Skip to content

One engine, many doors

There is exactly one AGS4 engine — a Rust parser + validator — and every way you reach it is a door onto that same engine. Pick the door that fits where you are; the behaviour behind it is identical.

One engine

The parser that tokenises an AGS4 file, the dictionary that types each column, and the numbered-rules validator are all Rust. Nothing reimplements them per language. So a file that reads clean in Python reads clean at the CLI, in Node, and inside DuckDB — same dtypes, same findings, same rule numbers. There is no "Python parser" drifting from a "JavaScript parser"; there is one engine with several front doors:

  • Python — import laterite (this site).
  • lat CLI — lat validate delivery.ags --json (see the CLI reference).
  • Node — @laterite/* on npm.
  • DuckDB — the laterite_ags4 loadable extension.

Many input doors

Inside Python, read itself has three doors — and they all feed the same engine, so it doesn't matter which one your data arrives through:

import laterite

# A minimal AGS4 string — GROUP / HEADING / UNIT / TYPE rows, then DATA.
ags4_text = '''"GROUP","LOCA"
"HEADING","LOCA_ID","LOCA_GL"
"UNIT","","m"
"TYPE","ID","2DP"
"DATA","BH01","23.68"
"DATA","BH02","32.49"
'''

ags = laterite.read(text=ags4_text)   # the text= door
loca = ags["LOCA"]
print(loca)
print({h: str(loca[h].dtype) for h in ("LOCA_ID", "LOCA_GL")})

shape: (2, 2)
┌─────────┬─────────┐
│ LOCA_ID ┆ LOCA_GL │
│ ---     ┆ ---     │
│ str     ┆ f64     │
╞═════════╪═════════╡
│ BH01    ┆ 23.68   │
│ BH02    ┆ 32.49   │
└─────────┴─────────┘
{'LOCA_ID': 'String', 'LOCA_GL': 'Float64'}

The same 2DP → Float64 typing you get from a file on disk falls out of an in-memory string, because the door is just a way in — the engine behind it is the same. The three doors are:

  • read("delivery.ags") — a path on disk.
  • read(text=...) — an in-memory AGS4 string.
  • read(data=raw_bytes) — raw bytes (an upload, an HTTP body, a blob).

All three return the same object, so the rest of your code never asks where the data came from.

Why it matters

Identical behaviour across surfaces is a guarantee, not a coincidence: validate in CI with lat, then read the same file in Python and get the same verdict. And because the input doors converge, a web upload (data=), a pasted snippet (text=), and a file (path) all flow through one code path — no special-casing.

← Back to Read