Add function to enumerate devices

[jlHertel]

Adds a basic DRM Listing functionality.
Please note that, different from libDRM, we don't process the devices to include bus information or anything else. We also don't de-duplicate devices like libDRM does, so duplicates might occur.

This change addresses #207

[notgull]

Could this instead return impl Iterator<Item = io::Result<DrmNode>>?

[jlHertel]

I'm not sure what would be the best approach here.
We can have IO Errors only by failing to open/read the /dev directory.
For the devices themselves, we will only get errors if they aren't DRM devices, which should be the case for the majority of stuff in /dev.

Would iterator improve anything in regards to the error handling?

Also, sorry if my answers are a bit confusing. I'm quite new to Rust, so I might be missing the big-picture on how to better handle errors/iterators.

[MarijnS95]

We can have IO Errors only by failing to open/read the /dev directory.

That could be handled with:

io::Result<impl Iterator<Item = io::Result<DrmNode>>>

For the devices themselves, we will only get errors if they aren't DRM devices, which should be the case for the majority of stuff in /dev.

And as just mentioned below, drmGetDevices2() skips errors while opening sub-devices, so we might as well return an iterator over DrmNode directly:

io::Result<impl Iterator<Item = DrmNode>>

For the devices themselves, we will only get errors if they aren't DRM devices, which should be the case for the majority of stuff in /dev.

And as below, on most platforms except FreeBSD we should be reading /dev/dri.

Also, sorry if my answers are a bit confusing. I'm quite new to Rust, so I might be missing the big-picture on how to better handle errors/iterators.

No worries, I think everyone is happy to educate new contributors on Rust as well as describing the "Rust style" used in drm-rs (as much as we're still discovering what that's supposed to be).

---

Regarding the drm_get_devices() function name (based on drmGetDevices(2)()), I'd drop the drm_ prefix since we're already in the drm crate (i.e. a fully qualified call would look like drm::node::drm_get_devices() which is somewhat superfluous) and then also drop the get_ prefix because of the no get_ prefix convention.

[jlHertel]

Thanks for all the feedback, it's highly appreciated.
I pushed the changes with everything you suggested. Do you mind to give another look?

[MarijnS95]

How about making this a .map().collect() rather than a side-effecting for_each()? If that, use regular control flow with for entry in fs::read_dir(..)? { .. }.

The .collect() can be dropped if using @notgull's suggestion to return an iterator instead.

[jlHertel]

Wouldn't we need to filter the results in some way if we use .map()? Asking as we check all devices and likely a lot of them aren't DRM devices, so we need to skip them. This validation is done right now by calling DrmNode::from_path().

[MarijnS95]

Such filtering can be implemented by a .filter_map(), but I'm hesitant to stat() every device on the system to see if it is a DRM device.

Since drmGetDevices2() was requested, I expected this feature to be built after that function. While it does also appear to ignore any error from processing individual devices, this function does open DRM_DIR_NAME which is /dev/dri on every platform except FreeBSD where it is /dev. This way we don't have to stat() devices that we know won't be DRM devices in the first place.

[jlHertel]

Hey thanks for the pointers. I somehow completely missed the fact that DRM_DIR_NAME was different in certain platforms. I will update the code to also use /dev/dri for all platforms except FreeBSD.