numbers-parser

numbers-parser is a Python module for parsing Apple Numbers .numbers files. It supports Numbers files generated by Numbers version 10.3, and all 11.x up to 11.2 (current as of November 2021).

It supports and is tested against Python versions from 3.6 onwards. It is not compatible with earlier versions of Python.

Currently supported features of Numbers files are:

Multiple sheets per document
Multiple tables per sheet
Text, numeric, date, currency, duration, percentage cell types

Formulas rely on Numbers storing current values which should usually be the case. Formulas themselves rather than the computed values can optionally be extracted. Styles are not supported.

Installation

python3 -m pip install numbers-parser

A pre-requisite for this package is python-snappy which will be installed by Python automatically, but python-snappy also requires that the binary libraries for snappy compression are present. The most straightforward way to achieve this is to use Homebrew and source Python from Homebrew rather than from macOS:

brew install snappy python3
python3 -m pip install numbers-parser

Usage

Reading documents:

from numbers_parser import Document
doc = Document("my-spreasdsheet.numbers")
sheets = doc.sheets()
tables = sheets[0].tables()
rows = tables[0].rows()

Referring to sheets and tables

Both sheets and names can be accessed from lists of these objects using an integer index (list syntax) and using the name of the sheet/table (dict syntax):

# list access method
sheet_1 = doc.sheets()[0]
print("Opened sheet", sheet_1.name)

# dict access method
table_1 = sheets["Table 1"]
print("Opened table", table_1.name)

Accessing data

Table objects have a rows method which contains a nested list with an entry for each row of the table. Each row is itself a list of the column values. Empty cells in Numbers are returned as None values.

data = sheets["Table 1"].rows()
print("Cell A1 contains", data[0][0])
print("Cell C2 contains", data[2][1])

Cell references

In addition to extracting all data at once, individual cells can be referred to as methods

doc = Document("my-spreasdsheet.numbers")
sheets = doc.sheets()
tables = sheets["Sheet 1"].tables()
table = tables["Table 1"]

# row, column syntax
print("Cell A1 contains", table.cell(0, 0))
# Excel/Numbers-style cell references
print("Cell C2 contains", table.cell("C2"))

Merged cells

When extracting data using data() merged cells are ignored since only text values are returned. The cell() method of Table objects returns a Cell type object which is typed by the type of cell in the Numbers table. MergeCell objects indicates cells removed in a merge.

doc = Document("my-spreasdsheet.numbers")
sheets = doc.sheets()
tables = sheets["Sheet 1"].tables()
table = tables["Table 1"]

cell = table.cell("A1")
print(cell.merge_range)
print(f"Cell A1 merge size is {cell.size[0]},{cell.size[1]})

Row and column iterators

Tables have iterators for row-wise and column-wise iteration with each iterator returning a list of the cells in that row or column

for row in table.iter_rows(min_row=2, max_row=7, values_only=True):
    sum += row
for col in table.iter_cole(min_row=2, max_row=7):
    sum += col.value

Pandas

Since the return value of data() is a list of lists, you should be able to pass it straight to pandas like this

import pandas as pd

doc = Document("simple.numbers")
sheets = doc.sheets()
tables = sheets[0].tables()
data = tables[0].rows(values_only=True)
df = pd.DataFrame(data, columns=["A", "B", "C"])

Bullets and lists

Cells that contain bulleted or numbered lists can be identified by the is_bulleted property. Data from such cells is returned using the value property as with other cells, but can additionally extracted using the bullets property. bullets returns a list of the paragraphs in the cell without the bullet or numbering character. Newlines are not included when bullet lists are extracted using bullets.

doc = Document("bullets.numbers")
sheets = doc.sheets()
tables = sheets[0].tables()
table = tables[0]
if not table.cell(0, 1).is_bulleted:
    print(table.cell(0, 1).value)
else:
    bullets = ["* " + s for s in table.cell(0, 1).bullets]
    print("\n".join(bullets))

Bulleted and numbered data can also be extracted with the bullet or number characters present in the text for each line in the cell in the same way as above but using the formatted_bullets property. A single space is inserted between the bullet character and the text string and in the case of bullets, this will be the Unicode character seen in Numbers, for example "• some text".

Numbers File Formats

Numbers uses a proprietary, compressed binary format to store its tables. This format is comprised of a zip file containing images, as well as Snappy-compressed Protobuf .iwa files containing metadata, text, and all other definitions used in the spreadsheet.

Protobuf updates

As numbers-parser includes private Protobuf definitions extracted from a copy of Numbers, new versions of Numbers will inevitably create .numbers files that cannot be read by numbers-parser. As new versions of Numbers are released, the following steps must be undertaken:

Run proto-dump on the new copy of Numbers to dump new Proto files.
- proto-dump assumes version 2.5.0 of Google Protobuf which may need changes to build on more modern OSes. The version linked here is maintained by the author and tested on recent macOS for both arm64 and x86_64 architectures.
- Any . characters in the Protobuf definitions must be changed to _ characters manually, or via the rename_proto_files.py script in the protos directory of this repo.
Connect to a running copy of Numbers with lldb (or any other debugger) and manually copy and reformat the results of po [TSPRegistry sharedRegistry] into mapping.py.
- Versions of macOS >= 10.11 may protect Numbers from being attached to by a debugger - to attach, temporarily disable System IntegrityProtection to get this data.
- The generate_mapping.py script in protos should help turn the output from this step into a recreation of mapping.py

Running make bootstrap will perform all of these steps and generate the Python protos files as well as mapping.py. The makefile assumes that proto-dump is in a repo parallel to this one, but the make variable PROTO_DUMP can be overridden to pass the path to a working version of proto-dump.

Credits

numbers-parser was built by Jon Connell but derived enormously from prior work by Peter Sobot. Both modules are derived from previous work by Sean Patrick O'Brien.

Decoding the data structures inside Numbers files was helped greatly by previous work by Steven Lott.

Formula tests were adapted from JavaScript tests used in fast-formula-parser.

License

All code in this repository is licensed under the MIT License

abrasive / numbers-parser