#[pg_test(expected = ...)] corrupts raw-string literals and panics on malformed input

[isdaniel]

Summary

The #[pg_test] attribute parses its expected = "..." argument with a hand-rolled TokenTree walker instead of syn. This causes bug:

Raw-string corruption — expected values written as raw strings (needed whenever the expected Postgres error message contains a ") are silently mangled, so the test never matches the real error.

pre-prepare

/// Raises an error whose message contains a literal double-quote: foo "bar"
#[pg_extern]
fn raise_quoted_error() -> &'static str {
    error!(r#"foo "bar""#);
}

Bug — raw-string corruption

#[pg_test(error = r#"foo "bar""#)]
fn bug1_raw_string_expected_is_corrupted() {
	Spi::get_one::<&str>("SELECT raise_quoted_error()").unwrap();
}

Result:

Client Error:
foo "bar"
postgres location: lib.rs:8
rust location: <unknown>


note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

failures:
    tests::pg_bug1_raw_string_expected_is_corrupted

The real error message from raise_quoted_error() is foo "bar", so the test fails with a mismatch even though the author wrote the correct expected string.

Environment

• pgrx: 0.18.1
• OS: Linux

[isdaniel]

fix via #2317