Finished substitutions erroneously shown as in progress

[Atemu]

This happened during a remote store + remote build (--store ssh-ng://machine where the store machine has remote builders configured).

┃                      │     │                             ┌─ ↓ ⏵ ncompress-5.0 from [1] ⏱ 22m46s
┃                      │     │                          ┌─ ↓ ✔ libXpm-3.5.17 from [1]
┃                      │     │                       ┌─ ✔ gd-2.3.3 on [5] ⏱ 14s
┃                      │     │                       │  ┌─ ↓ ⏵ pango-1.52.2.tar.xz from [1] ⏱ 22m40s
┃                      │     │                       ├─ ✔ pango-1.52.2 on [4] ⏱ 39s
┃                      │     │                    ┌─ ✔ graphviz-10.0.1 on [4] ⏱ 2m9s

nom says it's substituting some paths whose dependents have already finished building which obviously can't be correct.

Speculation:

This might be due to the fact that the path was already present on the remote (but not the local machine nom runs on) and therefore isn't considered?

[maralorn]

thx for the report. I witness something like that regularly myself. Not sure yet what exactly is going on there.

[Atemu]

As a workaround, you could at least consider a build/substitute to be done when a dependent starts realising.

[maralorn]

I hope we can find a better solution.

[me-and]

I've seen this behaviour when I've been using local builds and stores. I've not (yet) managed to characterise it or produce a minimal repro, but I am able to reproduce it fairly regularly.