Pass edition flags to compiler from rustdoc as expected

[GuillaumeGomez]

Fixes #52357.

r? @QuietMisdreavus

[QuietMisdreavus]

I'm surprised that setting the edition in the session options didn't work:

rust/src/librustdoc/test.rs

Lines 214 to 233 in 3d5753f

	let sessopts = config::Options {
	maybe_sysroot: maybe_sysroot.or_else(
	|| Some(env::current_exe().unwrap().parent().unwrap().parent().unwrap().to_path_buf())),
	search_paths: libs,
	crate_types: vec![config::CrateTypeExecutable],
	output_types: outputs,
	externs,
	cg: config::CodegenOptions {
	prefer_dynamic: true,
	linker,
	..cg
	},
	test: as_test_harness,
	unstable_features: UnstableFeatures::from_environment(),
	debugging_opts: config::DebuggingOptions {
	..config::basic_debugging_options()
	},
	edition,
	..config::basic_options().clone()
	};

What does this extra call do that that doesn't? Do we need to add this to our compilation runs for building docs?

[GuillaumeGomez]

It sets the syntax parser edition. The important here being that it's done inside rustc_driver::in_rustc_thread.

[QuietMisdreavus]

Right, but those session options are created before any parsing or compilation work is done. run_test is called (well, a closure created that calls it) inside that same in_rustc_thread. If the test runner is taking that closure and running it inside a different context, then there may be another problem at hand.

Either way, if it didn't work before and this makes it work, then i won't fight it. I'm just concerned that there seems to be more than one way to set the edition, and the most obvious of them doesn't work.

@bors r+ rollup

[bors]

📌 Commit 65e6b2b has been approved by QuietMisdreavus