| 6 Nov 2023 |
 @atemu12:matrix.org | So something occured to me. If mesa doesn't like mixing different versions between compositor and app; how the hell does flatpak work? | 11:22:31 |
 K900 | In reply to @atemu12:matrix.org
So something occured to me. If mesa doesn't like mixing different versions between compositor and app; how the hell does flatpak work? They have their own Mesa | 12:49:07 |
| @atemu12:matrix.org | K900 (drm/amd#1417 when): Yeah, that's what I mean. That Mesa is different from the system Mesa which the compositor is running with and that should cause issues with libdrm, shouldn't it? | 13:51:18 |
| K900 | No | 13:51:26 |
| @atemu12:matrix.org | Why not? | 13:51:44 |
| K900 | You just need to match libgbm and mesa | 13:51:46 |
| K900 | You don't need to match mesa to other mesa | 13:51:54 |
| @atemu12:matrix.org | Ah sorry, confused them | 13:52:02 |
| @atemu12:matrix.org | But the same still applies | 13:52:07 |
| @atemu12:matrix.org | (Also isn't libgbm == mesa?) | 13:52:19 |
| K900 | libgbm is part of Mesa | 13:53:43 |
| K900 | But the NixOS compositor issues stem from the fact that we load the Mesa drivers impurely | 13:53:54 |
| K900 | And the drivers and libgbm need to match | 13:54:06 |
| @atemu12:matrix.org | But isn't that the same for flatpak on any distro ever? | 13:54:34 |
| @atemu12:matrix.org | The distro runs the compositor with mesa version x. Apps in flatpak are linked against libgbm a b and c | 13:55:33 |
| K900 | It doesn't matter | 13:55:53 |
| @atemu12:matrix.org | Oh, they link against mesa a b and c at runtime because that's what's in the sandbox.. | 13:55:55 |
| K900 | As long as they're the same within one app | 13:55:59 |
| @atemu12:matrix.org | So compositor mesa/libgbm doesn't matter | 13:56:17 |
| K900 | Yes | 13:59:20 |
| @atemu12:matrix.org | So, wouldn't a good solution to this be to propagate the version of mesa an app was linked against and use that rather than the impure driver? | 14:05:55 |
| @atemu12:matrix.org | Why do we have the impure driver anyways for mesa? | 14:06:12 |
| @atemu12:matrix.org | (Other than for running impure apps of course.) | 14:06:27 |
| K900 | 1) it can be Nvidia and not Mesa | 14:06:48 |
| K900 | 2) forward compatibility | 14:06:51 |
| K900 | I think 2) was an accident | 14:06:59 |
| K900 | But it's the reason I don't agree with "let's just shove mesa into the closure for everything" | 14:07:13 |
| K900 | Because we could technically just load Nvidia impurely now | 14:08:08 |
| K900 | But I actually think the impure Mesa is important, because it allows people to run older nixpkgs stuff with recent hardware | 14:08:35 |
| @atemu12:matrix.org | Yeah, yeah right | 14:08:50 |
