Usage of `extern crate` fails doc tests

[Osspial]

If a doc test attempts to import any types declared in it's own crate, including the phrase extern crate anywhere in the code will cause the test to not compile, throwing a "unresolved import" error at the attempted import. For example, if we have struct Foo in crate bar, with the crate baz declared under dev_dependencies, these tests will all fail to compile when there's no reason for them not to:

/// ```
/// extern crate baz;
/// use bar::Foo;
/// ```
pub struct Foo{}

/// ```
/// // The phrase "extern crate" breaks this!
/// use bar::Foo;
/// ```
pub struct Foo{}

/// ```
/// use bar::Foo;
/// let string = "extern crate";
/// ```
pub struct Foo{}

The following error log is given by cargo test:

running 1 test
test Foo_0 ... FAILED

failures:

---- Foo_0 stdout ----
        <anon>:2:9: 2:12 error: unresolved import `bar::Foo`. Maybe a missing `extern crate bar`? [E0432]
<anon>:2     use bar::Foo;
                 ^~~
error: aborting due to previous error(s)
thread 'Foo_0' panicked at 'Box<Any>', ../src/librustc\session/mod.rs:153
stack backtrace:
   0:         0x61cc34dc - strncmp
   1:         0x61cc2b5c - strncmp
   2:         0x61c84d6f - strncmp
   3:         0x67f193ba - strncmp
   4:         0x67f96c9b - strncmp
   5:         0x6ca6943d - strncmp
   6:         0x6ca6e301 - strncmp
   7:         0x661e3803 - strncmp
   8:         0x61caf9ba - strncmp
   9:         0x661e3bba - strncmp
  10:         0x61cc1209 - strncmp
  11:     0x7fff457f8101 - strncmp


failures:
    Foo_0

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

Rust version:

rustc 1.8.0 (db2939409 2016-04-11)
binary: rustc
commit-hash: db2939409db26ab4904372c82492cd3488e4c44e
commit-date: 2016-04-11
host: x86_64-pc-windows-gnu
release: 1.8.0

[durka]

This is by design. It's a confusing design. The code in doc tests is (1) wrapped in fn main (unless main is already in there), and (2) the extern crate is put outside that fn (unless it's already in there). So putting in extern crate but not fn main causes it to skip step 2 but not step 1, making the imports unresolvable. The rules are described in the book.

[Osspial]

OK, I can see why it would be done that way. However, there are still two issues:

1. The error message doesn't indicate the actual problem
2. Examples 2/3 should still compile, as extern crate isn't an actual bit of code

It looks like rustdoc is just doing a blind search for the phrase extern crate in the code, which is causing some issues that shouldn't be present.

EDIT:
Additionally, the phrase "remainder of the text" in rule 4 seems to imply that everything after extern crate is supposed to be implicitly wrapped by an automatically generated fn main.

[durka]

Yeah, it just does a blind substring search. Possibly rustdoc could look
for the "unresolved import" error and give an extra hint?

On Thu, May 19, 2016 at 12:02 PM, Osspial notifications@github.com wrote:

OK, I can see why it would be done that way. However, there are still two
issues:

1. The error message doesn't indicate the actual problem
2. Examples 2/3 should still compile, as extern crate isn't an actual
   bit of code

It looks like rustdoc is just doing a blind search for the phrase extern
crate in the code, which is causing some issues that shouldn't be present.

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#33731 (comment)

[QuietMisdreavus]

Update: Still an issue. The offending code still does a naive string search before inserting extern crate my_crate; into the doctest:

rust/src/librustdoc/test.rs

Lines 403 to 410 in ae2f299

	if !s.contains("extern crate") && !opts.no_crate_inject && cratename != Some("std") {
	if let Some(cratename) = cratename {
	if s.contains(cratename) {
	prog.push_str(&format!("extern crate {};\n", cratename));
	line_offset += 1;
	}
	}
	}

A proper fix should scan the lines of the original string much like the way it checks for fn main:

rust/src/librustdoc/test.rs

Lines 414 to 423 in ae2f299

	let already_has_main = s.lines()
	.map(|line| {
	let comment = line.find("//");
	if let Some(comment_begins) = comment {
	&line[0..comment_begins]
	} else {
	line
	}
	})
	.any(|code| code.contains("fn main"));

Or even better, actually parse the code and look for the import that way. It would also fully fix #21299, since it has the same problem: the last example will fail if you replace "extern crate" with "fn main".

[Enet4]

Is this issue still reproducible? I stumbled upon a similar situation, but the doctest below seems to compile just fine since 1.26.0 (it fails in 1.25.0 and older).

//! # Examples
//!
//! ```
//! extern crate byteorder;
//! use byteorder::ReadBytesExt;
//! ```

[QuietMisdreavus]

That doctest doesn't exercise the issue. The problem is that you can't use any types from your own crate in there, right. (Unless your crate is byteorder, but then you're doing everything correctly.) Rustdoc still won't add extern crate my_crate; to the doctest.