| 5 May 2023 |
 @rnhmjoj:maxwell.ydns.eu | xserver.conf
I think this is due to find order being essentially undefined behavior in nixos/modules/services/x11/xserver.nix. it needs a | sort to be deterministic. | 16:55:05 |
 @trofi:matrix.org | Yeah. I think find does filesystem order (and each does it's own ordering). | 16:57:48 |
 delroth | In reply to @raboof:matrix.org
finally hacked in a way to get a report of just the runtime dependencies, https://reproducible.nixos.org/nixos-iso-minimal-runtime/ - no surprises there, the Linux BPF/BTF problem and Python remaining. Hoping to do the graphical iso later. I worry a bit that this might be trading off false positives for false negatives. Let's say I have a derivation "drv" which does "cp ${otherdrv}/foo $out/foo" (unlikely, but in practice this could be e.g. gzip, or static linking, or some other kind of bundling). As far as I understand now the report wouldn't catch unreproducibility of "otherdrv", and it would happily say that "drv" is reproducible, which is irrelevant (since its "non-runtime dependencies" are what matters). | 20:11:29 |
| delroth | (Does that happen in practice in nixpkgs? I don't know, I wouldn't be surprised if it did. Is it a major issue? No clue either.) | 20:11:54 |
 raitobezarius | (Yes, it happens through trivial builders sometimes I suppose) | 20:15:21 |
| raitobezarius | (Though I'm okay if we keep both pieces ?) | 20:15:33 |
| delroth | By "both pieces" here do you mean "both types of report, runtime-only and all"? As long as people understand that the "runtime dependencies" report actually might exclude some runtime dependencies, then sure :) | 20:20:40 |
| delroth | * By "both pieces" here do you mean "both types of report, runtime-only and all"? As long as people understand that the "runtime dependencies" report actually might accidentally miss some unreproducible runtime dependencies, then sure :) | 20:21:31 |
| raitobezarius | correct | 20:21:38 |
| raitobezarius | I guess we can probably engineer correctly the pages to reflect that | 20:21:51 |
| delroth | I have experience with teams accidentally carving out a metric that doesn't actually cover the whole problem they want to solve, get the metric to the desired goal, then declare victory too early :) it's a pretty common pattern | 20:22:25 |
| delroth | I actually wonder how the diff would look like between this definition of "runtime dependencies" and the definition of "runtime dependencies" you'd get by doing a cross-build of the same target and filtering by build vs. host system triple | 20:23:43 |
| delroth | I suspect "different", but I don't know in which direction (better or worse), and I don't think it's practical or a useful metric either because it's not like the ISOs being shipped are built this way | 20:24:26 |
