Tier adapters

Tier	Formats	Analysis
T5	Registered source dialects	CST structure and semantic concepts
T4	JSON/JSONC/JSON5, YAML, TOML, INI, ENV	Typed keys, values, and containers
T3	HTML/XML and related markup	Elements, attributes, and text
T2	Plain text/logs and selected extensionless documents	Word changes and paragraph moves
T1	Markdown, other textual formats, unsupported code	Raw line sequence
T0	Binary extensions or undecodable bytes	Changed bytes or optional plugin

classify_file assigns an initial tier from the path. is_parseable checks semantic-parser availability. Code/data/markup parser failure or analysis limits fall directly to raw lines. Numbered tiers describe abstraction levels, not a mandatory parser cascade. Data parsing is format-directed. Scalar type changes, nested YAML edits, and non-label HTML attribute changes remain visible. Original byte comparisons preserve binary differences that text decoding could erase. See the tier reference for Rust APIs and pipeline for fallback details.

Structured data: change one setting

database:
    pool:
        max: 10

database:
    pool:
        max: 25

differens before.yaml after.yaml

  ~ changed value of leaf `max` from `10` to `25` in object `pool`

The four-space nesting is preserved. The output names the containing object and the changed key; data ranges are synthetic tree ranges, not source offsets.

Markup: change an attribute

<a href="/old" title="a > b">Docs</a>

<a href="/new" title="a > b">Docs</a>

differens before.html after.html

  ~ changed file content (`/old` → `/new`)

The text and title stay the same. The href value changes even though it is not the element’s label. The > inside a quoted attribute remains part of its value.

Prose: edit a word

Ship the blue widget.

Ship the green widget.

differens before.txt after.txt

  ~ changed word (`blue` → `green`)

A paragraph that changes can produce word actions. Identical paragraphs can also move within a document. This handling applies to prose inputs, including .txt and .log files.

Markdown: keep line structure

# Release notes

Ship the blue widget.

# Release notes

Ship the green widget.

differens before.md after.md

  ~ changed line (`Ship the blue widget.` → `Ship the green widget.`)

The same wording change in Markdown is a line edit. Markdown formatting and code blocks are not parsed as an embedded programming language.

Binary: changed bytes of the same length

For binary data, equal lengths do not imply equal contents. You can reproduce a three-byte comparison with:

printf '\001\002\003' > before.bin
printf '\001\002\004' > after.bin
differens before.bin after.bin

  ~ changed file content (`3 bytes` → `3 bytes`)

The default binary description reports lengths. Use JSON to detect the Update action; the length text alone does not explain which byte changed. Format-specific plugins can provide more detail through the Rust library API.

Add an adapter

Choose an existing tier when it fits the format. Build a validated tree through TreeBuilder and add routing in crates/differens-tiers/src/router.rs; raw/prose adapters may use the shared sequence engine. Keep parser errors visible as line fallbacks. See contributing and the tiers API for the extension points.

Semantic Extractors