✨ Implement unix:dir and unix:tmpdir for --address

[jokeyrhyme]

No description provided.

[zeenix]

LGTM otherwise.

[zeenix]

LGTM. It's not being tested though. If you update the bus::default_address function to use a unix dir address (should be trivial) that would make sense anyway but also mean that this path is tested.

[jokeyrhyme]

I've changed default_address() and can confirm that cargo run causes a file like "/run/user/60222/dbus-3418628410" to exist on my system

[zeenix]

Nice! LGTM but I've one question as I wrote on the matrix channel:

I'm a bit curious though. The session bus config on my system does indeed have only one listen element set to listen>unix:tmpdir=/tmp however, the session bus is actually listening on /run/user/1000/bus

I wonder if that's because dbus-broker (the broker on my Fedora) is simply ignoring the listen directive in the config and defaulting to a hardcoded path for session. If so, I'd really want to follow their example. This would mean that I asked you to add the last commit for no good reason but in my defence, I didn't know about this discrepancy.

[zeenix]

I wonder if that's because dbus-broker (the broker on my Fedora) is simply ignoring the listen directive in the config and defaulting to a hardcoded path for session. If so, I'd really want to follow their example. This would mean that I asked you to add the last commit for no good reason but in my defence, I didn't know about this discrepancy.

Seems like when combined with #159, we end up still using the tmpdir from the config:

❯ cargo r -- --print-address
unix:tmpdir=/tmp,guid=3686ef2f3b7dd8236e7fc55c675b0474

We've 2 options:

1. Go the dbus-broker route completely on this and that means reverting this PR and ignoring the listen elements from the config in parse XML configuration files #159.
2. Forget about what dbus-broker is doing and re-add the removed commit to use tmpdir when no config is available/found.

Given that I can't really figure out what exactly dbus-broker is doing here (I searched though their issues but didn't find anything relevant), I'd vote for taking the simpler route of option 2 for now.

Moreover, the address being printed is not connectable and is server-side only address so this option can't be used like it's supposed to be. I'm not sure how you're testing the changes but the method I've been using to test changes, is to run DBUS_SESSION_BUS_ADDRESS=<address from busd> cargo t in zbus repo. Testing with any client would have shown this issue btw. 😉 We really need more tests.

Anyway, if you could fix these two issues in a new PR, that would be great.

[jokeyrhyme]

Ah, I wasn't using the --print-address at all, I was going directly to where I expected the socket file to be

[zeenix]

Ah, I wasn't using the --print-address at all, I was going directly to where I expected the socket file to be

Ah ok. I guess you were looking for the random socket file in the dir specified?

[jokeyrhyme]

Yeah, and that works
I just neglected to holistically test that along with the pre-existing functionality
I should be able to take a look at this over the weekend