hissbytenotation 0.3.0

Serialize and deserialize data to ordinary python source notation (hbn)

Hiss Byte Notation

Library to make it easy to use python literal syntax as a data format. No JavaScript and fast.

Have you seen people try to print a dict and then use the JSON library to parse the output? This library is some helper function for that scenario. It has an API similar to other serialization libraries with dumps, loads and so on.

It is a small wrapper all the ways that python has to convert python literals to live python

• ast.literal_eval
• import
• etc.

However, I don't recommend using any of those except Pickle because performance is so bad. Instead use hsbn with the rust speed ups which are now the default.

Safety

ast.literal_eval is safer than eval but the python docs still imply that there are malicious payloads. I'm not sure if they are the same problems that could affect json or other formats.

The Rust parser (by_rust=True) is safe — it only parses literal syntax and never executes arbitrary code.

Installation

Install the base package with:

pip install hissbytenotation

To install every optional Python integration in one shot, use:

pip install "hissbytenotation[all]"

If you use uv, the equivalent command is:

uv sync --extra all

Install just one optional feature with one of these extras:

• fmt for hbn fmt / Black integration: pip install "hissbytenotation[fmt]" or uv sync --extra fmt
• glom for query and mutation commands: pip install "hissbytenotation[glom]" or uv sync --extra glom
• validate for Cerberus-backed schema validation: pip install "hissbytenotation[validate]" or uv sync --extra validate

The optional Rust speedups are bundled in supported platform wheels for hissbytenotation; if your platform does not have a wheel yet, install the package normally and fall back to the pure-Python path, or build the extension locally:

cd rust && maturin develop --release

Usage

Learn how to use it with the GUI which demos all the features

hissbytenotation gui

Normal usage is as a library.

import hissbytenotation as hbn

data = {
    "mammal": "cat",
    "reptile": ["snake", "lizard"],
    "version": 1
}
data_as_string = hbn.dumps(data)

rehydrated = hbn.loads(data_as_string)
print(rehydrated)
# {'mammal': 'cat', 'reptile': ['snake', 'lizard'], 'version': 1}

CLI

It now has a Bash-first CLI with the hbn command:

uv run hbn dump --arg "{'mammal': 'cat', 'version': 1}"
uv run hbn convert --from json --to hbn --arg '{"mammal": "cat", "version": 1}'
uv run hbn keys --arg "{'mammal': 'cat', 'version': 1}" --lines
uv run hbn dump --arg "['cat', 'snake']" --bash-array animals
uv run hbn dump --arg "{'host': 'db', 'port': '5432'}" --bash-assoc cfg
uv run hbn get users.0.email users.hbn --raw
uv run hbn q --glom "{'emails': ('users', ['email'])}" users.hbn
uv run hbn set users.0.role --value "'admin'" users.hbn
uv run hbn merge left.hbn right.hbn
uv run hbn merge --strategy append-lists left.hbn right.hbn
uv run hbn diff left.hbn right.hbn
uv run hbn doctor --to json --pretty
uv run hbn repl users.hbn

Supported core formats:

• HBN input and output
• JSON input and output
• TOML input
• XML input and output via a simple @attrs / #text mapping
• Bash Map Notation (bmn) for flat dicts and scalar arrays

Useful shell-oriented output flags:

• --raw
• --lines
• --nul
• --shell-quote
• --shell-assign NAME
• --shell-export NAME
• --bash-array NAME
• --bash-assoc NAME
• --default VALUE

For nested traversal and mutation, install hissbytenotation[all] or just the glom extra:

uv sync --extra glom
uv run hbn q users.0.email users.hbn --raw
uv run hbn q --glom "{'emails': ('users', ['email'])}" users.hbn
uv run hbn append users --value "{'email': '[e-mail]'}" users.hbn

It now has glom commands, which is a query language for python dicts and types:

• q / query
• get
• set
• del
• append
• insert

Glom query mode supports:

• simple dot-path lookups such as users.0.email
• --glom SPEC for explicit glom specs written in HBN / Python literal syntax
• --spec-file PATH for reusable glom specs

It supports merge and mutation workflows:

• merge
• merge strategies: replace, shallow, deep, append-lists, set-union-lists
• conflict policies: error, left-wins, right-wins
• file mutation flags: --in-place, --backup SUFFIX, --atomic, --check

Examples:

uv run hbn merge --conflict right-wins --left-arg "{'a': 1}" --right-arg "{'a': 2}"
uv run hbn set config.port --value 6432 --in-place settings.hbn
uv run hbn append users --value "{'email': '[e-mail]'}" --backup .bak users.hbn

It supports some discoverability and interactive workflows:

• repl
• doctor
• friendly aliases such as show, format, delete, and count
• richer help and examples topics via hbn help TOPIC and hbn examples TOPIC
• shell completion scripts for Bash, Zsh, Fish, and PowerShell

Examples:

uv run hbn show --arg "{'a': 1}"
uv run hbn delete users.0.role users.hbn
uv run hbn count --arg "[1, 2, 3]"
uv run hbn doctor --compact
uv run hbn completion powershell
uv run hbn repl users.hbn

Inside the REPL:

load {'users': [{'email': '[e-mail]'}]}
get users.0.email --raw
set users.0.role --value "'admin'"
merge --value "{'users': [{'email': '[e-mail]'}]}" --strategy append-lists
write session.json --to json --pretty

The doctor command reports optional capabilities and install hints for:

• formatter support via black
• diff helper support, preferring git diff --no-index when available
• optional glom integration
• optional hbn_rust acceleration
• uv and git availability on PATH

Install shell completions by evaluating or sourcing the generated script for your shell:

uv run hbn completion bash
uv run hbn completion zsh
uv run hbn completion fish
uv run hbn completion powershell

For formatter support, install hissbytenotation[all] or just the fmt extra and use hbn fmt:

uv sync --extra fmt
uv run hbn fmt --arg "{'b':2,'a':1}"

hbn fmt prefers shelling out to the black executable when it is available and falls back to the Python package API when it is installed without the command on PATH.

It supports optional extras and helpers:

• fmt for formatter integration
• diff for canonicalized text diffs

Examples:

uv run hbn diff left.hbn right.hbn
uv run hbn diff --to json left.hbn right.hbn
uv run hbn diff --tool builtin --left-arg "{'a': 1}" --right-arg "{'a': 2}"
uv run hbn fmt --arg "{'b':2,'a':1}"

The diff command:

• canonicalizes both inputs as HBN or JSON
• prefers git diff --no-index when git is available
• falls back to a builtin unified diff renderer otherwise
• returns 0 for no diff and 1 when differences are found

Rust-accelerated parsing

Install hissbytenotation normally, or pip install "hissbytenotation[all]" if you also want all Python extras. Platform wheels can include the optional Rust parser for a ~14x speedup over ast.literal_eval; source installs fall back to the pure-Python implementation:

pip install hissbytenotation
# or build the optional extension from source:
cd rust && maturin develop --release

fast = hbn.loads(data_as_string, by_rust=True)

How it works

Serialization is done by calling repr, checking if ast.literal_eval can read it. Repr can be called on more data structures than ast.literal_eval can handle.

Because ast.literal_eval is so slow, there are other options for deserialization:

• Rust parser (by_rust=True): ~14x faster than ast.literal_eval. Safe — no code execution. Available when the optional Rust extension is present.
• default: ast.literal_eval with validation enabled. Very slow, very safe.
• eval (by_eval=True): Slow, only for trusted data.
• exec (by_exec=True): Slow, only for trusted data.
• import (by_import=True): Two times faster than exec, only for trusted data.

Deserialization benchmark — 1,000 iterations

Method	Time	Notes
Pickle	0.084s
JSON	0.252s
HBN (Rust parser)	0.510s	safe
HBN (eval)	4.968s	unsafe
HBN (ast.literal_eval)	6.984s	safe
HBN (exec)	5.357s	unsafe
HBN (import)	10.790s	unsafe
repr() only	0.339s	serialize

Rust parser is ~14x faster than ast.literal_eval and the fastest safe option.

Run the benchmark yourself:

python -m benchmark
python -m benchmark -n 5000  # more iterations

Prior art

• literal-python-to-pickle A faster replacement for ast.literal_eval and corresponding question on stackoverflow.

• You could just call repr and ast.literal_eval directly.

Possibly astor which serializes to a string representation of the AST, which looks nothing like the source code, nor json.