Determine which prefix we should use

[pyca]

Determine which prefix we should use

[dstufft]

An important consideration, I'm pretty sure I selected $2a$ to maximize compatibility, particularly with py-bcrypt since $2y$ was custom to the C lib. However now that $2b$ is a "standard" prefix too, py-bcrypt might either support it as well, or it might be sane to say we want to pick theoretical security over comparability.

[sarahbx]

I'm currently implementing bcrypt password usage in a project and wanted to drop my two cents after reading through jazzyb's changes. The hashes in my system are accessed both by Python and PHP. Where $2y$ is the indicator that it is using the patched $2a$, should not $2y$ be the default output, with $2b$ or $2a$ as an option? Since I'm talking about compaitibility and just re-read your first post, I guess keeping $2a$ as default and either $2b$ or $2y$ as the options. Right now, $2a$ and $2y$(default output) are supported by the 'stable' PHP 5.5 library, where $2b$ is not. If $2b$ is algorithmically the same as $2y$, would it be better to stay compatible? And later updating the default to $2b$ when it is better supported?

php > $pwd = 'hello world';
php > $hasha = '$2a$12$CPeaPhxqTrBzJXP1KB1C/.t.2TJ7EzHhyCCmiLKAEKZovzmvdMwge';
php > $hashb = '$2b$12$CPeaPhxqTrBzJXP1KB1C/.t.2TJ7EzHhyCCmiLKAEKZovzmvdMwge';
php > $hashy = '$2y$12$CPeaPhxqTrBzJXP1KB1C/.t.2TJ7EzHhyCCmiLKAEKZovzmvdMwge';
php > var_dump(password_verify($pwd, $hasha));
bool(true)
php > var_dump(password_verify($pwd, $hashb));
bool(false)
php > var_dump(password_verify($pwd, $hashy));
bool(true)

[sarahbx]

I'm adding a check in my code to handle this issue if $2b$ becomes the default, but had I not seen this post, I would have had a lot of very unhappy users.