It seems to be a general issue (maybe more so when using wayland) that upgrades of nixpkgs can break things. The issues I myself encountered most (being on nixpkgs=nixos-unstable and hyprland) are occasional/frequent issues with screen sharing not working and vscode crashes. From most online reports I suspect that these issues are related to mismatches in GPU accelerated (*GL) code in user programs vs. the mesa version, which both are often fast-moving targets (cases in point chrome rendering engine (incl. electron), wlroots).
Recently, when trying an update of X-Plane 12, it gave me a core dump with ERROR elf_dynamic_array_reader.h:64] tag not found, which also appears to be somehow mesa related. (I’m running this in an FHS shell as a “declarative replacement” of using steam-run, which normally appears to behave the same).
While researching that, I found that I need a newer mesa version, which prompted me to consider to just get a new mesa for my whole system via an overlay.
But since this is my daily driver and I can’t afford “too much instability”, I wanted to get a general idea how complex the relationship between mesa and the OS installation actually is.
Q1: is the above alleged dep complexity the main reason that the mesa version in nixpkgs is updated less frequently, because it is difficult to have a version that “works ok with most packages” and because so many packages depend on it?
Q2: In what way is the amdgpu kernel driver related to / used by mesa and should I consider the kernel-version vs. mesa-version when testing?
Q3: Suppose testing a few indicative programs (Hyprland, vscode, screensharing (pipewire ↔ wlroots?), an odd game or two), could this be considered an indication that the used mesa version is “generally ok”, or is there still a high chance of unpleasant surprises?