rustdoc --test only detects rust vs. markdown based on filename

[lambda]

If I run rustdoc --test on something that doesn't have a file extension, it interprets it as Rust code. This makes it inconvenient to run on something piped in, say via Bash process redirection:

$ rustdoc --test <(curl http://words.steveklabnik.com/structure-literals-vs-constructors-in-rust | pandoc -f html -t markdown)
/dev/fd/63:1:1: 1:2 error: unknown start of token: \u{a0}
/dev/fd/63:1
             ^
/dev/fd/63:1:1: 1:2 help: unicode character ' ' (No-Break Space) looks much like ' ' (Space), but it's not
/dev/fd/63:1
             ^
thread '<unnamed>' panicked at 'Box<Any>', ../src/libsyntax/parse/lexer/mod.rs:198
note: Run with `RUST_BACKTRACE=1` for a backtrace.

$ curl http://words.steveklabnik.com/structure-literals-vs-constructors-in-rust | pandoc -f html -t markdown > test.md
$ rustdoc --test test.md

running 9 tests
test _0 ... FAILED
# etc ...

There isn't any way to override this, by providing a --markdown flag or the like.

Also, for this use case, would be nice if you could provide input on stdin instead of via bash process redirection.

[Mark-Simulacrum]

This appears to work today. I think it could use a run-make test, so E-needstest.

running 1 test
test /dev/fd/63 - main (line 1) ... FAILED

failures:

---- /dev/fd/63 - main (line 1) stdout ----
	thread 'rustc' panicked at 'test executable failed:

thread 'main' panicked at 'assertion failed: `(left == right)` (left: `10`, right: `5`)', <anon>:2
note: Run with `RUST_BACKTRACE=1` for a backtrace.

', /checkout/src/librustdoc/test.rs:313
note: Run with `RUST_BACKTRACE=1` for a backtrace.


failures:
    /dev/fd/63 - main (line 1)

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured