headson 0.6.4

Budget‑constrained JSON preview renderer (Python bindings)

heal/tail for JSON, YAML - but structure‑aware. Get a compact preview that shows both the shape and representative values of your data, all within a strict character budget. (Just like head/tail, headson can also work with unstructured text files.)

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 you want to see
• Output formats: auto | json | yaml | text
  • Styles: strict | default | detailed
    • JSON family: strict → strict JSON, default → human‑friendly Pseudo, detailed → JS with inline comments
    • YAML: always YAML; strict has no comments, default uses “# …”, detailed uses “# N more …”
    • Text: prints raw lines. In default style, omissions are shown as a single line …; in detailed, as … N more lines …. strict omits array‑level summaries.
• Multiple inputs: preview many files at once with a shared or per‑file budget
• Fast: processes gigabyte‑scale files in seconds (mostly disk‑bound)
• 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 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:

• -c, --bytes <BYTES>: per‑file output budget. For multiple inputs, default total budget is <BYTES> * number_of_inputs.
• -C, --global-bytes <BYTES>: total output budget across all inputs. With --bytes, the effective total is the smaller of the two.
• -f, --format <auto|json|yaml|text>: output format (default: auto).
  • Auto: stdin → JSON family; filesets → per‑file based on extension (.json → JSON family, .yaml/.yml → YAML, unknown → Text).
• -t, --template <strict|default|detailed>: output style (default: default).
  • JSON family: strict → strict JSON; default → Pseudo; detailed → JS with inline comments.
  • YAML: always YAML; style only affects comments (strict none, default “# …”, detailed “# N more …”).
• -i, --input-format <json|yaml|text>: ingestion format (default: json). For filesets in auto format, ingestion is chosen by extensions.
• -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. Display styles place omission markers accordingly; strict JSON remains unannotated. Mutually exclusive with --tail.
• --tail: prefer the end of arrays when truncating (keep last N). Strings are unaffected. Display styles place omission markers accordingly; strict JSON remains unannotated. Mutually exclusive with --head.

Notes:

• Multiple inputs:
  • With newlines enabled, file sections are rendered with human‑readable headers. In compact/single‑line modes, headers are omitted.
• In --format auto, each file uses its own best format: JSON family for .json, YAML for .yaml/.yml.
  • Unknown extensions are treated as Text (raw lines) — safe for logs and .txt files.
  • --global-bytes may truncate or omit entire files to respect the total budget.
  • The tool finds the largest preview that fits the budget; even if extremely tight, you still get a minimal, valid preview.
  • Directories and binary files are ignored; a notice is printed to stderr for each. Stdin reads the stream as‑is.
  • Head vs Tail sampling: these options bias which part of arrays are kept before rendering. Display styles may still insert internal gap markers to honor very small budgets; strict JSON stays unannotated.

Quick one‑liners:

• Peek a big JSON stream (keeps structure):

  zstdcat huge.json.zst | headson -c 800 -f json -t default
  

• Many files with a fixed overall size:

  headson -C 1200 -f json -t strict logs/*.json
  

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

  headson -c 400 -f json -t detailed data.json
  

• YAML with detailed comments:

  headson -c 400 -f yaml -t detailed config.yaml
  

Text mode

• Single file (auto):

  headson -c 200 notes.txt
  

• Force Text ingest/output (useful when mixing with other extensions):

  headson -c 200 -i text -f text notes.txt
  

• Many text files (fileset):

  headson -c 800 -i text -f text logs/*.txt
  

• Styles on Text:

  • default: omission as a standalone … line.
  • detailed: omission as … N more lines ….
  • strict: no array‑level omission line (individual long lines may still truncate with …).

Show help:

headson --help

Note: flags align with head/tail conventions (-c/--bytes, -C/--global-bytes).

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 (JSON family, default style → Pseudo):

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

Machine‑readable preview (JSON family, strict style → strict JSON):

headson -c 120 -f json -t strict 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, *, format: str = "auto", style: str = "default", input_format: str = "json", character_budget: int | None = None, skew: str = "balanced") -> str
    • format: "auto" | "json" | "yaml" (auto maps to JSON family for single inputs)
    • style: "strict" | "default" | "detailed"
    • input_format: "json" | "yaml" (ingestion)
    • character_budget: maximum output size in characters (default: 500)
    • skew: "balanced" | "head" | "tail" (affects display styles; strict JSON remains unannotated)

Examples:

import json
import headson

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

# Prefer the tail of arrays (annotations show with style="default"/"detailed")
print(
    headson.summarize(
        json.dumps(list(range(100))),
        format="json",
        style="detailed",
        character_budget=80,
        skew="tail",
    )
)

# YAML support
doc = "root:\n  items: [1,2,3,4,5,6,7,8,9,10]\n"
print(headson.summarize(doc, format="yaml", style="default", input_format="yaml", character_budget=60))

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