config_dir() is inconsistent about type it returns

[Jelmer Vernooij]

depending on how it determines the config directory, bzrlib.config.config_dir() is inconsistent in what type it returns.

In some cases it returns a unicode string, and in some cases it returns a byte string. It would be nice if it was consistent, so callers could rely on it.

[Jelmer Vernooij (jelmer)]

On 13/08/11 17:40, Vincent Ladeuil wrote:
> ** Tags added: config
What would be the best thing to return in this case, a _fs_enc-encoded
bytestring?

Cheers,

Jelmer

[Vincent Ladeuil (vila)]

> What would be the best thing to return in this case, a _fs_enc-encoded bytestring?

An fs-decoded URL :)

Failing that, whatever the transport API is expecting (do we even need a bytestring ? Shouldn't a transport be enough ?) .

The config dir should probably not leaks anywhere else than to operations related to its creation or its content which should be handled via the transport API anyway.

[Vincent Ladeuil (vila)]

By the way, how did you find that config_dir does not always return the some type ? Do you have (by chance) tests reproducing the different cases ?

[Jelmer Vernooij (jelmer)]

On 08/13/2011 07:31 PM, Vincent Ladeuil wrote:
> By the way, how did you find that config_dir does not always return the
> some type ? Do you have (by chance) tests reproducing the different
> cases ?
bzr-svn asserts that it returns a bye string. Apparently, the only way
to not get a byte string back (on Linux) is to set the BZR_HOME
environment variable.

Cheers,

Jelmer

[Vincent Ladeuil (vila)]

> Apparently, the only way to not get a byte string back (on Linux) is to set the BZR_HOME environment variable.

Rats, related to my recent fix ?

[Jelmer Vernooij (jelmer)]

On 13/08/11 21:34, Vincent Ladeuil wrote:
>> Apparently, the only way to not get a byte string back (on Linux) is
> to set the BZR_HOME environment variable.
>
> Rats, related to my recent fix ?
Yeah, it looks like it :-/

Cheers,

Jelmer