rebis-dev: "[]" is not handled correctly

[triska]

Reviewing what I wrote in #1193 (comment), I noticed that rebis-dev in fact still contains a mistake in its handling of the aad/1 option!

Specifically, we currently get on rebis-dev, for the query:

?- Key = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
   Nonce = [0,0,0,0,0,0,0,0,0,0,0,0],
   crypto_data_encrypt("", 'chacha20-poly1305', Key, Nonce, _, [tag(Tag),aad("")]).

The result: Tag = [78,185,114,201,168,251,58,27,56,43,180,211,111,95,250,209], which is correct.

However, if we specify for the aad/1 option the string "[]", i.e., if we post:

?- Key = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
   Nonce = [0,0,0,0,0,0,0,0,0,0,0,0],
   crypto_data_encrypt("", 'chacha20-poly1305', Key, Nonce, _, [tag(Tag),aad("[]")]).

Then we get the same Tag as before, i.e., Tag = [78,185,114,201,168,251,58,27,56,43,180,211,111,95,250,209], which is incorrect! The correct result is: [129,92,253,207,2,86,106,239,60,201,211,254,151,176,184,80], which is what master yields.

It is as if the string "[]" is not taken into account at all to compute the authenticator in the rebis-dev branch. Note that already a small change to the authenticated data is expected to effect a huge change in the tag. For example, if we use the string "[" as additional data:

?- Key = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
   Nonce = [0,0,0,0,0,0,0,0,0,0,0,0],
   crypto_data_encrypt("", 'chacha20-poly1305', Key, Nonce, _, [tag(Tag),aad("[")]).

Then we get: Tag = [160,214,128,75,75,133,105,42,194,18,70,151,67,32,242,183], which is correct, and completely different from the tag we get for "" even though only a single character was added.

[triska]

I think this is a more general problem with how rebis-dev treats the string "[]", maybe due to an issue in the function value_to_str_like.

For example, I get with rebis-dev:

$ touch "[]"
$ scryer-prolog
?- use_module(library(files)).
   true.
?- file_exists("[]").
false.    % expected: true

master correctly yields true.

[triska]

This may be one of the simplest queries to test the issue:

?- Key = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
   Nonce = [0,0,0,0,0,0,0,0,0,0,0,0],
   crypto_data_encrypt("[]", 'chacha20-poly1305', Key, Nonce, "", []).

Since this unexpectedly succeeds, we see that the string "[]" is erroneously "encrypted" to "", i.e., the empty list.

[triska]

An even shorter query to test this:

?- crypto_data_hash("", A, []),
   crypto_data_hash("[]", B, []),
   dif(A, B).

This query is expected to succeed.

I have filed #1222 in an attempt to address this issue.

[triska]

This is now resolved, thank you a lot!