fix(highlight_source): Apply styling to multiline annotations

[momomuchu]

Fixes #427.

highlight_source was silently ignored on multiline annotations
(MultilineStart/MultilineLine/MultilineEnd). Two things conspired to
cause this:

• render_source_line has a couple of fast paths (the "short start with
  no leading text" case, and the "only connector lines, nothing to
  show" case) that can return before the per-annotation styling logic
  ever runs.
• Even when those aren't hit, the per-annotation match arm that
  applies highlight_source is placed after the arm that draws the
  MultilineStart/MultilineEnd connector, so it never matches for those
  variants, since Rust match arms are first-match-wins.

Styling multiline annotations up front, before any of that, sidesteps
both issues instead of trying to thread a fix through three separate
spots.

Added a regression test (ascii + unicode goldens) using the repro from
the issue.

[estebank]

Note that eventually we'd have to add a mechanism to provide highlighting exclusions (so that something like rustc can avoid highlighting trailing comments within the multiline span).

[epage]

I'll probably leave any further reviewing to others but at least noticed this in pre-review that might help with the main review.

[epage]

Huh, we don't have a CONTRIBUTING.md for this but something we've found very helpful is to add tests in the commit before, with them passing, showing the bad behavior. This makes it so when we diff for code changes, we also diff for the effect of that code that change.

[momomuchu]

Done. Restructured into two commits: the test first with pre-fix goldens showing the missing styling, then the fix with the goldens regenerated. Tree content is unchanged from the previous head, only the history was split.

[Muscraft]

Thanks for taking the time to fix this issue! I'm sorry for the extremely long delay in reviewing this! My life has been very busy recently, leaving me with very little time to review this properly.

[Muscraft]

Suggested change

	LineAnnotationType::Singleline => continue,
	LineAnnotationType::Singleline => (ann.start.char, ann.end.char),

Note: You will need to remove the existing code for SingleLine:

annotate-snippets-rs/src/renderer/render.rs

Lines 1192 to 1200 in c203ec6

	_ if annotation.highlight_source => {
	buffer.set_style_range(
	line_offset,
	(code_offset + annotation.start.char).saturating_sub(left),
	(code_offset + annotation.end.char).saturating_sub(left),
	underline.style,
	annotation.is_primary(),
	);
	}

[Muscraft]

ann.start.byte is the byte offset from the start of the source, not the current line

Suggested change

	let rest_of_line = line_info.line[ann.start.byte..].chars().count();
	(ann.start.char, ann.start.char + rest_of_line)
	(ann.start.char, line_info.line.chars().count())

[Muscraft]

The current code does not account for tabs (\t), and should do something similar to style_substitution_highlights:

Suggested change

	buffer.set_style_range(
	line_offset,
	(code_offset + start_char).saturating_sub(left),
	(code_offset + end_char).saturating_sub(left),
	underline.style,
	ann.is_primary(),
	);
	let extra_width_start = extra_width_from_tabs(line_info.line, start_char);
	let extra_width_end = extra_width_from_tabs(line_info.line, end_char);
	buffer.set_style_range(
	line_offset,
	(code_offset + start_char + extra_width_start).saturating_sub(left),
	(code_offset + end_char + extra_width_end).saturating_sub(left),
	underline.style,
	ann.is_primary(),
	);

Possible regression test

use annotate_snippets::{
    AnnotationKind, Group, Level, Patch, Renderer, Snippet, renderer::DecorStyle,
};
use snapbox::{assert_data_eq, file};

#[test]
fn case() {
    let source = "prefix\tstart\n\tmiddle\nend";

    let report = &[Group::with_level(Level::ERROR).element(
        Snippet::source(source)
            .annotation(AnnotationKind::Primary.span(7..24).highlight_source(true)),
    )];

    let expected_ascii = file!["<test-name>.ascii.term.svg": TermSvg];
    let renderer = Renderer::styled();
    assert_data_eq!(renderer.render(report), expected_ascii);

    let expected_unicode = file!["<test-name>.unicode.term.svg": TermSvg];
    let renderer = renderer.decor_style(DecorStyle::Unicode);
    assert_data_eq!(renderer.render(report), expected_unicode);
}

[Muscraft]

While playing around with these changes, I realized that when .fold(false) is set, lines that would normally be folded are not highlighted (when they should be). I suspect that annotated_lines is the source of the issue, but I have not checked.

Possible regression test

use annotate_snippets::{
    AnnotationKind, Group, Level, Patch, Renderer, Snippet, renderer::DecorStyle,
};
use snapbox::{assert_data_eq, file};

#[test]
fn case() {
    let source = r#"line
line
line
line
normally folded
normally folded
normally folded
line
primary end
normally folded
normally folded
normally folded
line
context end
"#;

    let report = &[Group::with_level(Level::ERROR).element(
        Snippet::source(source)
            .fold(false)
            .annotation(AnnotationKind::Primary.span(1..84).highlight_source(true))
            .annotation(AnnotationKind::Context.span(0..149).highlight_source(true)),
    )];

    let expected_ascii = file!["<test-name>.ascii.term.svg": TermSvg];
    let renderer = Renderer::styled();
    assert_data_eq!(renderer.render(report), expected_ascii);

    let expected_unicode = file!["<test-name>.unicode.term.svg": TermSvg];
    let renderer = renderer.decor_style(DecorStyle::Unicode);
    assert_data_eq!(renderer.render(report), expected_unicode);
}