Disable aiohttp requoting of redirection URL

[albertvillanova]

The library aiohttp performs a requoting of redirection URLs that unquotes the single quotation mark character: %27 => '

This is a problem for our Hugging Face Hub, which requires exact URL from location header.

Specifically, in the query component of the URL (https://netloc/path?query), the value for response-content-disposition contains %27:

response-content-disposition=attachment%3B+filename*%3DUTF-8%27%27sample.jsonl.gz%3B+filename%3D%22sample.jsonl.gz%22%3B

and after the requoting, the %27 characters get unquoted to ':

response-content-disposition=attachment%3B+filename*%3DUTF-8''sample.jsonl.gz%3B+filename%3D%22sample.jsonl.gz%22%3B

This PR disables the aiohttp requoting of redirection URLs.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[albertvillanova]

Comment by @lhoestq:

Do you think we need this in datasets if it's fixed on the moon landing side ? In the aiohttp doc they consider those symbols as "non-safe"

[albertvillanova]

The lib requests does not perform that requote on redirect URLs.

[lhoestq]

I'm fine with aligning with requests - this way we have less differences between streaming and non-streaming.

Maybe add a comment on why you needed to add this parameter ?

[albertvillanova]

Indeed, the requests library does perform a requoting, but this does not unquote %27:

In [1]: from requests.utils import requote_uri

In [2]: url = "https://netloc/path?param=param%27%27value"

In [3]: url
Out[3]: 'https://netloc/path?param=param%27%27value'

In [4]: requote_uri(url)
Out[4]: 'https://netloc/path?param=param%27%27value'

However, the aiohttp library uses yarl.ULR and this does unquote %27:

In [5]: from yarl import URL

In [6]: url
Out[6]: 'https://netloc/path?param=param%27%27value'

In [7]: str(URL(url))
Out[7]: "https://netloc/path?param=param''value"

If we pass requote_redirect_url=False to aiohttp, then it passes encoded=True to yarl.ULR: https://github.com/aio-libs/aiohttp/blob/4635161ee8e7ad321cca46e01ce5bfeb1ad8bf26/aiohttp/client.py#L578-L580

parsed_url = URL(
    r_url, encoded=not self._requote_redirect_url
)

which does not unquote %27:

In [8]: url
Out[8]: 'https://netloc/path?param=param%27%27value'

In [9]: str(URL(url, encoded=True))
Out[9]: 'https://netloc/path?param=param%27%27value'

[lhoestq]

Thanks for the explanations :)

[albertvillanova]

See the issues we opened in the respective libraries:

• aiohttp
  • Requoting unquotes %27 aio-libs/yarl#1077
• requests
  • Requoting does not unquote %27 psf/requests#6341

[github-actions]

Show benchmarks

PyArrow==6.0.0

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.012399 / 0.011353 (0.001047)	0.006388 / 0.011008 (-0.004620)	0.134173 / 0.038508 (0.095665)	0.037059 / 0.023109 (0.013949)	0.420697 / 0.275898 (0.144799)	0.473981 / 0.323480 (0.150502)	0.009857 / 0.007986 (0.001871)	0.004791 / 0.004328 (0.000463)	0.106886 / 0.004250 (0.102636)	0.044871 / 0.037052 (0.007818)	0.429843 / 0.258489 (0.171354)	0.461569 / 0.293841 (0.167728)	0.057285 / 0.128546 (-0.071261)	0.018809 / 0.075646 (-0.056837)	0.432613 / 0.419271 (0.013342)	0.058086 / 0.043533 (0.014553)	0.413064 / 0.255139 (0.157925)	0.444407 / 0.283200 (0.161207)	0.119102 / 0.141683 (-0.022581)	1.875954 / 1.452155 (0.423799)	1.916392 / 1.492716 (0.423676)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.267489 / 0.018006 (0.249483)	0.567554 / 0.000490 (0.567064)	0.005901 / 0.000200 (0.005701)	0.000134 / 0.000054 (0.000079)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.031248 / 0.037411 (-0.006164)	0.123014 / 0.014526 (0.108489)	0.140001 / 0.176557 (-0.036556)	0.191476 / 0.737135 (-0.545659)	0.141687 / 0.296338 (-0.154652)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.637481 / 0.215209 (0.422272)	6.255969 / 2.077655 (4.178314)	2.559811 / 1.504120 (1.055691)	2.118154 / 1.541195 (0.576960)	2.079487 / 1.468490 (0.610997)	1.201079 / 4.584777 (-3.383698)	5.592625 / 3.745712 (1.846913)	5.143344 / 5.269862 (-0.126517)	2.764716 / 4.565676 (-1.800960)	0.142539 / 0.424275 (-0.281736)	0.015541 / 0.007607 (0.007934)	0.771407 / 0.226044 (0.545363)	7.631657 / 2.268929 (5.362728)	3.279684 / 55.444624 (-52.164940)	2.587566 / 6.876477 (-4.288911)	2.624622 / 2.142072 (0.482549)	1.427878 / 4.805227 (-3.377350)	0.257759 / 6.500664 (-6.242906)	0.078616 / 0.075469 (0.003147)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	1.609305 / 1.841788 (-0.232483)	18.258792 / 8.074308 (10.184484)	20.345242 / 10.191392 (10.153850)	0.267366 / 0.680424 (-0.413058)	0.047035 / 0.534201 (-0.487166)	0.568881 / 0.579283 (-0.010402)	0.662763 / 0.434364 (0.228399)	0.668927 / 0.540337 (0.128590)	0.755766 / 1.386936 (-0.631170)

PyArrow==latest

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.010017 / 0.011353 (-0.001336)	0.006816 / 0.011008 (-0.004192)	0.105038 / 0.038508 (0.066529)	0.038689 / 0.023109 (0.015580)	0.482113 / 0.275898 (0.206215)	0.540072 / 0.323480 (0.216592)	0.007738 / 0.007986 (-0.000248)	0.005134 / 0.004328 (0.000806)	0.102203 / 0.004250 (0.097953)	0.054080 / 0.037052 (0.017028)	0.501057 / 0.258489 (0.242568)	0.567186 / 0.293841 (0.273345)	0.060330 / 0.128546 (-0.068217)	0.020059 / 0.075646 (-0.055587)	0.123102 / 0.419271 (-0.296170)	0.063426 / 0.043533 (0.019893)	0.494171 / 0.255139 (0.239032)	0.538238 / 0.283200 (0.255039)	0.119613 / 0.141683 (-0.022069)	1.853728 / 1.452155 (0.401574)	1.984621 / 1.492716 (0.491904)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.282511 / 0.018006 (0.264505)	0.563190 / 0.000490 (0.562700)	0.000465 / 0.000200 (0.000265)	0.000086 / 0.000054 (0.000032)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.029267 / 0.037411 (-0.008144)	0.135618 / 0.014526 (0.121093)	0.146286 / 0.176557 (-0.030271)	0.188570 / 0.737135 (-0.548565)	0.155839 / 0.296338 (-0.140499)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.671660 / 0.215209 (0.456451)	6.718775 / 2.077655 (4.641120)	3.004601 / 1.504120 (1.500481)	2.640504 / 1.541195 (1.099309)	2.666788 / 1.468490 (1.198298)	1.242655 / 4.584777 (-3.342122)	5.780119 / 3.745712 (2.034407)	3.247935 / 5.269862 (-2.021927)	2.114007 / 4.565676 (-2.451669)	0.147546 / 0.424275 (-0.276729)	0.014408 / 0.007607 (0.006801)	0.824407 / 0.226044 (0.598362)	8.278185 / 2.268929 (6.009257)	3.733463 / 55.444624 (-51.711161)	2.976732 / 6.876477 (-3.899745)	3.132758 / 2.142072 (0.990686)	1.446095 / 4.805227 (-3.359132)	0.258628 / 6.500664 (-6.242036)	0.085513 / 0.075469 (0.010043)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	1.702681 / 1.841788 (-0.139106)	18.725123 / 8.074308 (10.650815)	19.622808 / 10.191392 (9.431416)	0.215845 / 0.680424 (-0.464579)	0.029246 / 0.534201 (-0.504955)	0.554819 / 0.579283 (-0.024464)	0.630926 / 0.434364 (0.196562)	0.637663 / 0.540337 (0.097325)	0.837948 / 1.386936 (-0.548988)

[github-actions]

Show benchmarks

PyArrow==6.0.0

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.008540 / 0.011353 (-0.002813)	0.004538 / 0.011008 (-0.006470)	0.101507 / 0.038508 (0.062999)	0.029751 / 0.023109 (0.006641)	0.292608 / 0.275898 (0.016710)	0.354734 / 0.323480 (0.031254)	0.007430 / 0.007986 (-0.000556)	0.003365 / 0.004328 (-0.000964)	0.078703 / 0.004250 (0.074452)	0.034858 / 0.037052 (-0.002194)	0.303518 / 0.258489 (0.045029)	0.336523 / 0.293841 (0.042682)	0.033741 / 0.128546 (-0.094805)	0.011460 / 0.075646 (-0.064186)	0.319551 / 0.419271 (-0.099721)	0.041102 / 0.043533 (-0.002431)	0.295914 / 0.255139 (0.040775)	0.322142 / 0.283200 (0.038943)	0.084694 / 0.141683 (-0.056989)	1.481308 / 1.452155 (0.029153)	1.530271 / 1.492716 (0.037554)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.180516 / 0.018006 (0.162510)	0.405741 / 0.000490 (0.405251)	0.002806 / 0.000200 (0.002606)	0.000072 / 0.000054 (0.000018)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.023359 / 0.037411 (-0.014052)	0.096950 / 0.014526 (0.082424)	0.103991 / 0.176557 (-0.072566)	0.143700 / 0.737135 (-0.593435)	0.106764 / 0.296338 (-0.189575)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.416966 / 0.215209 (0.201757)	4.145601 / 2.077655 (2.067946)	1.838258 / 1.504120 (0.334139)	1.629396 / 1.541195 (0.088201)	1.649707 / 1.468490 (0.181217)	0.689624 / 4.584777 (-3.895153)	3.414584 / 3.745712 (-0.331129)	1.874295 / 5.269862 (-3.395566)	1.251930 / 4.565676 (-3.313746)	0.081782 / 0.424275 (-0.342493)	0.012868 / 0.007607 (0.005261)	0.523904 / 0.226044 (0.297859)	5.251032 / 2.268929 (2.982104)	2.301549 / 55.444624 (-53.143075)	1.942110 / 6.876477 (-4.934367)	2.023014 / 2.142072 (-0.119058)	0.816492 / 4.805227 (-3.988736)	0.150107 / 6.500664 (-6.350558)	0.065118 / 0.075469 (-0.010351)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	1.226433 / 1.841788 (-0.615355)	13.852569 / 8.074308 (5.778261)	13.862779 / 10.191392 (3.671387)	0.146361 / 0.680424 (-0.534062)	0.028652 / 0.534201 (-0.505549)	0.398251 / 0.579283 (-0.181032)	0.403590 / 0.434364 (-0.030774)	0.492184 / 0.540337 (-0.048154)	0.581040 / 1.386936 (-0.805896)

PyArrow==latest

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.006859 / 0.011353 (-0.004494)	0.004632 / 0.011008 (-0.006376)	0.076653 / 0.038508 (0.038145)	0.027865 / 0.023109 (0.004755)	0.354472 / 0.275898 (0.078573)	0.385462 / 0.323480 (0.061982)	0.005125 / 0.007986 (-0.002861)	0.003420 / 0.004328 (-0.000909)	0.076018 / 0.004250 (0.071768)	0.040197 / 0.037052 (0.003144)	0.353675 / 0.258489 (0.095186)	0.394911 / 0.293841 (0.101070)	0.032909 / 0.128546 (-0.095637)	0.011713 / 0.075646 (-0.063933)	0.085921 / 0.419271 (-0.333350)	0.044462 / 0.043533 (0.000929)	0.349997 / 0.255139 (0.094858)	0.375207 / 0.283200 (0.092008)	0.091288 / 0.141683 (-0.050394)	1.536515 / 1.452155 (0.084361)	1.581878 / 1.492716 (0.089162)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.273284 / 0.018006 (0.255277)	0.424457 / 0.000490 (0.423967)	0.044659 / 0.000200 (0.044459)	0.000247 / 0.000054 (0.000192)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.025473 / 0.037411 (-0.011938)	0.100014 / 0.014526 (0.085488)	0.108551 / 0.176557 (-0.068006)	0.147913 / 0.737135 (-0.589223)	0.112729 / 0.296338 (-0.183610)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.448162 / 0.215209 (0.232953)	4.472701 / 2.077655 (2.395046)	2.078384 / 1.504120 (0.574264)	1.861292 / 1.541195 (0.320097)	1.920482 / 1.468490 (0.451991)	0.706968 / 4.584777 (-3.877809)	3.433109 / 3.745712 (-0.312603)	1.898684 / 5.269862 (-3.371178)	1.174375 / 4.565676 (-3.391302)	0.083666 / 0.424275 (-0.340609)	0.012388 / 0.007607 (0.004781)	0.546011 / 0.226044 (0.319966)	5.487514 / 2.268929 (3.218585)	2.534124 / 55.444624 (-52.910500)	2.168441 / 6.876477 (-4.708036)	2.203458 / 2.142072 (0.061386)	0.813333 / 4.805227 (-3.991894)	0.153169 / 6.500664 (-6.347495)	0.067151 / 0.075469 (-0.008318)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	1.277815 / 1.841788 (-0.563972)	13.920545 / 8.074308 (5.846237)	13.473801 / 10.191392 (3.282409)	0.129035 / 0.680424 (-0.551389)	0.016737 / 0.534201 (-0.517464)	0.388413 / 0.579283 (-0.190870)	0.388785 / 0.434364 (-0.045579)	0.481735 / 0.540337 (-0.058602)	0.576390 / 1.386936 (-0.810546)