headson 0.6.1

Budget‑constrained JSON preview renderer (Python bindings)

Head/tail for JSON — but structure‑aware. Get a compact preview that shows both the shape and representative values of your data, all within a strict character budget.

Available as:

• CLI (see Usage)
• Python library (see Python Bindings)

Install

Using Cargo:

cargo install headson

From source:

cargo build --release
target/release/headson --help

Features

• Budgeted output: specify exactly how much JSON you want to see
• Multiple output formats : json (machine‑readable), pseudo (human‑friendly), js (valid JavaScript, most detailed metadata).
• Multiple inputs: preview many files at once with a shared or per‑file budget.
• Fast: can process gigabyte-scale files in seconds (mostly disk-constrained)
• Available as a CLI app and as a Python library

Fits into command line workflows

If you’re comfortable with tools like head and tail, use headson when you want a quick, structured peek into a JSON file without dumping the entire thing.

• head/tail operate on bytes/lines - their output is not optimized for tree structures
• jq you need to craft filters to preview large JSON files
• headson is like head/tail for trees: zero config but it keeps structure and represents content as much as possible

Usage

headson [FLAGS] [INPUT...]

• INPUT (optional, repeatable): file path(s). If omitted, reads JSON from stdin. Multiple input files are supported.
• Prints the preview to stdout. On parse errors, exits non‑zero and prints an error to stderr.

Common flags:

• -n, --budget <BYTES>: per‑file output budget. When multiple input files are provided, the default total budget equals <BYTES> * number_of_inputs.
• -N, --global-budget <BYTES>: total output budget across all inputs. Useful when you want a fixed-size preview across many files (may omit entire files).
  • When used together with --budget, the final total budget is min(global, per_file * number_of_inputs). Files are only truncated if they don't fit into this final global limit, and no single file expands beyond the per‑file budget.
• -f, --template <json|pseudo|js>: output style (default: pseudo)
• -m, --compact: no indentation, no spaces, no newlines
• --no-newline: single line output
• --no-space: no space after : in objects
• --indent <STR>: indentation unit (default: two spaces)
• --string-cap <N>: max graphemes to consider per string (default: 500)
• --head: prefer the beginning of arrays when truncating (keep first N). Strings are unaffected. In pseudo/js templates the omission marker appears near the end; json remains strict. Mutually exclusive with --tail.
• --tail: prefer the end of arrays when truncating (keep last N). Strings are unaffected. In pseudo/js templates the omission marker appears at the start; json remains strict. Mutually exclusive with --head.

Notes:

• With multiple input files:
  • JSON template outputs a single JSON object keyed by the input file paths.
  • Pseudo and JS templates render file sections with human-readable headers when newlines are enabled.
    • If you use --compact or --no-newline (both disable newlines), fileset output falls back to standard inline rendering (no per-file headers) to remain compact.
  • Using --global-budget may truncate or omit entire files to respect the total budget.
  • The tool finds the largest preview that fits the budget; if even the tiniest preview exceeds it, you still get a minimal, valid preview.
  • When passing file paths, directories and binary files are ignored; a notice is printed to stderr for each (e.g., Ignored binary file: ./path/to/file). Stdin mode reads the stream as-is.
  • Head vs Tail sampling: these options bias which part of arrays are kept before rendering. They guarantee the kept segment is contiguous at the chosen side (prefix for --head, suffix for --tail). Display templates may still insert additional internal gap markers inside that kept segment to honor very small budgets; json remains strict and unannotated.

Quick one‑liners:

• Peek a big JSON stream (keeps structure):

  zstdcat huge.json.zst | headson -n 800 -f pseudo
  

• Many files with a fixed overall size:

  headson -N 1200 -f json logs/*.json
  

• Glance at a file, JavaScript‑style comments for omissions:

  headson -n 400 -f js data.json
  

Show help:

headson --help

Examples: head vs headson

Input:

{"users":[{"id":1,"name":"Ana","roles":["admin","dev"]},{"id":2,"name":"Bo"}],"meta":{"count":2,"source":"db"}}

Naive cut (can break mid‑token):

jq -c . users.json | head -c 80
# {"users":[{"id":1,"name":"Ana","roles":["admin","dev"]},{"id":2,"name":"Bo"}],"me

Structured preview with headson (pseudo):

headson -n 120 -f pseudo users.json
# {
#   users: [
#     { id: 1, name: "Ana", roles: [ "admin", … ] },
#     …
#   ]
#   meta: { count: 2, … }
# }

Machine‑readable preview (json):

headson -n 120 -f json users.json
# {"users":[{"id":1,"name":"Ana","roles":["admin"]}],"meta":{"count":2}}

Terminal Demos

Regenerate locally:

• Place tapes under docs/tapes (e.g., docs/tapes/demo.tape)
• Run: cargo make tapes
• Outputs are written to docs/assets/tapes

Python Bindings

A thin Python extension module is available on PyPI as headson.

• Install: pip install headson (ABI3 wheels for Python 3.10+ on Linux/macOS/Windows).
• API:
  • headson.summarize(text: str, *, template: str = "pseudo", character_budget: int | None = None, skew: str = "balanced") -> str
    • template: one of "json" | "pseudo" | "js"
    • character_budget: maximum output size in characters (default: 500)
    • skew: one of "balanced" | "head" | "tail" (focus arrays on start vs end; only affects display templates; json remains strict).

Example:

import json
import headson

data = {"foo": [1, 2, 3], "bar": {"x": "y"}}
preview = headson.summarize(json.dumps(data), template="json", character_budget=200)
print(preview)

# Prefer the tail of arrays (annotations show in pseudo/js only)
print(
    headson.summarize(
        json.dumps(list(range(100))),
        template="pseudo",
        character_budget=80,
        skew="tail",
    )
)

Algorithm

Footnotes

• ^[1] Optimized tree representation: An arena‑style tree stored in flat, contiguous buffers. Each node records its kind and value plus index ranges into shared child and key arrays. Arrays are ingested in a single pass and may be deterministically pre‑sampled: the first element is always kept; additional elements are selected via a fixed per‑index inclusion test; for kept elements, original indices are stored and full lengths are counted. This enables accurate omission info and internal gap markers later, while minimizing pointer chasing.
• ^[2] Priority order: Nodes are scored so previews surface representative structure and values first. Arrays can favor head/mid/tail coverage (default) or strictly the head; tail preference flips head/tail when configured. Object properties are ordered by key, and strings expand by grapheme with early characters prioritized over very deep expansions.
• ^[3] Choose top N nodes (binary search): Iteratively picks N so that the rendered preview fits within the character budget, looping between “choose N” and a render attempt to converge quickly.
• ^[4] Render attempt: Serializes the currently included nodes using the selected template. Omission summaries and per-file section headers appear in display templates (pseudo/js); json remains strict. For arrays, display templates may insert internal gap markers between non‑contiguous kept items using original indices.
• ^[5] Diagram source: The Algorithm diagram is generated from docs/diagrams/algorithm.mmd. Regenerate the SVG with cargo make diagrams before releasing.

License

MIT