Can't extract files from `.7z` zipfile using `download_and_extract`

Describe the bug

I’m adding a new dataset which is a .7z zip file in Google drive and contains 3 json files inside. I’m able to download the data files using download_and_extract but after downloading it throws this error:

>>> dataset = load_dataset("./datasets/mantis/")
Using custom data configuration default
Downloading and preparing dataset mantis/default to /Users/bhavitvyamalik/.cache/huggingface/datasets/mantis/default/1.1.0/611affa804ec53e2055a335cc1b8b213bb5a0b5142d919967729d5ee23c6bab4...
Downloading data: 100%|█████████████████████████████████████████████████████████| 77.2M/77.2M [00:23<00:00, 3.28MB/s]
/Users/bhavitvyamalik/.cache/huggingface/datasets/downloads/fc3d70123c9de8407587a59aa426c37819cf2bf016795d33270e8a1d558a34e6
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/bhavitvyamalik/Desktop/work/hf/datasets/src/datasets/load.py", line 1745, in load_dataset
    use_auth_token=use_auth_token,
  File "/Users/bhavitvyamalik/Desktop/work/hf/datasets/src/datasets/builder.py", line 595, in download_and_prepare
    dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
  File "/Users/bhavitvyamalik/Desktop/work/hf/datasets/src/datasets/builder.py", line 690, in _download_and_prepare
    ) from None
OSError: Cannot find data file. 
Original error:
[Errno 20] Not a directory: '/Users/bhavitvyamalik/.cache/huggingface/datasets/downloads/fc3d70123c9de8407587a59aa426c37819cf2bf016795d33270e8a1d558a34e6/merged_train.json'

just before generating the splits. I checked fc3d70123c9de8407587a59aa426c37819cf2bf016795d33270e8a1d558a34e6 file and it’s 7z zip file (similar to downloaded Google drive file) which means it didn’t get unzip. Do I need to unzip it separately and then pass the paths for train,dev,test files in SplitGenerator?

Environment info

• datasets version: 1.18.4.dev0
• Platform: Darwin-19.6.0-x86_64-i386-64bit
• Python version: 3.7.8
• PyArrow version: 5.0.0

[albertvillanova]

Hi @bhavitvyamalik, thanks for reporting.

Yes, currently we do not support 7zip archive compression: I think we should.

As a workaround, you could uncompress it explicitly, like done in e.g. samsum dataset:

https://github.com/huggingface/datasets/blob/fedf891a08bfc77041d575fad6c26091bc0fce52/datasets/samsum/samsum.py#L106-L110

[albertvillanova]

Hi @bhavitvyamalik, thanks for your investigation.

On Monday, I started a PR that will eventually close this issue as well: I’m linking it to this.

• #4672

Let me know what you think.