Documentation

Head of the padded base32 encoding loop.

This function takes an alphabet in the form of an unboxed Addr#, allocates the correct number of bytes that will be written, and executes the inner encoding loop against that data.

Head of the unpadded base32 encoding loop.

This function takes an alphabet in the form of an unboxed Addr#, allocates the correct number of bytes that will be written, and executes the inner encoding loop against that data.

Head of the base32 decoding loop.

This function takes a base32-decoding lookup table and base32-encoded bytestring, allocates the correct number of bytes that will be written, and executes the inner decoding loop against that data.

Validate a base32-encoded bytestring against some alphabet.

This function checks that the last N-chars of a bytestring are '=' and, if true, fails with a message or completes some io action.

This is necessary to check when decoding permissively (i.e. filling in padding chars). Consider the following 8 cases of a string of length l:

• l = 0 mod 8: No pad chars are added, since the input is assumed to be good.
• l = 1 mod 8: Never an admissible length in base32
• l = 2 mod 8: 6 padding chars are added. If padding chars are present in the string, they will fail as to decode as final quanta
• l = 3 mod 8: Never an admissible length in base32
• l = 4 mod 8: 4 padding chars are added. If 2 padding chars are present in the string this can be "completed" in the sense that it now acts like a string `l == 2 mod 8` with 6 padding chars, and could potentially form corrupted data.
• l = 5 mod 8: 3 padding chars are added. If 3 padding chars are present in the string, this could form corrupted data like in the previous case.
• l = 6 mod 8: Never an admissible length in base32
• l = 7 mod 8: 1 padding char is added. If 5 padding chars are present in the string, this could form corrupted data like the previous cases.

Hence, permissive decodes should only fill in padding chars when it makes sense to add them. That is, if an input is degenerate, it should never succeed when we add padding chars. We need the following invariant to hold:

  B32.decodeUnpadded | B32.decodePadded ~ B32.decode