Support query! for cargo-free systems

[kshramt]

Does your PR solve an issue?

Eagerly calling

sqlx/sqlx-macros-core/src/query/mod.rs

Line 80 in 929af41

pub fn workspace_root(&self) -> PathBuf {

results in build failure in Bazel builds.

Delete this text and add "fixes #(issue number)"

[kshramt]

To provide additional context, I am attempting to get the following code to work.

# BUILD.bazel
rust_binary(
    name = "app",
    srcs = glob(["src/**/*.rs"]),
    aliases = aliases(),
    compile_data = [
        "@//:.sqlx",
    ],
    proc_macro_deps = all_crate_deps(
        proc_macro = True,
    ),
    rustc_env = {
        "SQLX_OFFLINE_DIR": ".sqlx",
    },
    visibility = ["//visibility:public"],
    deps = all_crate_deps(),
)

[grim7reaper]

I can confirm that this fixes our build issue with the 0.7.3 as well (we're also working with Bazel).

BTW, how did you troubleshoot this one @kshramt ?
In our case the build failure wasn't really explicit (process_wrapper failed: error executing command with no error message nor stderr output whatsoever), did you use some magic Bazel flags to get more info?

[kshramt]

Thank you for confirming the resolution. As for your question about troubleshooting, I think I didn't do anything out of the ordinary. However, I did come across the error message "help: message: CARGO must be set: NotPresent." at https://github.com/kshramt/evidence_based_scheduling/actions/runs/7158836343/job/19491388204#step:5:207. I'm not entirely sure what made this error message visible in this instance.

[rbtcollins]

@grim7reaper Did you consider the patch I put up in #2922 ? I think they will be ~functionally equivalent, and I expect this patch would work for us too.

[sbernauer]

Believe it or not, I patched sqlx to be able to build it in Nix and ended up independently with a 100% identical patch 😅 Just found this PR before opening my own.
Changing it from eagerly to lazy seems like such an obvious and good addition, fixed all my problems 👍

[rbtcollins]

Thank you for merging this! I just realised that 2 years ago I did the first implementation in #1415 - do we need some sort of test to prevent this reverting again in future?

[NilsIrl]

Sorry if this is not the appropriate place to ask but I don't think there's many issues/threads with this many bazel and sqlx users so I'll shoot my shot.

What workflow do you use to update the offline dir? By default it only updates inside of the bazel sandbox so what I've done is just set SQLX_OFFLINE_DIR to /tmp but it's not very nice. Any advice?