Hi, not sure I'm missing something, new to the lib. I am having the reported error, on these lines of the lib:

is it my fault? My code is like this:

def deserializeFromCsv(self, base_folder: str) -> MondrianAccount:
with open(base_folder + "FakeAccount.csv") as users_csv:
reader = DataclassReader(users_csv, MondrianAccount)
print(reader)
for row in reader:
print(row)

    return reader[0]

I'm not using Union, if that helps

Description

At the moment the DataclassReader behave exactly the same as the DictReader when the csv file contains duplicated headers. By default, it will get the value of the last column, for example, assuming that we have a csv file with the following content:

firstname,lastname,age,age
John,Test,41,23
Edit,Test,3,34

Using the DictReader, we get the values:

[
    OrderedDict([('firstname', 'John'), ('lastname', 'Test'), ('age', '23')]), 
    OrderedDict([('firstname', 'Edit'), ('lastname', 'Test'), ('age', '34')])
]

Note the the values for the field age are the values of the second age column. The DataclassReader has the same behaviour, if we would have a data class containing the field age, it would get the values of the second age column (23 and 34).

This can easily lead to mistakes and unpredictable results when reading the csv file.

Proposed behavior

It would be nice if instead of just getting the values of the last column with the same name, the DataclassReader could raise an error saying that the "The CSV file {filename} contains duplicated column names: {column names}".

The DataclassReader could also get a boolean argument called duplicate_columns_error (we can find a better name) which would disable this verification and have the same behavior as the DictReader.

Need also to discuss if it would be better to have this feature disabled or enabled by default.

Other information

OS: Debian 4.9.130-2
Python version: 3.7.1
dataclass-csv version: 1.0.1

I ran into a problem with dataclasses with fields that have default values. I managed to reproduce the issue with a small change in one of the tests: Modify UserWithDateFormatDecorator in tests/mocks.py to contain a datetime with a default value:

@dateformat('%Y-%m-%d')
@dataclasses.dataclass
class UserWithDateFormatDecorator:
    name: str
    create_date: datetime
    date_with_default: datetime = datetime.now()

Executing the test shows the error:

TypeError: strptime() argument 1 must be str, not datetime.datetime

Apparently, dataclass-csv tries to parse the default value as if it were an input string from the CSV, although in reality it is already a datetime (and can't be changed to a str either).

CSV files and other databases commonly have fields that are NULL (missing / empty). As an easy example, a string field in a row might be of zero length. Handling such situations is currently complicated and confusing with dataclass-csv.

The current code allows a workaround by specifying a default value.

The example given for default values is this:

@dataclass
class User:
    firstname: str
    email: str = 'Not specified'
    age: int

But that generates this error:

Traceback (most recent call last):
  File "example.py", line 5, in <module>
    class User:
  File "/usr/lib/python3.8/dataclasses.py", line 1019, in dataclass
    return wrap(cls)
  File "/usr/lib/python3.8/dataclasses.py", line 1011, in wrap
    return _process_class(cls, init, repr, eq, order, unsafe_hash, frozen)
  File "/usr/lib/python3.8/dataclasses.py", line 925, in _process_class
    _init_fn(flds,
  File "/usr/lib/python3.8/dataclasses.py", line 502, in _init_fn
    raise TypeError(f'non-default argument {f.name!r} '
TypeError: non-default argument 'age' follows default argument

It actually works to simply put all the fields which have default values at the end of the class, since the order of class members doesn't matter:

@dataclass
class User:
    firstname: str
    age: int
    email: str = ''

A quick, partial solution would seem to be fixing the example, and noting the need to put the fields with default values at the end.

But requiring that the fields not appear in order reduces the clarify of the code and is more convoluted than necessary.

I think it would be good to also automatically move the fields needing default values to the end of the generated class.

It is perhaps also worthwhile to provide a way to specify that NULL values are ok for all fields, though that requires figuring out what value to use as a default default, I guess....

Say If I have the following classes for serialization to csv.

@dataclass
class Link:
    title: str
    url: str
    
@dataclass
class SearchResult:	
    paper_name: Link
    authors: list[Link]
    publication: Link

I want to split paper_name into paper_name.title and paper_name.url columns in the generated csv file.

See also #33 (comment).

Maybe some annotation can be used on the classes of the property type to indicate the purpose.

While the existing datetime support is nice, some additional flexibility would be nice. When reading date strings, for example, the ability to use a well known parser such as dateutil.parser or, the ability to manipulate the datetime object as it's created (such as making sure the conversion to UTC happens correctly, if needed), or even to instantiate a datetime from a non-human format, such as a unix timestamp.

To that end, in addition to the dateformat metadata that exists today, maybe some sort of datefactory or dateformatter metadata, which would be a callable, passing in the value, and returning a well-formed datetime object would be useful.

I came across this problem when trying to work around #17.

Given a data file with a column 'date', which contains an integer (a unix timestamp), and
a dataclass with a field of date which is of type datetime.datetime. Since there isn't a way to convert that in dataclass_csv today, I munged the dataclass a bit, made date an init=False field, introduced an InitVar of unix_ts, and used post_init to convert the unix timestamp to a datetime field like so:

@dataclass
class SomeData:
    date: datetime.datetime = field(init=False)
    unix_ts: InitVar[int]

    def __post_init__(self, unix_ts: int):
        self.date = datetime.datetime.utcfromtimestamp(unix_ts)

This works exactly as expected: the init signature is looking for unix_ts instead of date, date gets set with the value of unix_ts, all my consumers of the dataclass object don't have to change.

... until I try to load the data from dataclass_csv, at any rate.

reader = DataclassReader(inp, SomeData)
reader.map('date').to('unix_ts')
for record in reader:
   ...

This generates the standard 'hey, you're telling me to map onto a datetime field, and didn't tell me how to convert:

AttributeError: Unable to parse the datetime string value. Date format not specified. To specify a date format for all datetime fields in the class, use the @dateformat decorator. To define a date format specifically for this field, change its definition to: `date: datetime = field(metadata={'dateformat': <date_format>})`

Okay, fine, I figure I'll open an issue (here it is!), and I'll rename the date field on my dataclass to something else. Then I run into #9 because despite having init=False, it's still trying to build it into the construction of my class.

So, between #9 and #17, I'm actually stuck at the moment, but while working both of those out, I encountered this issue and thought I'd raise it.

The following example from the README works.

import dataclasses
import io

from dataclass_csv import DataclassReader


@dataclasses.dataclass
class User:
    firstname: str
    email: str = "Not specified"


reader = DataclassReader(io.StringIO("firstname,e-mail\nElsa,[email protected]"), User)
reader.map("e-mail").to("email")
print(list(reader))  # [User(firstname='Elsa', email='[email protected]')]

Unfortunately introducing a typo leads to silent data loss.

reader.map("e_mail").to("email")
print(list(reader))  # [User(firstname='Elsa', email='Not specified')]

I think raising when an explicitly mapped column does not occur in the header of the data would help prevent errors.

It seems that distutils will be soon be obsolete.

/usr/local/lib/python3.10/site-packages/dataclass_csv/dataclass_reader.py:5
  /usr/local/lib/python3.10/site-packages/dataclass_csv/dataclass_reader.py:5: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
    from distutils.util import strtobool

I am hoping this is just a naive user question:

refer to this line of code in the DataClassReader:

Task:

Using subclasses defining a logical data model, read a csv file with name different than in the data model. In the DataClassReader using the .map method to map columns names in the csv file to field names in the sales class. The class 'sales' includes 'markets' and 'products' master data as subclasses.

code and traceback:

from dataclasses import dataclass, field
import dataclass_csv
import datetime

#define dataclasses to align with a logical dimensional data model
@DataClass
class markets:
market_id : str = field(metadata={'accept_whitespaces': True},init=False)
market : str = field(metadata={'accept_whitespaces': True},init=True)

@DataClass
class products:
upc : str = field(metadata={'accept_whitespaces': True},init=True)
product : str = field(metadata={'accept_whitespaces': True},init=True)

@DataClass
@dataclass_csv.dateformat('%Y-%m-%d')
class sales(markets, products):
weekend : datetime = field(metadata={'accept_whitespaces': False},init=True)
total_units : float = field(metadata={'accept_whitespaces': False},init=True,default=0.00000)
total_revenue : float = field(metadata={'accept_whitespaces': False},init=True,default=0.00000)

#assume all data is provided in a single file.

with open('sales.csv','r') as infile:
reader = dataclass_csv.DataclassReader(infile, sales)
reader.map('Week').to('weekend')
reader.map('Market').to('market')
reader.map('EANUPC').to('upc')
reader.map('Product Description').to('product')
reader.map('Sales Units').to('total_units')
reader.map('Sales Dollars').to('total_revenue')
for row in reader:
print(row)

At this point I am expecting the see records like:
sales(weekend = ..., ..., total_revenue = 100.00)
...

However I get the following traceback on the first iteration of the for loop:

Traceback (most recent call last):

File "", line 44, in
for row in reader:

File "/Users/user/opt/anaconda3/lib/python3.7/site-packages/dataclass_csv/dataclass_reader.py", line 211, in next
return self._process_row(row)

File "/Users/user/opt/anaconda3/lib/python3.7/site-packages/dataclass_csv/dataclass_reader.py", line 195, in _process_row
transformed_value = field_type(value)

TypeError: 'module' object is not callable

Looking at the dataclass_reader.py I see "field_type = field.type", but not defined as a callable function anywhere.

Python 3.7.0 64 bit
Spyder IDE 3.3.6
Mac OS

Hi,

thanks for this very nice package. Do you have plans to support more custom fields as well (e.g. by passing in conversion functions?).

For me, the one type I very often use is "Path" as a lot of csv-files I have reference paths somewhere. Of course I can easily convert later, but if I define a dataclass, would be nice to define it correctly from the start!

Thanks

Description

At the moment the code does not have any docstrings. Docstrings are helpful when developing since the developers can get help right in their text editor, IDE or the python REPL.

When reading a csv file, it would be nice if when iterating through DataclassReader the type checker could recognise the objects as being the type that the DataclassReader was instantiated with. At the moment it just returns Any.

This generalises #20.

I completed the example from the README. The following works.

import dataclasses
import io
import re

from dataclass_csv import DataclassReader


class SSN:
    def __init__(self, val):
        if re.match(r"\d{9}", val):
            self.val = f"{val[0:3]}-{val[3:5]}-{val[5:9]}"
        elif re.match(r"\d{3}-\d{2}-\d{4}", val):
            self.val = val
        else:
            raise ValueError(f"Invalid SSN: {val!r}")


@dataclasses.dataclass
class User:
    name: str
    ssn: SSN


print(list(DataclassReader(io.StringIO("name,ssn\nAlice,123456789"), User)))

After replacing the SSN in the data with a default value, the error below occurs.

class User:
    name: str
    ssn: SSN = SSN("123456789")


print(list(DataclassReader(io.StringIO("name,ssn\nAlice,"), User)))

  File "readme.py", line 10, in __init__
    if re.match(r"\d{9}", val):
TypeError: expected string or bytes-like object

The reason being that the SSN constructor is called with the default value, which is an object. The next snippet works around that.

class User:
    name: str
    ssn: SSN = "123456789"

But now this class can no longer be instantiated directly without creating broken objects, and type checkers will complain.

I think if a default value exists it should be used as is, and not passed to the constructor of its class.

As a convenient alternative to explicitly mapping column to field names, in case they don't directly match, it would be handy if dataclass-csv could populate dataclass' fields in the order they are declared. This would be particularly handy in the presence of CSV files that have many columns with cryptic headers (whose generation is not under the users control).

E.g. given the following CSV file

firstname, age
Jean, 77
Jim, 22
...

and dataclass

@dataclass
class Person:
  forename: str
  age: int

I would expected the following Person instances to be created:

User(forename='Jean', age=77)
User(forename='Jim', age=22)
...

That is, the fields from top to bottom are initialised with the row data from left to right.

It'd be nice to have some way of allowing users to specify custom field types. E.g.:

• a metadata field that maps raw strings to the desired type
• an ABC that users can subclass and override some parse method that the library will call (similar to how Go's csv lib works)

When reading datetime fields from a file the @dateformat() works well but when writing the same dataclass with DataclassWriter it does not write them to the file in this format.

It would be useful if, rather than having to add a default (potentially meaningless) value to a field, instead you could mark it as Optional.

from dataclasses import dataclass
from typing import Optional

@dataclass
class User:
    firstname: str
    email: Optional[str]
    age: int

If the field is not provided in the row, the field is set to None. This will allow for a type-safe way to detect and handle missing values.

This code:

#!/usr/bin/env python3

from dataclasses import dataclass

from dataclass_csv import DataclassReader


@dataclass
class User:
    firstname: str
    email: str
    age: int


def main() -> None:
    with open('test.csv') as users_csv:
        reader = DataclassReader(users_csv, User)
        for row in reader:
            print(row)


if __name__ == '__main__':
    main()

Runs correctly:

User(firstname='Elsa', email='[email protected]', age=11)
User(firstname='Astor', email='[email protected]', age=7)
User(firstname='Edit', email='[email protected]', age=3)
User(firstname='Ella', email='[email protected]', age=2)

With the following test.csv:

firstname,email,age
Elsa,[email protected], 11
Astor,[email protected], 7
Edit,[email protected], 3
Ella,[email protected], 2

But using __future__.annotations, which will be mandatory in Python 4.0 and is useful in Python 3.7+, breaks DataclassReader:

#!/usr/bin/env python3

from __future__ import annotations

from dataclasses import dataclass

from dataclass_csv import DataclassReader


@dataclass
class User:
    firstname: str
    email: str
    age: int


def main() -> None:
    with open('test.csv') as users_csv:
        reader = DataclassReader(users_csv, User)
        for row in reader:
            print(row)


if __name__ == '__main__':
    main()

Traceback (most recent call last):
  File "test.py", line 25, in <module>
    main()
  File "test.py", line 20, in main
    for row in reader:
  File "venv/lib/python3.7/site-packages/dataclass_csv/dataclass_reader.py", line 211, in __next__
    return self._process_row(row)
  File "venv/lib/python3.7/site-packages/dataclass_csv/dataclass_reader.py", line 159, in _process_row
    or '__origin__' in field_type.__dict__

Nested dataclasses are not handled by the DataclassReader. While the DataclassWriter serializes a nested dataclass as a tuple (thanks to the recursive behavior of dataclasses.astuple()), the DataclassReader tries to initialize the nested dataclass with:

This cannot work because value is a string representing a tuple, whereas field_type (that in this case is a dataclass) constructor expects the fields necessary to build the dataclass.

The solution is pretty simple:

if dataclasses.is_dataclass(field_type):
    import ast
    fields = ast.literal_eval(value)
    values[field.name] = field_type(*fields)
    continue

This IF-statement could be placed in dataclass_reader.py, line 230. This is just a draft, but if you think it would be a good improvement I can propose a PR with error handling and tests as well.

Description

The issue occurs when defining a field in the dataclass with the property init=False. When it init property is set to false it means that the auto-generated dataclass initializer won't add the field in the list of arguments.

How to reproduce

1. Create a dataclass and set init=False in one of the fields, for example:

from dataclasses import dataclass, field

from dataclass_csv import DataclassReader


@dataclass
class User:
    firstname: str
    lastname: str
    age: int = field(init=False)

2. Try reading a csv file containing the expected data

def main():

    with open('users.csv') as f:
        reader = DataclassReader(f, User)

        for row in reader:
            print(row)


if __name__ == '__main__':
    main()

3. You will see the error on the terminal:

Traceback (most recent call last):
  File "app.py", line 25, in <module>
    main()
  File "app.py", line 20, in main
    for row in reader:
  File "/home/daniel/Envs/test-oGAe6Cfg/lib/python3.7/site-packages/dataclass_csv/dataclass_reader.py", line 161, in __next__
    return self._process_row(row)
  File "/home/daniel/Envs/test-oGAe6Cfg/lib/python3.7/site-packages/dataclass_csv/dataclass_reader.py", line 157, in _process_row
    return self.cls(*values)
TypeError: __init__() takes 3 positional arguments but 4 were given

Expected behavior

When setting init=False in a dataclass field, the field should not be passed to the dataclass initializer.

Other information

OS: Debian 4.9.130-2
Python version: 3.7.1
dataclass-csv version: 1.0.1

It would be great if, given a List[SomeDataclass] it were possible to write back out a series of dataclasses into their corresponding CSV representation.

This would allow someone to read a CSV into dataclasses, modify the data in a type safe way, and then write it back out to CSV.

This generalises #17.

Currently any class not special-cased by the library must have a constructor that takes a single string.

Even for built-in types this leads to workarounds like the following.

class Time(datetime.time):
    def __new__(cls, value: str):
        hour, minute = value.split(":")
        return super().__new__(cls, int(hour), int(minute))

I think supporting transformation functions would add much-needed flexibility.

One possible syntax might be this.

def strptime(value: str):
    hour, minute = value.split(":")
    return datetime.time(int(hour), int(minute))

reader.map('time').using(strptime)  # With or without .to()

dataclass-csv 1.1.2 .docx
Example for french documentation

Description

Support various string representations of bool values.

Use Case

I'd like to use dataclass-csv as part of a data pipeline which imports data from a CSV intermediate format into PostgreSQL. Currently any non-null boolean value is parsed as True, which prevents me from using this nice package.

Proposed solution

    def _parse_bool_value(self, bool_value):
        """Parses `str` representation of a boolean value to a `bool`. Case-insensitive.
        Values parsed as True: 'yes', 'true', 't', 'y', '1'
        Values parsed as False: 'no', 'false', 'f', 'n', '0
        All other values(including empty/NULL) parsed as None.
        """
        return str2bool(bool_value)

We can use this in _process_row(). This leverages a very lightweight package called str2bool and I put a little wrapper around it just so we can have a unit test around this parsing in case we decide to change the implementation later.

If you'd rather not add this dependency, we could re-implement str2bool pretty easily.

I've got this ready to roll locally, so feel free to assign this one to me and I can get a PR submitted shortly.

Btw, I'm really liking this package and dataclasses in general and would be glad to help out more in the future. Thanks!

Since CSV data is often used without changing it, it would be great if dataclass-csv supported frozen dataclasses. This way, users could enjoy the convenience of dataclass-csv and the additional safety of immutable data.

When just exploring lots of different csv files, it would be very handy to have an optional way to scan a csv file and generate a dataclass that would work with it.

It could start off just reading row headers and using them to come up with legal attribute names.

Next, automatically generating attribute types by analyzing the contents of each column would also be handy.

If that funtionality could be provided as module functions and called from other software, it could eventually be possible to start playing with csv files that you run across, and have the module automatically read a csv in and provide access to the rows as objects which could be introspected.

The type annotations for the DataclassReader are incorrect:

Even though *args and **kwds are technically a List and a Dict respectively, they are not typed the same way. See mypy docs or PEP-484. Changing both to Any solves this problem.

Mypy output when calling the DataclassReader Ctor:

reader = DataclassReader(
    lines,
    MyDataclass,
    delimiter="+",
#             ^ Argument "delimiter" to "DataclassReader" has incompatible type "str"; expected "Dict[str, Any]"
)