Function source_position_at 

pub fn source_position_at(
    read: impl Fn(usize) -> Option<String>,
    cursor: usize,
    cursor_byte: u32,
) -> Option<SourcePos>

MG.50: the SOURCE position the cursor is looking at, inside a diff.

<CR> in emacs magit opens the file at the code under the cursor, not at the top. The diff already carries the answer: a hunk’s @@ header names where its body starts in the new file, so the target is that start plus however many new-side rows precede the cursor within the hunk.

Which rows count. Only those that exist on the NEW side — context ( ) and additions (+). A deletion (-) is not in the file being opened, so it advances nothing; a cursor sitting on one resolves to the position where the deleted text was, which is where a reader looking at it wants to land. \ No newline at end of file belongs to neither side.

The byte offset. Every hunk body row carries the one-character diff marker ( / + / -) at byte 0, so source byte = cursor byte − 1. Subtracting a BYTE rather than a column is what makes this correct on rows containing multibyte text: the marker is always exactly one byte, whatever follows it. saturating_sub handles a cursor parked ON the marker, which resolves to the start of the source line rather than wrapping. On the @@ row there is no code under the cursor to align to, so the offset is 0.

Line and offset are answered together, by one walk, rather than by a source_line_at plus a separate offset helper. The bug this replaced was precisely a caller that had the line and defaulted the offset to 0; a shape where you cannot obtain one without the other cannot regress that way again.

None when the cursor is not inside a hunk (a file entry, a section header, a diff --git line) — the caller then opens at the top, which is what emacs does for a file entry too.