Allow to parse opam 2.1 switch import files containing extra-files

[kit-ty-kate]

Fixes #5941

opam 2.1 used the Base64 module from extlib which does not seem to pad encoded outputs. In contrast, opam 2.2 uses the Base64 module from the base64 library (after extlib's Base64 module was removed in extlib.1.7.8) which by default pads outputs and requires padded inputs when decoding them.

cc-ing base64 implementations maintainers @dinosaure @ygrek

• Question for @dinosaure: is this fix fine? Should Base64.encode be also called with ~pad:false? What are the pitfalls of disabling padding?

• Question for @ygrek: Just to double-check, am i correct in assuming the output of extlib's base64 implementation is always not padded?

[dinosaure]

As far as Base64.decode_exn is concerned, the default setting is that values should have a pad as if there were a '\000' for a C string. Of course, in reality, this is not necessarily the case. The pad becomes useful if you concatenate base64 values and want to decode them cleanly without there being a mismatch between the values (since you have this pad as a delimiter).

If the base64 value exists on its own (or is delimited by something else such as "), the pad is not necessary. What Base64.decode_exn tries to do is check for the existence of this pad by default, but if you are manipulating base64 values where the pad is not needed, pad:false is indeed advisable.

In other words, pad:true is for a paranoid mode of retrieving values.

[kit-ty-kate]

Thanks! In that case I will use this option for the encoder as well.
I'll also add some tests in a minute

[rjbou]

Tiny comment in the test (see commit), otherwise ltgm! Thanks!