annotate-snippets-rs
annotate-snippets-rs copied to clipboard
rust-lang
Source code with tabs did not highlight correctly
Here some example Slice where t symbol in source code have one space behind:
That Slice give us that output:
Now i change space behind t from space to tab(which still one symbol, so range of highlight should not be changed):
And highlight in output is wrong now:
Yeah, I think annotate snippets should replace tab characters with a user configurable number of spaces, and take it into account when computing underlines. This is what I do in codespan at least!
agree! I am swamped a bit these days but will try to debug it! PRs welcomed!
agree! I am swamped a bit these days but will try to debug it! PRs welcomed!
@legendiguess - can you retest against master? #27 got merged and I can't reproduce it anymore locally!
@zbraniecki
Just retested it - problem still here:
Here gist with minimal reproducible example https://gist.github.com/legendiguess/ccb197dac4bc77e3ae7571befc8ffbe2
Ah, I see. Now I understand the issue.
@brendanzab 's solution is one option.
The other would be to only adjust the underline calculation to account for tabs in line, but I'm not sure if display of "\tt" is always 5 ascii characters.
The other would be to only adjust the underline calculation to account for tabs in line, but I'm not sure if display of
"\tt"is always 5 ascii characters.
\t move next characters to the next tabulation column. so basically it's adds from 1 to 4 ascii characters, dependent on caret position off last tabulation column
Yeah, so it may be tricky to predict. Wondering how rustc handles that in error reporting now.
Yeah, so it may be tricky to predict. Wondering how rustc handles that in error reporting now.
https://github.com/rust-lang/rust/blob/5e8897b7b51636f157630e6639b711d698e1d101/src/librustc_errors/styled_buffer.rs#L16
They don't handle \t in a fully "proper way"
I do this using a custom writer in codespan-reporting: https://github.com/brendanzab/codespan/blob/a77e646944765dfb338185b0ba9504500afb7304/codespan-reporting/src/term/config.rs#L56-L84
Oh, and importantly I also measure the length of strings taking the tab width into account: https://github.com/brendanzab/codespan/blob/a77e646944765dfb338185b0ba9504500afb7304/codespan-reporting/src/term/config.rs#L35-L45
Here's the relevant tests (excuse the not-so-nice test setup!): https://github.com/brendanzab/codespan/blob/a77e646944765dfb338185b0ba9504500afb7304/codespan-reporting/tests/term.rs#L415-L542
\t move next characters to the next tabulation column. so basically it's adds from 1 to 4 ascii characters, dependent on caret position off last tabulation column
Oh wait, I understand what you're saying now, @legendiguess - yes I don't handle this properly in codespan-reporting either 🤦♂
https://github.com/rust-lang/annotate-snippets-rs/issues/58 can be a duplicate.
a user configurable number of spaces
When I dig into this issue yesterday, it turns out to be quite subtle even the expand POSIX command can produce "unexpected" result.
That said, leading \t (only) is a easy case, while tab in the medium can be quite complex, especially mixed with other whitespace. Generally, tab means making table according to tabstop so you need to translate the tab into whitespace to ensure that the whitespace width are aligned at the tabstop. But it's very unclear that even expand handles \t and \t differently within one tabstop slot.
Concretely, with tabstop=4, \t is translate to four spaces while \t is translated to five spaces. I suspect it's because is treated either as part of the whitespace or part of the next "column".
Generally, users are aware of the "final" result displayed on their screen, and don't care what the whitespaces there are. So, I guess there is no consensus on how a tab should be formatted "properly". We can make our rule, while it can still be different to what users "expected".