long-xls 0.1.1

Recover data from XLS files that exceed the 65,536-row limit

GitHub | 한국어

long-xls

A simple tool to recover data from XLS files that exceed the 65,536-row limit.

The XLS format is structurally limited to 65,536 rows per sheet. However, some programs — legacy reporting utils, converting tools, stock trading platform exporters, etc. — keep writing BIFF cell records past this limit. The data is physically present in the file, but most programs and libraries (Excel, pandas, xlrd) either silently truncate it or throw an error, leaving no way to read the stored data.

long-xls reads the raw BIFF binary stream, detects row-index wrap-arounds at the 65,536 boundary, and reconstructs the full dataset.

The Problem

┌─────────────────────────────────────────────────────┐
│  Your XLS file (e.g. 370,000 rows of trade data)    │
│                                                     │
│  Row 1 ............ ✓ visible in Excel              │
│  Row 65,536 ....... ✓ visible in Excel              │
│  Row 65,537 ....... ✗ INVISIBLE — data is there     │
│  Row 370,000 ...... ✗ INVISIBLE — but recoverable!  │
└─────────────────────────────────────────────────────┘

Install

pip install long-xls              # xlsx output (default)
pip install "long-xls[parquet]"   # + parquet support
pip install "long-xls[all]"       # everything

Or download a standalone executable from Releases — no Python required.

Quick Start

# Convert to xlsx (default)
long-xls data.xls

# Convert to csv
long-xls data.xls -f csv

# Convert to parquet
long-xls data.xls -f parquet

# Include a JSON schema sidecar file
long-xls data.xls --schema

# Multiple files at once
long-xls *.xls -f csv -o output/

Commands

Command	Description
long-xls data.xls	Convert to xlsx (default)
long-xls data.xls -f csv	Convert to csv
long-xls data.xls -f parquet	Convert to parquet
long-xls data.xls --schema	Also write .schema.json
long-xls schema data.xls	Print JSON schema to stdout
long-xls scan data.xls	Quick file scan (record counts only)

Options

Option	Default	Description
-f, --format	xlsx	Output format: xlsx, csv, parquet
-o, --output-dir	same as input	Output directory
-e, --encoding	cp949	Text encoding for string cells
-y, --force	off	Overwrite existing files without asking
--schema	off	Write a .schema.json alongside output

Python API

from long_xls import parse, parse_to_dataframe, schema_json

# Parse and inspect
sheet = parse("data.xls")
print(f"{sheet.num_data_rows:,} rows recovered")

# Get a pandas DataFrame
df, sheet = parse_to_dataframe("data.xls")

# Export schema as JSON
import json
print(json.dumps(schema_json(sheet), indent=2))

Schema Output Example

{
  "file": "data.xls",
  "file_size": 29736094,
  "encoding": "cp949",
  "num_columns": 4,
  "num_data_rows": 371700,
  "row_limit_exceeded": true,
  "wraps": {"0": 5, "1": 5, "2": 5, "3": 5},
  "columns": [
    {"index": 0, "name": "date", "type": "string", "non_null_count": 371700},
    {"index": 1, "name": "time", "type": "string", "non_null_count": 371700},
    {"index": 2, "name": "price", "type": "float", "non_null_count": 371700},
    {"index": 3, "name": "volume", "type": "integer", "non_null_count": 371700}
  ]
}

Test Files

Synthetic test files (generator included)

Run tests/generate_test_xls.py to generate long-XLS test files of various sizes. It writes raw BIFF2 records to reproduce the row wrap-around behaviour.

python tests/generate_test_xls.py

File	Rows	Wraps	Encoding	Purpose
test_100k_rows.xls	100,000	1	UTF-8	Basic recovery verification
test_200k_rows.xls	200,000	3	UTF-8	Multi-wrap verification
test_70k_cp949.xls	70,000	1	CP949	Korean encoding verification

Real-world example: Kiwoom Securities HTS chart data

Futures tick chart data exported from Kiwoom Securities HTS. Contains 371,700 rows, but Excel only shows up to 65,535. long-xls recovers the entire dataset.

• 20240808_KOSPI200_Tick_Kiwoom.xls — 371,700 rows (5 wraps), 29.7 MB

Building a Standalone Executable

pip install pyinstaller
python build_exe.py          # produces dist/long-xls.exe (Windows)

Author

seonhwa17kim (with help from GPT-5.5, Gemini 3.5, Claude Opus 4.8)

License

MIT Copyright (c) 2026 seonhwa17kim