longecho 0.6.0

Added

• A durable format list, longecho.formats.FORMATS. Each entry records its
  extensions, media type (plus common aliases such as audio/x-wav),
  category, the readers that can open it (standard tool, web browser, text
  editor), and the reason it qualifies under the Library of Congress
  sustainability factors, which are now the written criteria for the list.
  DURABLE_EXTENSIONS and DURABLE_FORMAT_CATEGORIES are derived from it and
  keep working. is_durable_mimetype() answers for tools that record media
  types (any case, parameters allowed).
• New durable formats: .pdf, .epub, .eml, .mbox, .svg, .tif,
  .tiff, .wav, .flac, .mp3, .ogg, .opus, .mp4, .webm, with new
  Mail, Audio, and Video categories. A directory holding only one of these
  plus a README is now compliant.
• Compliance warnings, which never change the verdict or exit code:
  ComplianceResult.warnings and EchoSource.warnings list
  ComplianceWarning(kind, message, paths) for files that are neither in a
  durable format nor UTF-8 text (grouped by extension), SQLite databases with
  a non-empty -wal or -journal sidecar, and directories more than two
  levels down that nothing inspected. Telling text from binary reads the
  first 4 KB of a file whose extension is not on the list; that is the only
  place longecho reads inside a data file, and warnings are computed only
  when something asks for them, so query without --json reads no file
  contents. check prints warnings (--verbose names the files);
  query --json includes them when present.
• check --verbose shows which readers can open the source's own files.
• longecho formats shows each format's readers and reason;
  longecho formats --json prints the list as data.
• EchoSource.files: every file a source owns, durable or not.
• Inline previews in the generated site for images, audio, and video that
  browsers play, using the browser's own elements (lazy-loaded).

Changed

• Behavior change: durable files inside site/ or tool folders
  (node_modules, venv, dist, and the like) no longer make a directory
  compliant. A viewer is not the data. Durable files in nested sources still
  count.
• Behavior change: lowercase readme.md and readme.txt no longer count
  as their own data file, the same as README.md always has.
• The generated site lists every file a source owns (within two levels,
  minus SQLite sidecars), marking the ones check warns about, instead of
  only the durable ones.
• README HTML in the site is sanitized by an allowlist (an HTMLParser that
  keeps known tags, attributes, and URL schemes) instead of regular
  expressions.
• The degradation ladder is one table that pairs each kind of reader with
  the formats it can open, with a new "standard tools" rung. It replaces the
  separate format ladder of the early essays.
• The top-level README is the archive's "start here" note; the spec now says
  so and describes what it should contain. START_HERE.txt from the early
  essays is retired.
• longecho spec fits 80 columns and covers warnings and the description
  cascade.
• should_skip_directory and SKIP_DIRECTORIES live in checker.py
  (still importable from discovery).

Fixed

• Links in the generated site were computed relative to the archive root,
  but the page lives in site/, so every "Data Files" link and every "Open
  interactive viewer" link pointed into site/ and returned not found.
  Links are now relative to the page and percent-encoded, so names with
  spaces, #, ?, %, or a colon work and can never become a URL scheme.
  A site written outside the archive, or a source that is a symlink to
  somewhere else, links by absolute file:// URI.
• An entity-encoded javascript: link in a README (javascript:) got
  past the old sanitizer.
• longecho build --output pointing at the archive itself, or at a
  directory containing it, overwrote that directory's README. It is now
  refused, with or without --force.
• Building twice into the same custom --output inside the archive turned
  the first build into a sub-source of the second.
• Filenames and README text containing Rich markup such as [red] were
  rendered as styles by the CLI, and unbalanced markup crashed check.
  They are now printed literally.

Install: pip install longecho==0.6.0. Docs: https://queelius.github.io/longecho/