Resource Can't load data from some file-like objects

[LincolnPuzey]

frictionless version: 5.17.0
python version: 3.12
platform: linux

The docs here https://framework.frictionlessdata.io/docs/schemes/stream.html say data can be loaded from file-like objects,

this does work for the example given when the object is an opened file:

from frictionless import Resource

with open("example.csv", "rb") as f:
    resource = Resource(
        source=f,
        format="csv",
        encoding="utf-8",
)
print(resource.read_rows())

# prints [{'A': 1, 'B': 2}]

However for classes like BytesIO and SpooledTemporaryFile it doesn't work

e.g. Bytes IO

from io import BytesIO
from frictionless import Resource

file_like_object = BytesIO(b"A,B\n1,2")
resource = Resource(
    source=file_like_object,
    format="csv",
    encoding="utf-8",
)
print(resource.read_rows())

Error:

Traceback (most recent call last):
    print(resource.read_rows())
          ^^^^^^^^^^^^^^^^^^^^
  File "python3.12/site-packages/frictionless/resources/table.py", line 423, in read_rows
    with helpers.ensure_open(self):
  File "python3.12/contextlib.py", line 137, in __enter__
    return next(self.gen)
           ^^^^^^^^^^^^^^
  File "python3.12/site-packages/frictionless/helpers/general.py", line 97, in ensure_open
    thing.open()
  File "python3.12/site-packages/frictionless/resources/table.py", line 161, in open
    self.__open_parser()
  File "python3.12/site-packages/frictionless/resources/table.py", line 178, in __open_parser
    self.__parser.open()
  File "python3.12/site-packages/frictionless/system/parser.py", line 95, in open
    self.__loader = self.read_loader()
                    ^^^^^^^^^^^^^^^^^^
  File "python3.12/site-packages/frictionless/system/parser.py", line 126, in read_loader
    return loader.open()
           ^^^^^^^^^^^^^
  File "python3.12/site-packages/frictionless/system/loader.py", line 107, in open
    self.__byte_stream = self.read_byte_stream()
                         ^^^^^^^^^^^^^^^^^^^^^^^
  File "python3.12/site-packages/frictionless/system/loader.py", line 137, in read_byte_stream
    byte_stream = self.read_byte_stream_create()
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "python3.12/site-packages/frictionless/schemes/stream/loader.py", line 17, in read_byte_stream_create
    if not os.path.isfile(byte_stream.name):  # type: ignore
                          ^^^^^^^^^^^^^^^^
AttributeError: '_io.BytesIO' object has no attribute 'name'

e.g. SpooledTemporaryFile

from tempfile import SpooledTemporaryFile
from frictionless import Resource

file_like_object = SpooledTemporaryFile(max_size=1024, mode="w+b")
file_like_object.write(b"A,B\n1,2")
file_like_object.seek(0)

resource = Resource(
    source=file_like_object,
    format="csv",
    encoding="utf-8",
)

print(resource.read_rows())

Error mostly the same except for:

File "python3.12/site-packages/frictionless/schemes/stream/loader.py", line 17, in read_byte_stream_create
    if not os.path.isfile(byte_stream.name):  # type: ignore
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen genericpath>", line 30, in isfile
TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType

---

Is there reason frictionless expects the file object to have .name that is an actual file
in read_byte_stream_create?

Being able to pass a SpooledTemporaryFile would be particularly useful,
because FastAPI presents uploaded files as SpooledTemporaryFile instances.

[LincolnPuzey]

With some further testing I found calling .rollover() on the SpooledTemporaryFile made it work,

e.g. this works

from tempfile import SpooledTemporaryFile
from frictionless import Resource

file_like_object = SpooledTemporaryFile(max_size=1024, mode="w+b")
file_like_object.write(b"A,B\n1,2")
file_like_object.seek(0)
file_like_object.rollover()  # force file to disk
resource = Resource(
    source=file_like_object,
    format="csv",
    encoding="utf-8",
)

print(resource.read_rows())
# prints [{'A': 1, 'B': 2}]

However it also appears that frictionless closes the file at some point, because when we try to

file_like_object.seek(0)
print(file_like_object.read())

after frictionless does its work, we get a ValueError: seek of closed file error.
This is not ideal since we want to use frictionless to process the file, then read it again to save it somewhere else.

[pierrecamilleri]

Thanks for the report and for the workaround. I agree that automatically closing the file may not be the best behaviour.

The part that crashes is part of an exception handling :

    if not os.path.isfile(byte_stream.name):  # type: ignore
       note = f"only local streams are supported: {byte_stream}"
       raise FrictionlessException(errors.SchemeError(note=note))

So supporting file-like in-memory data seems like a new feature to me.

However, in the meanwhile, I can at least : 

• make sure that the right exception with explicit error message is raised
• Update the docs to make it clearer.