scan_csv does not work on s3 path while read_csv works

[michael72]

Checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

some_s3_file = f"s3://{BUCKET}/data.csv"
pl.read_csv(some_s3_file) # works
pl.scan_csv(some_s3_file) # doesn't work

Log output

2023-12-19 07:43:57,983 - analysis_helper - ERROR - No such file or directory (os error 2): s3://*****************/data.csv

Issue description

scan_csv fails reading s3 data

Expected behavior

scan_csv should at least support the same locations as read_csv

Installed versions

--------Version info---------
Polars:               0.20.1
Index type:           UInt32
Platform:             Linux-5.19.0-46-generic-x86_64-with-glibc2.35
Python:               3.11.4 (main, Jun  7 2023, 12:45:48) [GCC 11.3.0]

----Optional dependencies----
adbc_driver_manager:  <not installed>
cloudpickle:          2.2.1
connectorx:           <not installed>
deltalake:            <not installed>
fsspec:               2023.12.2
gevent:               <not installed>
matplotlib:           3.7.2
numpy:                1.26.0
openpyxl:             <not installed>
pandas:               2.0.3
pyarrow:              14.0.1
pydantic:             <not installed>
pyiceberg:            <not installed>
pyxlsb:               <not installed>
sqlalchemy:           <not installed>
xlsx2csv:             <not installed>
xlsxwriter:           <not installed>

[ritchie46]

scan_csv doesn't support reading from the cloud yet.

[edblackburn]

Hi @ritchie46 is there a plan to introduce scanning CSV files from object storage like S3? Is it on a roadmap? Is there a roadmap I can look at, sorry if this is in neon lights elsewhere and I missed it. Are there any significant blockers, and is there anything the community can do to help? Sorry for so many questions!

[edblackburn]

I am following up on my previous comment and inquiring about the possibility of a piecemeal approach to adding object storage features. Specifically, is it possible to explore a route similar to the one taken for Parquet files, where storage options are introduced as arguments for scan_csv as they exist for scan_parquet?

If seeking parity between scan and read for CSV files is a stated objective, I assume such a change represents a substantial engineering endeavour. However, a phased approach might offer a way to incrementally build the capability without deviating too much from the goal of achieving parity between read_csv and scan_csv. I note this has been requested prior in issue #7225

Are there any significant blockers, or is there anything the community can contribute to help? Please let us know.

Thank you, Ed

[nameexhaustion]

Closed by #16674