aiocsv

Asynchronous CSV reading and writing.

Installation

pip install aiocsv. Python 3.11+ is required.

This module contains an extension written in C. Pre-build binaries may not be available for your configuration. You might need a C compiler and Python headers to install aiocsv.

Usage

AsyncReader & AsyncDictReader accept any object that has a read(size: int) coroutine, which should return a string.

AsyncWriter & AsyncDictWriter accept any object that has a write(b: str) coroutine.

Reading is implemented using a custom CSV parser, which should behave exactly like the CPython parser.

Writing is implemented using the synchronous csv.writer and csv.DictWriter objects - the serializers write data to a StringIO, and that buffer is then rewritten to the underlying asynchronous file.

Example

Example usage with aiofiles.

import asyncio
import csv

import aiofiles
from aiocsv import AsyncReader, AsyncDictReader, AsyncWriter, AsyncDictWriter

async def main():
    # simple reading
    async with aiofiles.open("some_file.csv", mode="r", encoding="utf-8", newline="") as afp:
        async for row in AsyncReader(afp):
            print(row)  # row is a list

    # dict reading, tab-separated
    async with aiofiles.open("some_other_file.tsv", mode="r", encoding="utf-8", newline="") as afp:
        async for row in AsyncDictReader(afp, delimiter="\t"):
            print(row)  # row is a dict

    # simple writing, "unix"-dialect
    async with aiofiles.open("new_file.csv", mode="w", encoding="utf-8", newline="") as afp:
        writer = AsyncWriter(afp, dialect="unix")
        await writer.writerow(["name", "age"])
        await writer.writerows([
            ["John", 26], ["Sasha", 42], ["Hana", 37]
        ])

    # dict writing, all quoted, "NULL" for missing fields
    async with aiofiles.open("new_file2.csv", mode="w", encoding="utf-8", newline="") as afp:
        writer = AsyncDictWriter(afp, ["name", "age"], restval="NULL", quoting=csv.QUOTE_ALL)
        await writer.writeheader()
        await writer.writerow({"name": "John", "age": 26})
        await writer.writerows([
            {"name": "Sasha", "age": 42},
            {"name": "Hana"}
        ])

asyncio.run(main())

Differences with csv

aiocsv strives to be a drop-in replacement for Python's builtin csv module. However, there are a few notable differences, due to technical limitations or CPython bugs:

• Readers accept objects with async read methods, instead of an AsyncIterable over lines from a file.
• AsyncDictReader.fieldnames can be None - use await AsyncDictReader.get_fieldnames() instead.
• Changes to csv.field_size_limit are not picked up by existing Reader instances. The field size limit is cached on Reader instantiation to avoid expensive function calls on each character of the input.
• QUOTE_NOTNULL and QUOTE_STRINGS work on readers even in 3.12. aiocsv does not replicate CPython bug #113732.

Other, minor, differences include:

• AsyncReader.line_num, AsyncDictReader.line_num and AsyncDictReader.dialect are not settable,
• AsyncDictReader.reader is of AsyncReader type,
• AsyncDictWriter.writer is of AsyncWriter type,
• AsyncDictWriter provides an extra, read-only dialect property.

Reference

aiocsv.AsyncReader

AsyncReader(
    asyncfile: aiocsv.protocols.WithAsyncRead,
    dialect: str | csv.Dialect | Type[csv.Dialect] = "excel",
    **csv_dialect_kwargs: Unpack[aiocsv.protocols.CsvDialectKwargs],
)

An object that iterates over records in the given asynchronous CSV file. Additional keyword arguments are understood as dialect parameters.

Iterating over this object returns parsed CSV rows (List[str]).

Methods:

• __aiter__(self) -> self
• async __anext__(self) -> List[str]

Read-only properties:

• dialect (aiocsv.protocols.DialectLike): The dialect used when parsing
• line_num (int): The number of lines read from the source file. This coincides with a 1-based index of the line number of the last line of the recently parsed record.

aiocsv.AsyncDictReader

AsyncDictReader(
    asyncfile: aiocsv.protocols.WithAsyncRead,
    fieldnames: Optional[Sequence[str]] = None,
    restkey: Optional[str] = None,
    restval: Optional[str] = None,
    dialect: str | csv.Dialect | Type[csv.Dialect] = "excel",
    **csv_dialect_kwargs: Unpack[aiocsv.protocols.CsvDialectKwargs],
)

An object that iterates over records in the given asynchronous CSV file. All arguments work exactly the same was as in csv.DictReader.

Iterating over this object returns parsed CSV rows (Dict[str, str]).

Methods:

• __aiter__(self) -> self
• async __anext__(self) -> Dict[str, str]
• async get_fieldnames(self) -> List[str]

Properties:

• fieldnames (List[str] | None): field names used when converting rows to dictionaries
  ⚠️ Unlike csv.DictReader, this property can't read the fieldnames if they are missing - it's not possible to await on the header row in a property getter. Use await reader.get_fieldnames().

  reader = csv.DictReader(some_file)
  reader.fieldnames  # ["cells", "from", "the", "header"]
  
  areader = aiofiles.AsyncDictReader(same_file_but_async)
  areader.fieldnames   # ⚠️ None
  await areader.get_fieldnames()  # ["cells", "from", "the", "header"]
  

• restkey (str | None): If a row has more cells then the header, all remaining cells are stored under this key in the returned dictionary. Defaults to None.
• restval (str | None): If a row has less cells then the header, then missing keys will use this value. Defaults to None.
• reader: Underlying aiofiles.AsyncReader instance

Read-only properties:

• dialect (aiocsv.protocols.DialectLike): Link to self.reader.dialect - the current csv.Dialect
• line_num (int): The number of lines read from the source file. This coincides with a 1-based index of the line number of the last line of the recently parsed record.

aiocsv.AsyncWriter

AsyncWriter(
    asyncfile: aiocsv.protocols.WithAsyncWrite,
    dialect: str | csv.Dialect | Type[csv.Dialect] = "excel",
    **csv_dialect_kwargs: Unpack[aiocsv.protocols.CsvDialectKwargs],
)

An object that writes csv rows to the given asynchronous file. In this object "row" is a sequence of values.

Additional keyword arguments are passed to the underlying csv.writer instance.

Methods:

• async writerow(self, row: Iterable[Any]) -> None: Writes one row to the specified file.
• async writerows(self, rows: Iterable[Iterable[Any]]) -> None: Writes multiple rows to the specified file.

Readonly properties:

• dialect (aiocsv.protocols.DialectLike): Link to underlying's csv.writer's dialect attribute

aiocsv.AsyncDictWriter

AsyncDictWriter(
    asyncfile: aiocsv.protocols.WithAsyncWrite,
    fieldnames: Sequence[str],
    restval: Any = "",
    extrasaction: Literal["raise", "ignore"] = "raise",
    dialect: str | csv.Dialect | Type[csv.Dialect] = "excel",
    **csv_dialect_kwargs: Unpack[aiocsv.protocols.CsvDialectKwargs],
)

An object that writes csv rows to the given asynchronous file. In this object "row" is a mapping from fieldnames to values.

Additional keyword arguments are passed to the underlying csv.DictWriter instance.

Methods:

• async writeheader(self) -> None: Writes header row to the specified file.
• async writerow(self, row: Mapping[str, Any]) -> None: Writes one row to the specified file.
• async writerows(self, rows: Iterable[Mapping[str, Any]]) -> None: Writes multiple rows to the specified file.

Properties:

• fieldnames (Sequence[str]): Sequence of keys to identify the order of values when writing rows to the underlying file
• restval (Any): Placeholder value used when a key from fieldnames is missing in a row, defaults to ""
• extrasaction (Literal["raise", "ignore"]): Action to take when there are keys in a row, which are not present in fieldnames, defaults to "raise" which causes ValueError to be raised on extra keys, may be also set to "ignore" to ignore any extra keys
• writer: Link to the underlying AsyncWriter

Readonly properties:

• dialect (aiocsv.protocols.DialectLike): Link to underlying's csv.reader's dialect attribute

aiocsv.protocols.WithAsyncRead

A typing.Protocol describing an asynchronous file, which can be read.

aiocsv.protocols.WithAsyncWrite

A typing.Protocol describing an asynchronous file, which can be written to.

aiocsv.protocols.DialectLike

Type of an instantiated dialect property. Thank CPython for an incredible mess of having unrelated and disjoint csv.Dialect and _csv.Dialect classes.

aiocsv.protocols.CsvDialectArg

Type of the dialect argument, as used in the csv module.

aiocsv.protocols.CsvDialectKwargs

Keyword arguments used by csv module to override the dialect settings during reader/writer instantiation.

Development

Contributions are welcome, however please open an issue beforehand. aiocsv is meant as a replacement for the built-in csv, any features not present in the latter will be rejected.

Building from source

To create a wheel (and a source tarball), run python -m build.

For local development, use a virtual environment. pip install --editable . will build the C extension and make it available for the current venv. This is required for running the tests. However, due to the mess of Python packaging this will force an optimized build without debugging symbols. If you need to debug the C part of aiocsv and build the library with e.g. debugging symbols, the only sane way is to run python setup.py build --debug and manually copy the shared object/DLL from build/lib*/aiocsv to aiocsv.

Tests

This project uses pytest with pytest-asyncio for testing. Run pytest after installing the library in the manner explained above.

Linting & other tools

This library uses black and isort for formatting and pyright in strict mode for type checking.

For the C part of library, please use clang-format for formatting and clang-tidy linting, however this are not yet integrated in the CI.

Installing required tools

pip install -r requirements.dev.txt will pull all of the development tools mentioned above, however this might not be necessary depending on your setup. For example, if you use VS Code with the Python extension, pyright is already bundled and doesn't need to be installed again.

Recommended VS Code settings

Use Python, Pylance (should be installed automatically alongside Python extension), and Ruff extensions.

You will need to install all dev dependencies from requirements.dev.txt, except for pyright. Recommended .vscode/settings.json:

{
    "C_Cpp.codeAnalysis.clangTidy.enabled": true,
    "python.testing.pytestArgs": [
        "."
    ],
    "python.testing.unittestEnabled": false,
    "python.testing.pytestEnabled": true,
    "[python]": {
        "editor.defaultFormatter": "charliermarsh.ruff",
        "editor.formatOnSave": true,
        "editor.codeActionsOnSave": {
            "source.organizeImports": "always"
        }
    },
    "[c]": {
        "editor.formatOnSave": true
    }
}

For the C part of the library, C/C++ extension is sufficient. Ensure that your system has Python headers installed. Usually a separate package like python3-dev needs to be installed, consult with your system repositories on that. .vscode/c_cpp_properties.json needs to manually include Python headers under includePath. On my particular system this config file looks like this:

{
    "configurations": [
        {
            "name": "Linux",
            "includePath": [
                "${workspaceFolder}/**",
                "/usr/include/python3.14"
            ],
            "defines": [],
            "compilerPath": "/usr/bin/clang",
            "cStandard": "c17",
            "cppStandard": "c++17",
            "intelliSenseMode": "linux-clang-x64"
        }
    ],
    "version": 4
}