Improve spans for indexing expressions
fixes#114388
Indexing is similar to method calls in having an arbitrary left-hand-side and then something on the right, which is the main part of the expression. Method calls already have a span for that right part, but indexing does not. This means that long method chains that use indexing have really bad spans, especially when the indexing panics and that span in coverted into a panic location.
This does the same thing as method calls for the AST and HIR, storing an extra span which is then put into the `fn_span` field in THIR.
r? compiler-errors
Improve spans for indexing expressions
fixes#114388
Indexing is similar to method calls in having an arbitrary left-hand-side and then something on the right, which is the main part of the expression. Method calls already have a span for that right part, but indexing does not. This means that long method chains that use indexing have really bad spans, especially when the indexing panics and that span in coverted into a panic location.
This does the same thing as method calls for the AST and HIR, storing an extra span which is then put into the `fn_span` field in THIR.
r? compiler-errors
Lots of tiny incremental simplifications of `EmitterWriter` internals
ignore the first commit, it's https://github.com/rust-lang/rust/pull/114088 squashed and rebased, but it's needed to use to use `derive_setters`, as they need a newer `syn` version.
Then this PR starts out with removing many arguments that are almost always defaulted to `None` or `false` and replace them with builder methods that can set these fields in the few cases that want to set them.
After that it's one commit after the other that removes or merges things until everything becomes some very simple trait objects
Lots of tiny incremental simplifications of `EmitterWriter` internals
ignore the first commit, it's https://github.com/rust-lang/rust/pull/114088 squashed and rebased, but it's needed to use to use `derive_setters`, as they need a newer `syn` version.
Then this PR starts out with removing many arguments that are almost always defaulted to `None` or `false` and replace them with builder methods that can set these fields in the few cases that want to set them.
After that it's one commit after the other that removes or merges things until everything becomes some very simple trait objects
Indexing is similar to method calls in having an arbitrary
left-hand-side and then something on the right, which is the main part
of the expression. Method calls already have a span for that right part,
but indexing does not. This means that long method chains that use
indexing have really bad spans, especially when the indexing panics and
that span in coverted into a panic location.
This does the same thing as method calls for the AST and HIR, storing an
extra span which is then put into the `fn_span` field in THIR.
Indexing is similar to method calls in having an arbitrary
left-hand-side and then something on the right, which is the main part
of the expression. Method calls already have a span for that right part,
but indexing does not. This means that long method chains that use
indexing have really bad spans, especially when the indexing panics and
that span in coverted into a panic location.
This does the same thing as method calls for the AST and HIR, storing an
extra span which is then put into the `fn_span` field in THIR.
It's the same as `Delimiter`, minus the `Invisible` variant. I'm
generally in favour of using types to make impossible states
unrepresentable, but this one feels very low-value, and the conversions
between the two types are annoying and confusing.
Look at the change in `src/tools/rustfmt/src/expr.rs` for an example:
the old code converted from `MacDelimiter` to `Delimiter` and back
again, for no good reason. This suggests the author was confused about
the types.
It's the same as `Delimiter`, minus the `Invisible` variant. I'm
generally in favour of using types to make impossible states
unrepresentable, but this one feels very low-value, and the conversions
between the two types are annoying and confusing.
Look at the change in `src/tools/rustfmt/src/expr.rs` for an example:
the old code converted from `MacDelimiter` to `Delimiter` and back
again, for no good reason. This suggests the author was confused about
the types.
Less `TokenTree` cloning
`TokenTreeCursor` has this comment on it:
```
// FIXME: Many uses of this can be replaced with by-reference iterator to avoid clones.
```
This PR completes that FIXME. It doesn't have much perf effect, but at least we now know that.
r? `@petrochenkov`
Resolves 5542
Prior to rust-lang/rust#101212 the `ast::TraitObjectSyntax` enum only
had two variants `Dyn` and `None`. The PR that introduced the `dyn*`
syntax added a new variant `DynStar`, but did not update the formatting
rules to account for the new variant.
Now the new `DynStar` variant is properly handled and is no longer
removed by rustfmt.
Currently, Clippy, Miri, Rustfmt, and rustc all use an environment variable to
indicate that output should be blessed, but they use different variable names.
In order to improve consistency, this patch applies the following changes:
- Emit `RUSTC_BLESS` within `prepare_cargo_test` so it is always
available
- Change usage of `MIRI_BLESS` in the Miri subtree to use `RUSTC_BLESS`
- Change usage of `BLESS` in the Clippy subtree to `RUSTC_BLESS`
- Change usage of `BLESS` in the Rustfmt subtree to `RUSTC_BLESS`
- Adjust the blessable test in `rustc_errors` to use this same
convention
- Update documentation where applicable
Any tools that uses `RUSTC_BLESS` should check that it is set to any value
other than `"0"`.
Update docs to include an example of running `rustfmt` built from src,
and show how users can set the `RUSTFMT` environment variable to test
`cargo-fmt` using the `rustfmt` they built from src.
Implement rust-lang/compiler-team#578.
When an ICE is encountered on nightly releases, the new rustc panic
handler will also write the contents of the backtrace to disk. If any
`delay_span_bug`s are encountered, their backtrace is also added to the
file. The platform and rustc version will also be collected.
Fixes 5730
Previously rustfmt was attempting to slice a string with an invalid
range (`start > end`), leading to the ICE.
When formatting a macro transcriber snippet consisting of a lone
semicolon, the snippet was being formatted into the empty string,
leading the enclosing `fn main() {\n}` added by `format_code_block` to
be formatted into `fn main() {}`. However, rustfmt was assuming that the
enclosing function string's length had been left unchanged. This was
leading to an invalid range being constructed when attempting to trim
off the enclosing function.
The fix is to just clamp the range's start to be less than or equal
to the range's end, since if `end < start` there's nothing to iterate
over anyway.
* Use matches!() macro to improve readability
1. Use `matches!()` macro in `is_line_comment` and `is_block_comment` to
improve readability.
2. Very sightly improve the wording of the doc comment for these two functions.
* Update wording on doc comment on is_line_comment()
There was an issue with the script when passing optional rustfmt configs
without specifying a commit hash. Because these optional values are
passed via positional arguments the configs ($4) would be used in place
of the commit hash ($3). Now that we set a default value for the
optional commit hash we avoid this problem.
The `set -e` option is used to immediately exit if any command exits
with a non zero exit status. This will help us catch errors in the
script, for example, needing the `LD_LIBRARY_PATH` to be set.
There were some upstream changes made a while back that requires this to
be set when building rustfmt from source like we do in the
`check_diff.sh` script.
See issue 5675 for more details.
By reversing the logic I felt that the code became a clearer. Also,
added a comment to make it clear that we need to take the trailing
semicolon for the `let-else` statement into account.
This allows users to configure the maximum length of a single line
`let-else` statements. `let-else` statements that otherwise meet the
requirements to be formatted on a single line will have their divergent
`else` block formatted over multiple lines if they exceed this length.
**Note**: `single_line_let_else_max_widt` will be introduced as a stable
configuration option.
