[drsamonline]

What I read

CLDR's unit conversion lived in one function, extras.rs::convert_units, over three tables — length (9), mass (7), temperature (3) — reached from the smart cascade after math and before path / PATH-binary / web. The amount side was split by a byte scan that only accepted [-+]?digits[.digits].

What was wrong

Everything outside those three dimensions fell silently through to a web search. Measured against the old logic: 2 h in min, 90 min in h, 10 GB in MB, 5 km/h in mph, 3 tbsp in ml, 1 bar in psi → None, i.e. a DuckDuckGo query. Two parse bugs in the same function: 10 in in cm (the inch collides with the " in " connector) and 1e3 m in km (no exponent in the literal).

What I changed

Units (minor). One table carries a dimension with every factor; conversion requires both units to share a dimension. Dimensions: length, mass, time, volume, area, speed, data, pressure, energy, temperature. Data is decimal (kb = 1000 B) with the binary forms spelled out (kib, mib, …). Factors follow Numbat's unit definitions (MIT/Apache-2.0) — physical constants only, no code, credited in the source. No dependency added.

Parsing. Candidate connector splits are tried right to left and the first that resolves on both sides wins, so 10 in in cm is ten inches into centimetres. A suffix is only taken as a unit when a number — or an expression the existing calc engine already handles — sits in front of it, so 1e3 m in km and 2*3 km in mi work while 5min stays minutes rather than 5 metres + "in".

/cmd truncation (patch). A line cut at 500 characters now ends … (+N chars) instead of a bare ellipsis, so a wrapped line and a 40 KB minified JSON line no longer look the same.

Docs follow the behaviour: docs/USER_GUIDE.md, docs/index.html, the README feature row and the /cmd note. Two changes/ files carry the bump levels; VERSION is untouched.

Evidence

CI on this branch — gh run view 42613276705: both Check jobs green (ubuntu-latest 26s, windows-latest 1m17s).

cargo build --release --locked   Finished `release` profile in 7.24s
cargo test  --release --locked   test result: ok. 20 passed; 0 failed; 0 ignored
cargo run --release --locked -- --self-test
                                 self-test: OK (18/18)

Five new tests (conversion_time_volume_data, conversion_speed_area_pressure_energy, conversion_parsing_resolves_the_split, long_lines_say_how_much_was_cut) plus four new --self-test checks (14 → 18). No benchmark is claimed.

What I deliberately did not do

• No dependency. clap, fend, kalker and numbat are references here, not crates: the binary is ~600 KB and advertises zero-bloat.
• No output-format change. The guide's example shows 10 km in mi → 6.213712 mi (with the unit); the code has never appended it. Pre-existing doc/behaviour mismatch, left alone rather than folded into this diff.
• No general compound-unit parser. km/h, m/s, mph are table entries; 5 ft 2 in in cm (two amount tokens) still does not parse.
• No tree-wide cargo fmt. The rustfmt step was already red on main (50 diffs across 7 files, all line width) and is non-blocking; my additions follow the file's existing long-line style and add 9 diffs of the same kind. Cleaning that is a separate sweep, not this PR.
• No version bump, per the brief.