Asking for help with sqlx prepare as a derivation

[git-girl]

Hey :)

I am trying to set up a rust project and mostly everything works. I followed the tutorial and API docs to get a decent setup. I am building with sqlx however, which mean adding a preBuild and postBuild and also for the checks as a commonArgs. I really wasn't a big fan of the prospect of always running the migrations on every iteration, so I thought I'd add a derivation that takes the cargo sqlx prepare output add that as a dependency for the crane cargo calls, that would be quite a bit more efficient.
I ran into the issue of cargo-sqlx wanting to check cargo metadata (it runs it as a Command::new.args("").output()).
That wants to have network, but also fails when building with --no-sandbox.
When I add the cargo metadata that fails complaining about not getting any network.
I tried inspecting why the cargo sqlx, cargo metadata command fails, but I didnt find anything that is very conclusive. So most likely the same thing that just calling cargo metadata runs into.

I have a reproduction at https://github.com/git-girl/min_repro_metadata
nix run github:git-girl/min_repro_metadata

It would be really kind if you know how I could set this up to give me some tips.

I'm kinda out of ideas (except writing a custom sqlx command but I suffered last time I tried doing that for cargo watch and now I'm scared).
Since in this repro I'm not overriding the scope i don't think it's related to #353.
I didn't see how this could be related to nix stripping file stat data. Cargo handles that okay otherwise as well, afaik. It seems quite connected to the metadata cargo invocation and I wanted to get around that by providing artifacts to it which didn't work.
I also haven't checked a workaround like this one:
cargo +nightly metadata -Z unstable-options -Zconfig-include --config config.toml --offline
from nix-community/crate2nix#148
(cargo metadata would have to then be smart enough not to try and make the request even if invoked without those flags again by sqlx though)

Best Regards

[ipetkov]

Hi @git-girl ! I think what you might be after is craneLib.mkCargoDerivation. Basically this function will basically handle source vendoring for you which will satisfy the cargo metadata calls (which are otherwise lazy and will try to hit the network if the sources aren't already available).

You can swap pkgs.stdenv.mkDerivation for craneLib.mkCargoDerivation in your flake and you'll be almost there. mkCargoDerivation has a few additional required arguments, but if you are already defining your own build/check/install phases you can use some dummy values. Take a look at the docs for more context, and feel free to ask follow up questions if things don't make sense!

[git-girl]

Oh, yes thank you! That makes a lot of sense. I think I had a setup that had the sqlx prepare as a mkCargoDerivation and iirc it did work similar to my current working setup, with just start postgres all the time (currently I just pass it as commonArgs, but I did start by having it on the derivation). But I didn't think to use this for sqlx prepare. I think maybe the issue I'd have there again is getting rid of this behavior for some things, where I don't wan't this (cargoFmt) but also reuuse the craneLib instance, because caching. I'll try doing some cursed variable thing for the cargoDerivation. Or just accept that I'll need 2 craneLib instances? I will check it out in the next few days and will report back :)
Thanks!