DiagnosticSpanLine pointing inside char boundary

[phansch]

A test from the rustc UI tests currently causes a crash when executed with rustfix:

Click to expand backtrace

thread 'everything' panicked at 'byte index 1 is not a char boundary; it is inside 'γ' (bytes 0..2) of `γ  //~ ERROR non-ascii idents are not fully supported`', src/libcore/str/mod.rs:2027:5
   0:     0x555eeaaef943 - std::sys::unix::backtrace::tracing::imp::unwind_backtrace::h3ea4e797e65a328d
                               at src/libstd/sys/unix/backtrace/tracing/gcc_s.rs:39
   1:     0x555eeaaeb6eb - std::sys_common::backtrace::_print::hf58574b7d5a741ab
                               at src/libstd/sys_common/backtrace.rs:71
   2:     0x555eeaaee5e6 - std::panicking::default_hook::{{closure}}::hed8b316a41353ca9
                               at src/libstd/sys_common/backtrace.rs:59
                               at src/libstd/panicking.rs:197
   3:     0x555eeaaee30e - std::panicking::default_hook::h8f4ba314e61c5eff
                               at src/libstd/panicking.rs:208
   4:     0x555eeaaeecef - std::panicking::rust_panic_with_hook::hde5d2bfa491b5add
                               at src/libstd/panicking.rs:474
   5:     0x555eeaaee871 - std::panicking::continue_panic_fmt::h928c357471014704
                               at src/libstd/panicking.rs:381
   6:     0x555eeaaee755 - rust_begin_unwind
                               at src/libstd/panicking.rs:308
   7:     0x555eeab09eac - core::panicking::panic_fmt::h5ab350f3b9a2b236
                               at src/libcore/panicking.rs:85
   8:     0x555eeab0b73f - core::str::slice_error_fail::h8551c49db625e656
                               at src/libcore/str/mod.rs:0
   9:     0x555eea8c2bff - core::str::traits::<impl core::slice::SliceIndex<str> for core::ops::range::Range<usize>>::index::{{closure}}::h2b84c0866fdd5130
                               at /rustc/0f88167f89fffe321590c5148f21b7d51d44388d/src/libcore/str/mod.rs:1751
  10:     0x555eea8c7a1f - core::option::Option<T>::unwrap_or_else::he9ebd301cf0dbcdb
                               at /rustc/0f88167f89fffe321590c5148f21b7d51d44388d/src/libcore/option.rs:386
  11:     0x555eea8c7095 - core::str::traits::<impl core::slice::SliceIndex<str> for core::ops::range::Range<usize>>::index::ha86fa6cbfda6a990
                               at /rustc/0f88167f89fffe321590c5148f21b7d51d44388d/src/libcore/str/mod.rs:1751
  12:     0x555eea8c2d80 - core::str::traits::<impl core::ops::index::Index<I> for str>::index::h12989e8413e52b49
                               at /rustc/0f88167f89fffe321590c5148f21b7d51d44388d/src/libcore/str/mod.rs:1616
  13:     0x555eea873b50 - <alloc::string::String as core::ops::index::Index<core::ops::range::Range<usize>>>::index::hc28f4c5a88ab385d
                               at /rustc/0f88167f89fffe321590c5148f21b7d51d44388d/src/liballoc/string.rs:1953
  14:     0x555eea87e464 - rustfix::parse_snippet::h40fa32a6f9ed0564
                               at src/lib.rs:110
  15:     0x555eea7ddc8f - rustfix::collect_suggestions::{{closure}}::h396f64ade2bb2873
                               at /home/phansch/code/rustfix/src/lib.rs:165

This is a trimmed down version of src/tests/ui/utf8_idents.rs which causes the crash in rustfix:

fn foo<
γ  //~ ERROR non-ascii idents are not fully supported
   //~^ WARN type parameter `γ` should have an upper camel case name
>() {}

pub fn main() {}

The DiagnosticSpanLine that rustfix receives looks like this:

DiagnosticSpanLine {
    text: "γ  //~ ERROR non-ascii idents are not fully supported",
    highlight_start: 1,
    highlight_end: 2
}

γ is 2 bytes long, which leads me to believe that this may not be a bug in rustfix, but somewhere in rustc, because highlight_end should be 3 in this case?

[estebank]

Digging a bit I believe that highlight_start and highlight_end weren't supposed to be interpreted as byte offsets but rather as character offsets. Couldn't rustfix interpret them that way?

[phansch]

Oh, you're right! It's even mentioned in the struct definition:

rust/src/libsyntax/json.rs

Line 134 in 9ebf478

/// 1-based, character offset in self.text.

Thanks for taking a look. I will have a fix for rustfix ready later today.