Tricky rustdoc warning messages when using `include_str`

[shnewto]

I wanted to share something that made me scratch my head in case it makes anyone else scratch their head too. But I'm not sure I'd call this a bug or an error. It just stumped me for a minute and thought I'd bring it up and let y'all decide whether it's either.

(and in case it helps set the tone of my comments below, I love rustdoc, I love the include_str feature, and I'm grateful for all the work and care that goes into it all!)

When using #![doc = include_str!("../README.md")], if the contents of README.md generate warnings, it's not immediately clear what's to be done.

Here's a screenshot of output from cargo doc when the badge urls in my README.md were bare rather than atomatic urls.

I guess my misguided thought process went something like, "hmm it must be thinking about the path ../README.md as a url?" and I think that derailed me really seeing that the urls provided in the output were the root of the issue and not just "helpful" examples.

...I don't really know what "better" would look like either, I can imagine myself being similarly confused if the terminal output showed me lines from a .md file 😅

Thanks for taking a look! And if you decide that I'm just too easily confused, please feel free to close this issue 😄

[fmease]

This is definitely a bug. Ideally, the lint diagnostic would highlight the lines in the included file (here README.md).

Addressing other rustdoc devs: I can't tell how easy it would be to fix this without looking at the relevant code first. I remember that rustc does map included files to SourceFiles and thus assigns a Span distinct from the including file (which would allow a Diagnostic to point into the README.md) but there might be some peculiarities I don't know of. cc #95189 (comment).

[connortsui20]

I actually just ran into something similar, and it is similar enough that I won't make a new issue.

It's common to have code blocks in README.md, and I think that people can forget to annotate the code blocks with syntax highlighting. For example:

This is annotated with "sh".

This is not annotated with anything. (no "sh")

The difference is subtle, but for documentation tests this is pretty important. When you have #![doc = include_str!("../README.md")] and you run cargo test, it will run documentation tests on top of all the other tests. The issue here is that if the code block is not annotated, cargo will treat it as a documentation test.

Example:

// lib.rs
#![doc = include_str!("../README.md")]

And then in README.md

# Testing code block in README

A few words

```
This is going run as a test (and not compile)!
```

Running cargo doc is fine, but when I run cargo test, I get this error output:

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

   Doc-tests test_include_str_doc

running 1 test
test src\lib.rs - (line 5) ... FAILED

failures:

---- src\lib.rs - (line 5) stdout ----
error: expected one of `!`, `.`, `::`, `;`, `?`, `{`, `}`, or an operator, found `is`
 --> src\lib.rs:6:6
  |
3 | This is going to be run as a test!
  |      ^^ expected one of 8 possible tokens

error: aborting due to previous error

Couldn't compile the test.

failures:
    src\lib.rs - (line 5)

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.10s

error: doctest failed, to rerun pass `--doc`

I think there are a couple of things that are confusing here:

• The line where it failed doesn't actually exist, lib.rs should only have 2 lines
• The error message itself can be cryptic

In this case, it is pretty obvious where we want to find the bug. But when I ran into this issue the first time, I had this in my README.md:

tar -cvf something.tar somethingelse.txt ...

I had forgot to put the $ in front, and I was also calling this in my build script, so it was actually not immediately obvious where the source of this bug was (I actually thought it was something wrong with build.rs).

I think these issues would probably be solved by what @fmease suggested, highlighting the lines in the included file rather than in the file that brought it in. Though it seems like it might be non-trivial to implement this, maybe it could be a lint instead? Make sure that if a doc test is in an included file that it compiles immediately?

Thanks for all the great work!

[GuillaumeGomez]

I'll try to take a look in the coming weeks if someone didn't already as I've been bitten by this issue quite a few times already.

[fmease]

I'll try to take a look in the coming weeks

Nice, thanks! If you are short on time, I can squeeze this into my ever-growing TODO list, too ;)

[GuillaumeGomez]

For now I'll put into my "oh my god how can I have enough time to do all that????" list. 🤣