v0.2.0

Other

• read the fixed patterns before paying for a whole grid
• imgproc, image: primitives survive absurd sizes, radii and non-finite geometry
• sampler believes a rune only if the whole 11x11 grid is that rune
• sampler checks the solid edge columns before sampling a grid
• bounded fiducial flood; aztec, microqr, rmqr: stop flooding the page
• imgproc, aztec: bound finder/bullseye hit clustering on textures
• analyse the symbol under a mask of its own quad
• bars at 45 degrees are not finder patterns
• noise- and shift-tolerant hint matching; docs
• live path: regression tests; UPC-E vote floor; Aztec blank-region reads
• pipeline, detect: read stacked codes from linear regions; grow matrix boxes
• demo, wasm, scan1d: reliable live decode loop; axis refinement
• qr, appclip: stop paying for false hypotheses on code-free frames
• try the largest few dark components as the symbol
• rotation-normalised box fill; liveeval diagnostics
• orientation-aware locator with local binarization
• expand macro codewords on decode; canonical mask for no-op forcing
• corner-forced masks and Reed-Solomon error correction
• fix decoder overflow panic and encoder silent truncation
• reject out-of-range metadata instead of panicking
• make numeral and byte modes conform to AIMD014
• fix 144x144 Reed-Solomon block interleave
• reject unsupported layer counts; add fuzz, capacity and RS tests
• decode latched Punct mode and FLG(n) escapes
• fix orientation marks to the ISO/IEC 24778 pattern
• exhaustive Text sub-mode transition round-trip test
• no-panic fuzz for Code 49 and Codablock F payload reconstruction
• decode Numeric Encodation
• a single-row Expanded Stacked symbol is the linear symbol
• code16k, codablockf: honour FNC4 extended ASCII when decoding
• use the Code 16K pattern 211133 for symbol value 106
• implement Expanded encodation methods 3-14
• reject Omnidirectional/Limited values beyond the GTIN range
• enforce the 928-codeword symbol limit when choosing a geometry
• deterministic no-panic fuzz for the stacked and DataBar decoders
• decode the 913 byte shift inside Text compaction
• reject a truncated Expanded method-1 header instead of panicking
• decode modes 5/6 as set C with implied Shift B
• scan1d, pipeline: scan along any axis; band-averaged profiles
• treat M1's EC codewords as detection-only
• reject undefined mode indicators instead of truncating
• qr, rmqr: reject ECI assignment numbers above 999999
• cross-check EC-block, alignment and remainder tables in a test
• make mask penalty rule 4 symmetric about 50% dark
• report content errors instead of retrying every format
• support ECI; report FNC1 symbols as unsupported, not empty
• deterministic no-panic fuzzing for QR, Micro QR, rMQR and Han Xin
• qr, microqr, rmqr: reject Kanji trail bytes that alias another character
• microqr, rmqr: keep empty segments; reject invalid numeric groups
• reject numeric groups that are not valid decimal numbers
• reject out-of-range numeric groups instead of panicking
• pad M1/M3 with 0xEC/0x11 codewords per ISO/IEC 18004
• scan1d, pipeline: segment scanlines into spans; accept 1D reads by consensus
• deterministic no-panic / re-encodable sweep over linear and postal decoders
• validate the termination bar
• apply FNC4 (extended ASCII) when reconstructing the payload
• decode rows cropped to their last bar
• itf, codabar, twoof5: bound wide elements and require narrow 2-of-5 spaces
• code39, code11: decode symbols printed at up to a 3:1 wide:narrow ratio
• use the standard two-module gap for one-track symbols
• reject POSTNET/PLANET/RM4SCC layouts with no data character
• reject an unparseable customer information field
• only accept canonical intermediate fields
• reject out-of-range post code values instead of panicking
• reject out-of-range codewords instead of panicking
• reject non-BCD bit groups and payload-less rows on decode
• check the all-codes decode-only and scan-only slices too
• reject surplus positional arguments
• scale resamples the edge instead of whitening the border
• decode_payload requires the color stream
• refuse PNGs that declare gigantic dimensions
• do not panic when stdout is closed early
• align CLI list text, README 1D note and doc comments with the code
• MatrixBuf::new reports overflowing dimensions as a capacity error
• composite PNG alpha over white before scanning
• reject absurd --scale instead of panicking or exhausting memory
• strict option parsing (flags, --, unknown options)
• encode hosts of the form appclip.
• keep leading zeros of numeric URL components
• reject non-ASCII near the scheme prefix instead of panicking
• liveeval (end-to-end live pipeline evaluation) and scantime
• gate decoding on ring structure and the template byte
• require quiet zones in the width-ratio reader; search every offset

Fixed

• Live detection. On camera-like scenes the locate → crop → decode loop read about
  half the codes it was shown and reported dozens of values that were not there, at
  ~1 s per decode pass. It now reads all of the 1D and 95% of the 2D codes on the same scenes
  with no wrong values, in tens of milliseconds (examples/liveeval.rs,
  tests/live_pipeline.rs):
  • scan1d treated a whole scan line as one barcode and required both crop edges to
    be light; it now thresholds locally, cuts each line into quiet-zone-delimited spans,
    fits module width and bar/space bias jointly, and samples band-averaged profiles at
    any angle.
  • pipeline::scan_1d believed any single decode although half its decoders have no
    check character; readings now need cross-scanline consensus, quiet zones, a
    plausible length, and must not be a fragment of a longer reading.
  • The EAN/UPC width reader ignored quiet zones, reading a UPC-E out of the left half
    of any EAN-13 and out of other symbologies and text.
  • detect::locate only recognised barcodes within ~15° of the frame axes, was blinded
    by uneven lighting (global Otsu), let a single false finder hit turn a text blob into
    a top-ranked "QR" that swallowed the barcode inside it, and gave decoders no
    orientation. It is now gradient-orientation based with local binarization, and
    reports each linear region's reading axis and an oriented box.
  • The App Clip scanner decoded ordinary barcodes and text into fluent-looking URLs
    (its payload RS is far too weak for thousands of hypotheses per frame); it now
    requires real ring structure and the template byte. Aztec no longer "decodes" a
    blank patch (all-zero codewords are a valid RS codeword).
  • The Data Matrix sampler assumed the largest dark blob in the image was the symbol
    and let any ink near it pull the corner and edge fits; it read a third of the
    symbols it was shown in a scene, now all of them.
  • QR and App Clip scanning no longer spend hundreds of milliseconds on code-free
    frames.
  • FrameDetector matched Hints by exact fingerprint equality, which sensor noise
    and hand shake defeat on every frame; it now matches by position plus a tolerant
    fingerprint.
  • Web demo: a wasm trap or failed fetch left a decode worker "busy" forever, silently
    disabling decoding for the session; the crop batch was truncated before barcodes
    were prioritised, so busy scenes never decoded theirs.
• Encoders now conform where they did not (output unreadable by other tools): Aztec
  orientation marks, Data Matrix 144×144 block interleave, Micro QR M1/M3 padding,
  Code 16K symbol value 106, one-track Pharmacode gap, Grid Matrix numeral/byte modes,
  DotCode corner-forced masks.
• Decoders now read what other encoders produce: DataBar Expanded methods 3–14, PDF417
  byte shift 913, Code 128 / Code 16K / Codablock F FNC4, Code 49 numeric mode, Code 16K
  modes 5/6, rMQR ECI, Aztec latched punctuation and FLG(n), DotCode macros and error
  correction, Code 39 / Code 11 at 3:1 print ratio, Telepen cropped to its last bar.
• Panics on hostile input: IMb, Mailmark, Han Xin, DataBar Expanded, DotCode and MaxiCode
  decoders; App Clip URLs with non-ASCII near the scheme; MatrixBuf::new overflow; CLI
  --scale, giant PNG headers, closed stdout.
• Many decoders accepted malformed or non-canonical input (out-of-range numeric groups,
  aliased Kanji, empty payloads, over-wide elements, unvalidated termination bars, Micro
  QR M1 "error correction") that either mis-decoded or could not re-encode.
• CLI: option parsing (flags swallowing data, unknown options ignored, surplus
  arguments), alpha-channel PNGs decoding as solid black.

Added

• pipeline::scan_1d_at, scan_stacked_at, scan_linear_at: decode a located linear
  region along its known reading axis (1D readers, then the PDF417 family).
• scan1d::scan_spans, scan1d::refine_axis, ScanOptions::around;
  imgproc::orient::gradient_angle_peaks; codes::ean::decode_edges_within;
  detect::fingerprint_distance.
• Location::rotation and an oriented outline on linear detect::locate candidates.

Changed

• LocateOptions::anisotropy is now a rotation-invariant gradient coherence (default
  0.7) and edge_density a fraction of edge pixels; ScanOptions::angles_deg accepts
  any angle.
• One-track Pharmacode renders with the standard two-module gap (max_modules 78).