HSORT

🌐 Languages

• 🇺🇸 English
• 🇯🇵 日本語
• 🇨🇳 简体中文
• 🇹🇼 繁體中文（台灣）

High-performance sort/merge Python extension backed by a C engine. Uses an internal memory pool for fast sorting of fixed-length, variable-length, and CSV files, plus merge of sorted inputs.

• Sort — Order records by text or numeric keys, ascending or descending
• Merge — Combine multiple sorted files into one stream

Highlights

• Install with pip from PyPI
• hsort CLI and import hsort Python API
• Windows, Linux, macOS
• Unix-style - / -- options

Sort modes

In-memory sort — When enough memory is available for the dataset, sorting avoids temp files.

External sort — When memory is insufficient, data is split, sorted in chunks, and merged using temporary files.

If -W is omitted, the engine estimates memory from input size; if allocation fails, it falls back to external sort.

Features (summary)

• Configurable memory budget (-W) with automatic sizing when omitted
• Fixed-length, variable-length (text/binary), and CSV
• Stdin/stdout; logs on stderr
• Multi-key sort; stable sort (-S); unique output (-U)
• Invalid records can be skipped or written to an error file (-E)
• ASCII vs numeric key modes; large files supported
• Cross-platform

Installation

pip install hsort

Test PyPI (when applicable):

pip install --index-url https://test.pypi.org/simple/ hsort

Check install:

hsort --version
hsort --help

Quick start

CSV (header row, sort column 1)

hsort -C -H -K1 -O output.csv input.csv

Fixed-length (64-byte records, key bytes 0–19)

hsort -L64 -K0,20 -O output.dat input.dat

Variable-length

hsort -K0,10 -O output.dat input.dat

CLI usage

hsort [options] [input files...]

Options are case-sensitive (-C ≠ -c). Short and long forms are equivalent (-C / --csv).

For the authoritative option list, run:

hsort --help

Option reference

Short	Long	Description
-h	--help	Show help
-V	--version	Show version and license info
-C	--csv	Input is CSV
-H	--header	Treat first CSV row as header
-S	--stable	Stable sort
-U	--unique	Unique output (first record per key)
-L BYTES	--record-length BYTES	Fixed-length records; length 1–640KB per file layout
-K KEY_SPEC	--key	Sort key (repeatable). Fixed / variable-length: start[,len][n][a|d] — start: 0-based byte offset; len: optional key length; n: numeric sort (default ASCII); a: ascending (default), d: descending. CSV: col[n][a|d] — column number from 1
-A	--all-asc	Sort whole record ascending. With -K, -K wins; with -R, the later flag wins
-R	--all-desc	Sort whole record descending. With -K, -K wins; with -A, the later flag wins
-P CODE	--newline CODE	Newline for variable-length / CSV: \n, \r, \r\n (default \n)
-D CHAR	--delimiter CHAR	CSV delimiter; use \t for tab (default comma)
-W SIZE	--memory SIZE	Max memory, e.g. 64MB, 1024KB; if omitted, engine sizes from input; minimum 16MB enforced. Not the free-tier total input file size cap (see Free tier below)
-T DIR	--temp-dir DIR	Temp directory for external sort (default: system temp); files named like hsort_<pid>.tmp, removed after sort
-O FILE	--output FILE	Output path (default stdout); may match an input path
-E FILE	--error-file FILE	Invalid records → this file; if omitted, bad records are skipped silently
-M	--merge	Merge already-sorted files
(paths)	—	Input files: multiple allowed, must be last; if omitted, read stdin

Notes

• Short and long options are equivalent (e.g. -C / --csv).
• Options are case-sensitive (-C ≠ -c).

License-related CLI

hsort --license YOUR_KEY      # activate
hsort --check-license         # status

Examples

I. CSV sorting

① Comma delimiter, max memory 100MB, whole-record ascending, stdin → stdout

hsort -C -W100M < in.csv

② LF newline, tab delimiter, multi-key (col1 asc, col3 desc), output + inputs

hsort -C -P'\n' -D'\t' -K1a -K3d -O out.csv in1.csv in2.csv

No -E: rows with fewer than 3 columns are not written to an error file.

③ Unique, whole-record descending, temp dir, output, error file, input

hsort -C -U -R -T /tmp -O out.csv -E err.csv in.csv

Bad records go to err.csv.

④ Stable sort; column 1 ASCII ascending, column 3 numeric descending

hsort -C -S -K1 -K3nd -O out.csv in.csv

II. Fixed-length sorting

① Record length 64, max memory 100MB, whole-record ascending, stdin → stdout

hsort -L64 -W100M < in.dat

② Record length 64, multi-key (asc + numeric desc), output + inputs

hsort -L64 -K0,5a -K3,8nd -O out.dat in1.dat in2.dat

No -E: rows shorter than required keys are not written to an error file.

③ Record length 64, unique, whole-record descending, temp dir, output, error file, input

hsort -L64 -U -R -T /tmp -O out.dat -E err.dat in.dat

④ Record length 64, stable, multi-key (desc + asc), output + input

hsort -L64 -S -K1,3d -K6,9a -O out.dat in.dat

III. Variable-length sorting

① CRLF newline, max memory 100MB, whole-record ascending, stdin → stdout

hsort -P'\r\n' -W100M < in.dat

② Multi-key (asc + numeric desc), output + inputs

hsort -K5,8a -K0,3nd -O out.dat in1.dat in2.dat

③ Unique, whole-record descending, temp dir, output, error file, input

hsort -U -R -T /tmp -O out.dat -E err.dat in.dat

④ Stable, multi-key (asc + desc), output + input

hsort -S -K1,3 -K5,9d -O out.dat in.dat

Notes

1. If format is unspecified, input is treated as variable-length.
2. Options are case-sensitive (e.g. -C ≠ -c).
3. Short and long options mean the same (e.g. -C / --csv).
4. Default newline for variable-length / CSV is \n (LF).
5. Input file paths must appear last; other options can be in any order.
6. After pip install hsort, the hsort command is available.
7. In code: import hsort and call hsort.hsort(args).

Language (UI)

CLI messages default to English. Japanese is used when detected from, in order:

1. HSORT_LANG (explicit)
2. Windows: system UI language
3. Unix/Linux: locale.getdefaultlocale()
4. LANG, LANGUAGE, LC_ALL, LC_MESSAGES

Forcing the language (HSORT_LANG)

Windows PowerShell

$env:HSORT_LANG="en"
hsort --help

$env:HSORT_LANG="ja"
hsort --help

# Persist for user account:
[System.Environment]::SetEnvironmentVariable("HSORT_LANG", "en", "User")

Linux / macOS (Bash)

export HSORT_LANG=en
hsort --help

export HSORT_LANG=ja
hsort --help

Windows CMD

set HSORT_LANG=en
hsort --help

set HSORT_LANG=ja
hsort --help

Accepted values: ja / japanese / jp (Japanese), en / english (English). On PowerShell, use $env:HSORT_LANG, not set.

Debug locale detection

$env:HSORT_DEBUG_LANG="1"
hsort --help

export HSORT_DEBUG_LANG=1
hsort --help

Details print to stderr.

Python API

HSORT can be used as a library.

Import

import hsort

Function

hsort.hsort(args) takes the same argv-style list as the CLI.

hsort.hsort(args: List[str]) -> int

• args: argument list (same as CLI)
• Return: exit code (0 = success)

Examples

Example 1 — Sort CSV by column 1

import hsort

ret = hsort.hsort([
    "-C",
    "-H",
    "-K1",
    "-O", "output.csv",
    "input.csv",
])

if ret == 0:
    print("Sort succeeded")
else:
    print(f"Error, exit code: {ret}")

Example 2 — CSV multi-key (col1 asc, col3 desc)

import hsort

ret = hsort.hsort([
    "-C",
    "-H",
    "-K1",
    "-K3d",
    "-O", "output.csv",
    "input.csv",
])

Example 3 — CSV numeric sort on column 1

import hsort

ret = hsort.hsort([
    "-C",
    "-H",
    "-K1n",
    "-O", "output.csv",
    "input.csv",
])

Example 4 — Fixed-length (64-byte records)

import hsort

ret = hsort.hsort([
    "-L64",
    "-K0,20",
    "-O", "output.dat",
    "input.dat",
])

Example 5 — Variable-length

import hsort

ret = hsort.hsort([
    "-K0,10",
    "-O", "output.dat",
    "input.dat",
])

Example 6 — Unique output

import hsort

ret = hsort.hsort([
    "-C",
    "-H",
    "-K1",
    "-U",
    "-O", "output.csv",
    "input.csv",
])

Example 7 — Stable sort

import hsort

ret = hsort.hsort([
    "-C",
    "-H",
    "-K1",
    "-S",
    "-O", "output.csv",
    "input.csv",
])

Example 8 — Error file

import hsort
import os

input_file = "input.csv"
output_file = "output.csv"
error_file = "errors.csv"

ret = hsort.hsort([
    "-C",
    "-H",
    "-K1",
    "-K2",
    "-E", error_file,
    "-O", output_file,
    input_file,
])

if ret == 0:
    if os.path.exists(output_file):
        print(f"OK: {output_file}")
    if os.path.exists(error_file) and os.path.getsize(error_file) > 0:
        print(f"Some rows in: {error_file}")
else:
    print(f"Failed, exit code: {ret}")

Example 9 — Memory limit

import hsort

ret = hsort.hsort([
    "-C",
    "-H",
    "-K1",
    "-W64MB",
    "-O", "output.csv",
    "input.csv",
])

Example 10 — Batch CSV files

import hsort
import os
from pathlib import Path

def sort_csv_files(input_dir, output_dir):
    input_path = Path(input_dir)
    output_path = Path(output_dir)
    output_path.mkdir(exist_ok=True)

    for csv_file in input_path.glob("*.csv"):
        output_file = output_path / f"sorted_{csv_file.name}"
        ret = hsort.hsort([
            "-C",
            "-H",
            "-K1",
            "-O", str(output_file),
            str(csv_file),
        ])
        if ret == 0:
            print(f"OK: {csv_file.name}")
        else:
            print(f"Fail: {csv_file.name} (exit {ret})")

sort_csv_files("data/input", "data/output")

API reference

hsort.hsort(args) — run a sort/merge; same rules as the CLI.

• args (List[str]): argv-style list.
• Returns (int): exit code.
• Raises
  • ImportError: extension not installed
  • RuntimeError: free-tier total input size exceeded (same condition as CLI)

args must follow the same rules as the command line.

Requirements

• Python 3.9+
• Windows, Linux, or macOS
• Suggest ≥ 16 MB available RAM (tune with -W)

Pricing & License

HSORT offers a free tier and paid licenses:

🟢 Free version

• Up to 100MB total input size
• No feature restrictions (size limit only)

🔵 Paid license

• Unlimited input size
• Full performance
• Commercial use

👉 Activate license:

hsort --license YOUR_KEY

Free tier: total input file size

When no valid license is activated, the combined size of regular input files passed on the command line or in hsort.hsort([...]) must not exceed 100 MiB (100 × 1024 × 1024 bytes). The check is shared by CLI and API.

• Activating a paid license removes this total input-size limit (subject to your license agreement).

License

Copyright (c) 2015–2026 株式会社GPO

This project is not open source. The software is proprietary; see the LICENSE file for full terms.

• Not open source — No general right to source, redistribution, or modification except as allowed by law or a written agreement.
• Free tier — May include limits (e.g. total input size as above). Does not grant full commercial rights.
• Paid license — Unlocks full features per your agreement with the publisher.

👉 Buy License (Instant Key Delivery):
https://github.com/xuhui-hou/hsort/blob/main/Payment.md

Licensing contact: soft@gpo-i.com

Links

• Homepage: https://github.com/xuhui-hou/hsort
• Repository: https://github.com/xuhui-hou/hsort
• Issues: https://github.com/xuhui-hou/hsort/issues