Fix ICE when rustdoc encounters certain usages of HRTBs
Fixes#51236
Under certain circumstances, `AutoTraitFinder` could end up computing a `ParamEnv` involving two trait predicates that differed only in the region parameters involved. One of these parameters would be a HRTB, while the other would be a normal region parameter.
When this `ParamEnv` was later passed to `SelectionContext`, an `Ambiguity` error would occur, since the erased versions of these predicates would be identical. To solve the issue, we de-duplicate our list of predicates as we build it up. Whenever we encounter two predicates that differ only in their assignment of region parameters (a HRTB vs a normal lifetime parameter), we pick the HRTB. This corresponds to selecting a 'stricter' bound to display in the generated documentation: we're requiring that a particular type works for all possible lifetime parameters if it's going to implement a particular auto trait.
Fix NLL migration mode so that reports region errors when necessary.
The code here was trying to be clever, and say "lets not report diagnostics when we 'know' NLL will report an error about them in the future."
The problem is that in migration mode, when no error was reported here, the NLL error that we "knew" was coming was downgraded to a warning (!).
Thus causing #53026
(I hope it is the only instance of such a scenario, but we will see.)
Anyway, this PR fixes that by only doing the "clever" skipping of region error reporting when we are not in migration mode. As noted in the FIXME, I'm not really thrilled with this band-aid, but it is small enough to be back-ported easily if that is necessary.
Rather than make a separate test for issue 53026, I just took the test that uncovered this in a first place, and extended it (via our revisions system) to explicitly show all three modes in action: AST-borrowck, NLL, and NLL migration mode.
(To be honest I hope not to have to add such revisions to many tests. Instead I hope to adopt some sort of new `compare-mode` for either borrowck=migrate or for the 2018 edition as a whole.)
Fix#53026
dead-code lint: say "constructed" for structs
Respectively.
This is a sequel to November 2017's #46103 / 1a9dc2e9. It had been
reported (more than once—at least #19140, #44083, and #44565) that the
"never used" language was confusing for enum variants that were "used"
as match patterns, so the wording was changed to say never "constructed"
specifically for enum variants. More recently, the same issue was raised
for structs (#52325). It seems consistent to say "constructed" here,
too, for the same reasons.
~~While we're here, we can also use more specific word "called" for unused
functions and methods. (We declined to do this in #46103, but the
rationale given in the commit message doesn't actually make sense.)~~
This resolves#52325.
make `everybody_loops` preserve item declarations
First half of https://github.com/rust-lang/rust/issues/52545.
`everybody_loops` is used by rustdoc to ensure we don't contain erroneous references to platform APIs if one of its uses is pulled in by `#[doc(cfg)]`. However, you can also implement traits for public types inside of functions. This is used by Diesel (probably others, but they were the example that was reported) to get around a recent macro hygiene fix, which has caused their crate to fail to document. While this won't make the traits show up in documentation (that step comes later), it will at least allow files to be generated.
rustdoc: refactor how passes are structured, and turn intra-doc-link collection into a pass
This builds on https://github.com/rust-lang/rust/pull/52751 and should not be merged until that finally finishes the bors queue
Part 2 of my passes refactor. This introduces the concept of an "early pass", which is run right before exiting the compiler context. This is important for passes that may want to ask the compiler about things. For example, i took out the intra-doc-link collection and turned it into a early pass. I also made the `strip-hidden`, `strip-private` and `strip-priv-imports` passes occur as early passes, so that intra-doc-link collection wouldn't run on docs that weren't getting printed.
Fixes https://github.com/rust-lang/rust/issues/51684, technically https://github.com/rust-lang/rust/issues/51468 too but that version of `h2` hits a legit intra-link error after that `>_>`
r? @rust-lang/rustdoc
[NLL] Use smaller spans for errors involving closure captures
Closes#51170Closes#46599
Error messages involving closures now point to the captured variable/closure args.
r? @pnkfelix
avoid computing liveness for locals that escape into statics
Fixes#52713
I poked at this on the plane and I think it's working -- but I want to do a bit more investigation and double check. The idea is to identify those local variables where the entire value will "escape" into the return -- for them, we don't need to compute liveness, since we know that the outlives relations from the return type will force those regions to be equal to free regions. This should help with html5ever in particular.
- [x] test performance
- [x] verify correctness
- [x] add comments
r? @pnkfelix
cc @lqd
Make left column of rustdoc search results narrower
To make more room for the description of the item
The description often has useful text that helps disambiguate between search results, but very little of it is shown.
As a side effect, this breaks the alignment between the search results and the "In Return Types" tab, which tends to line up above the description-- up until I started investigating this, I thought "In Names"/"In Parameters"/"In Return Types" were column headers and I just never saw search results that had info for the "In Parameters" middle column! Now, with the two columns of search results each taking up about a half, they look more like tabs than column headers.
Types that are long still wrap and look good-- I made some artificially long types in the following screenshots.
Before screenshot:
<img width="1258" alt="screen shot 2018-08-03 at 8 32 35 pm" src="https://user-images.githubusercontent.com/193874/43670805-56e3b3b4-975e-11e8-9296-600837d03de2.png">
After screenshot:
<img width="1239" alt="screen shot 2018-08-03 at 8 31 17 pm" src="https://user-images.githubusercontent.com/193874/43670810-6591f9ac-975e-11e8-9e12-4ea9ab1e5806.png">
Provide `{to,from}_{ne,le,be}_bytes` functions on integers
If one doesn't view integers as containers of bytes, converting them to
bytes necessarily needs the specfication of encoding.
I think Rust is a language that wants to be explicit. The `to_bytes`
function is basically the opposite of that – it converts an integer into
the native byte representation, but there's no mention (in the function
name) of it being very much platform dependent. Therefore, I think it
would be better to replace that method by three methods, the explicit
`to_ne_bytes` ("native endian") which does the same thing and
`to_{le,be}_bytes` which return the little- resp. big-endian encoding.
