There are very few tests for -Z unpretty expansion

[brson]

See #23616 (comment) for a current issue description. The original description follows below.

---

Pretty-expanded source that results in unstable code can't build. After it becomes illegal to not declare features this is going to cause a bunch of expanded source to not build.

[brson]

I'm going to be adding no-pretty-expanded to a bunch of tests.

[steveklabnik]

at least expanded is unstable, yeah?

[alexcrichton]

Unfortunately I think that this is just a fundamental limitation. We have a number of problems that make it such that expanded source cannot actually compile at all:

• #[allow_internal_unstable] make it such that macros can use unstable features without declaration
• #[no_std] which is injected is unstable
• Hygiene (e.g. a macro's local variable foo is very different than the outer variable foo)

I would also be fine just assuming that tests cannot be pretty printed when expected unless there is an explicit opt-in directive.

[brson]

Perhaps I will switch compiletest to support pretty-expanded and add that to a bunch that are passing now. It does seem like it would be pretty annoying to maintain all the no-pretty-expanded flags.

[nikomatsakis]

can we just disable the pretty-printing tests for the beta branch or something like that?

[nikomatsakis]

(Does that make sense?)

[steveklabnik]

Triage: not aware of any specific issue here, though I haven't heard about anyone running into this in practice in the last year.

[steveklabnik]

Triage: lots of FIXMEs referring to this issue still remain.

[jethrogb]

Ran into this while looking at the compiletest framework and was thoroughly confused. As far as I can tell, this is the situation as of early 2019:

Certain parts of the Rust test suite can be run in “pretty” mode. This will test the pretty-printing output of rustc. One of the pretty-printing outputs of rustc is the expanded mode, which may result in code that can't subsequently be compiled, that is what this issue is about. Tests that were in 2015 known to work with expanded mode have the pretty-expanded attribute to test that mode. All pretty-expanded annotations are marked with “FIXME #23616”, but this is misleading: the tests that have the attribute don't need to be fixed, it's the tests that don't have the pretty-expanded/FIXME annotation that are blocked on this issue.

I think it was a mistake to replace no-pretty-expanded with pretty-expanded in #23598, because the test coverage of expanded pretty printing has basically stagnated. Since the attribute is off by default and things just work if you don't test it, people have not been adding the pretty-expanded annotation to new tests even if it would work. Now, everyone has basically forgotten about this attribute.