Ignore spacing chars for base64decode

[Moelf]

fixes #32397

[JeffBezanson]

This implementation is not ideal since it copies the data. Would be best to handle it during decoding. I also believe it is only necessary to handle ASCII space and newline.

[Moelf]

So maybe do a || (space) || (newline) somwhere in here?

julia/stdlib/Base64/src/decode.jl

Line 67 in b56a9f0

if b1 < 0x40 && b2 < 0x40 && b3 < 0x40 && b4 < 0x40 && p + 2 < p_end

[bicycle1885]

Hi, I wrote the original code of this implementation to improve the performance based on my package (https://github.com/bicycle1885/CodecBase.jl). The implementation in this package ignores whitespace characters so it may help here. Otherwise, I can fix the issue instead of you if you are not in a rush.

[Moelf]

Thanks, I think I can pull this one off, Of course it would be easier if you can give me some hint on which part of the code is used to ignore invalid characters when decoding.

[bicycle1885]

I don't remember the code well, but I think you can modify the decoding behavior just by modifying its look-up table.

[Moelf]

I think that was the case in your CodecBase but not in Julia's code base.

[JeffBezanson]

Will need a test of course.

[Moelf]

Fun fact, python does this to handle invalid characters:
https://github.com/python/cpython/blob/5150d327924959639215ed0a78feffc0d88258da/Lib/base64.py#L85

[Moelf]

@JeffBezanson I have the test covered, and I don't think it's worth to do in the decoding first because there are copying and allocation anyways (if we need to check if a char is valid or not), second because I don't think people pass ~GB of data into base64decode (bonus, regular expression is okally fast?)

[JeffBezanson]

I disagree, and I also realized another problem which is that a user can directly call readbytes!(Base64DecodePipe(data), vector) where the same issue needs to be handled, plus vector is pre-allocated so no allocation necessarily occurs.