On Thu, Apr 19, 2012 at 08:45:22PM +0200, Erik Faye-Lund wrote:
> Here's an updated version of my Windows credential-helper.
>
> The most important difference is that it doesn't suck as much as
> it used to ;)
>
> Basically, I'm now using the attribute-system to store the meta-data
> for the credentials.
>
> This passes the test for me, and seems to work OK.
This looks much better. As usual, I can't comment on the Windows-y
parts, but the credential logic looks good. One minor comment:
> +int main(int argc, char *argv[])
> +{
> [...]
> + if (!protocol || !host)
> + return 0;
You could have a protocol that does not have a "host" component. Git
will produce this for a cached certificate password (which will then
always have a "path" field). I don't know if anybody is using it to
cache certificate passwords, and I admit that it is not very well tested
by me, either. So I'm not sure anybody will actually care.
The credential-store helper uses this logic:
$ sed -n '70,79p' credential-store.c
/*
* Sanity check that what we are storing is actually sensible.
* In particular, we can't make a URL without a protocol field.
* Without either a host or pathname (depending on the scheme),
* we have no primary key. And without a username and password,
* we are not actually storing a credential.
*/
if (!c->protocol || !(c->host || c->path) ||
!c->username || !c->password)
return;
I'm pretty sure the OS X helper does not handle certificate passwords at
all. It is harder there, because I map directly to native protocol
types, and I haven't checked to see whether OS X handles this type. But
since you are just storing whatever protocol information git hands you,
I think for you it is just a matter of letting it through (and handling
the "!host" case when writing).
-Peff
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html