=-N) is read as bytes=0-N — 'Range: bytes=-10' returns the FIRST 11 bytes with Content-Range 0-10/1024, and a range request on an empty file drops the connection

read_comments swallows a TOML parse error with a bare catch, so a corrupt comments.toml renders identically to no comments file — while add_comment on the same file refuses by name

bib_link double-prefixes a DOI that is already a URL: doi = {https://doi.org/10.1000/xyz123} renders as https://doi.org/https://doi.org/10.1000/xyz123 and does not resolve

cite.jl renders authors from first+last only: 'van der Vorst' becomes 'Vorst' and 'King, Jr., Martin Luther' becomes 'Martin King' — the parser supplies particle/middle/junior and the adapter discards them