coverage: llvm-cov expects column numbers to be bytes, not code points

Author: Zalathar

compiler/rustc_mir_transform/src/coverage/mod.rs

@@ -23,7 +23,7 @@ use rustc_middle::mir::{
 use rustc_middle::ty::TyCtxt;
 use rustc_span::def_id::LocalDefId;
 use rustc_span::source_map::SourceMap;
-use rustc_span::{Span, Symbol};
+use rustc_span::{BytePos, Pos, RelativeBytePos, Span, Symbol};
 
 /// Inserts `StatementKind::Coverage` statements that either instrument the binary with injected
 /// counters, via intrinsic `llvm.instrprof.increment`, and/or inject metadata used during codegen
@@ -258,7 +258,16 @@ fn inject_statement(mir_body: &mut mir::Body<'_>, counter_kind: CoverageKind, bb
     data.statements.insert(0, statement);
 }
 
-/// Convert the Span into its file name, start line and column, and end line and column
+/// Convert the Span into its file name, start line and column, and end line and column.
+///
+/// Line numbers and column numbers are 1-based. Unlike most column numbers emitted by
+/// the compiler, these column numbers are denoted in **bytes**, because that's what
+/// LLVM's `llvm-cov` tool expects to see in coverage maps.
+///
+/// Returns `None` if the conversion failed for some reason. This shouldn't happen,
+/// but it's hard to rule out entirely (especially in the presence of complex macros
+/// or other expansions), and if it does happen then skipping a span or function is
+/// better than an ICE or `llvm-cov` failure that the user might have no way to avoid.
 fn make_code_region(
     source_map: &SourceMap,
     file_name: Symbol,
@@ -272,20 +281,54 @@ fn make_code_region(
         source_map.span_to_diagnostic_string(body_span)
     );
 
-    let (file, mut start_line, mut start_col, mut end_line, mut end_col) =
-        source_map.span_to_location_info(span);
-    if span.hi() == span.lo() {
-        // Extend an empty span by one character so the region will be counted.
-        if span.hi() == body_span.hi() {
-            start_col = start_col.saturating_sub(1);
-        } else {
-            end_col = start_col + 1;
-        }
+    let lo = span.lo();
+    let hi = span.hi();
+
+    let file = source_map.lookup_source_file(lo);
+    if !file.contains(hi) {
+        debug!(?span, ?file, ?lo, ?hi, "span crosses multiple files; skipping");
+        return None;
+    }
+
+    // Column numbers need to be in bytes, so we can't use the more convenient
+    // `SourceMap` methods for looking up file coordinates.
+    let rpos_and_line_and_byte_column = |pos: BytePos| -> Option<(RelativeBytePos, usize, usize)> {
+        let rpos = file.relative_position(pos);
+        let line_index = file.lookup_line(rpos)?;
+        let line_start = file.lines()[line_index];
+        // Line numbers and column numbers are 1-based, so add 1 to each.
+        Some((rpos, line_index + 1, (rpos - line_start).to_usize() + 1))
     };
-    if let Some(file) = file {
-        start_line = source_map.doctest_offset_line(&file.name, start_line);
-        end_line = source_map.doctest_offset_line(&file.name, end_line);
+
+    let (lo_rpos, mut start_line, mut start_col) = rpos_and_line_and_byte_column(lo)?;
+    let (hi_rpos, mut end_line, mut end_col) = rpos_and_line_and_byte_column(hi)?;
+
+    // If the span is empty, try to expand it by one character so that it is
+    // more visible in `llvm-cov` reports. (These column numbers are bytes, so
+    // "one character" might be multiple bytes.)
+    if span.is_empty()
+        && body_span.contains(span)
+        && let Some(src) = &file.src
+    {
+        // Prefer to expand the end position, if it won't go outside the body span.
+        if hi < body_span.hi() {
+            let hi_rpos = hi_rpos.to_usize();
+            let nudge_bytes = src.ceil_char_boundary(hi_rpos + 1) - hi_rpos;
+            end_col += nudge_bytes;
+        } else if lo > body_span.lo() {
+            let lo_rpos = lo_rpos.to_usize();
+            let nudge_bytes = lo_rpos - src.floor_char_boundary(lo_rpos - 1);
+            // Subtract the nudge, but don't go below column 1.
+            start_col = start_col.saturating_sub(nudge_bytes).max(1);
+        }
+        // If neither nudge could be applied, stick with the empty span coordinates.
     }
+
+    // Apply an offset so that code in doctests has correct line numbers.
+    // FIXME(#79417): Currently we have no way to offset doctest _columns_.
+    start_line = source_map.doctest_offset_line(&file.name, start_line);
+    end_line = source_map.doctest_offset_line(&file.name, end_line);
+
     Some(CodeRegion {
         file_name,
         start_line: start_line as u32,

compiler/rustc_mir_transform/src/lib.rs

@@ -10,6 +10,7 @@
 #![feature(min_specialization)]
 #![feature(never_type)]
 #![feature(option_get_or_insert_default)]
+#![feature(round_char_boundary)]
 #![feature(trusted_step)]
 #![feature(try_blocks)]
 #![feature(yeet_expr)]

tests/coverage/unicode.cov-map

@@ -1,5 +1,5 @@
 Function name: unicode::main
-Raw bytes (67): 0x[01, 01, 09, 01, 05, 03, 05, 1e, 0d, 22, 09, 03, 05, 11, 1b, 1e, 0d, 22, 09, 03, 05, 09, 01, 0e, 01, 00, 0b, 05, 01, 09, 00, 0b, 03, 00, 0f, 00, 18, 05, 00, 19, 00, 24, 22, 02, 08, 00, 13, 09, 00, 17, 00, 22, 11, 00, 23, 02, 06, 1b, 02, 06, 00, 07, 17, 02, 05, 01, 02]
+Raw bytes (67): 0x[01, 01, 09, 01, 05, 03, 05, 1e, 0d, 22, 09, 03, 05, 11, 1b, 1e, 0d, 22, 09, 03, 05, 09, 01, 0e, 01, 00, 0b, 05, 01, 09, 00, 0c, 03, 00, 10, 00, 1b, 05, 00, 1c, 00, 28, 22, 02, 08, 00, 25, 09, 00, 29, 00, 46, 11, 00, 47, 02, 06, 1b, 02, 06, 00, 07, 17, 02, 05, 01, 02]
 Number of files: 1
 - file 0 => global file 1
 Number of expressions: 9
@@ -14,34 +14,34 @@ Number of expressions: 9
 - expression 8 operands: lhs = Expression(0, Add), rhs = Counter(1)
 Number of file 0 mappings: 9
 - Code(Counter(0)) at (prev + 14, 1) to (start + 0, 11)
-- Code(Counter(1)) at (prev + 1, 9) to (start + 0, 11)
-- Code(Expression(0, Add)) at (prev + 0, 15) to (start + 0, 24)
+- Code(Counter(1)) at (prev + 1, 9) to (start + 0, 12)
+- Code(Expression(0, Add)) at (prev + 0, 16) to (start + 0, 27)
     = (c0 + c1)
-- Code(Counter(1)) at (prev + 0, 25) to (start + 0, 36)
-- Code(Expression(8, Sub)) at (prev + 2, 8) to (start + 0, 19)
+- Code(Counter(1)) at (prev + 0, 28) to (start + 0, 40)
+- Code(Expression(8, Sub)) at (prev + 2, 8) to (start + 0, 37)
     = ((c0 + c1) - c1)
-- Code(Counter(2)) at (prev + 0, 23) to (start + 0, 34)
-- Code(Counter(4)) at (prev + 0, 35) to (start + 2, 6)
+- Code(Counter(2)) at (prev + 0, 41) to (start + 0, 70)
+- Code(Counter(4)) at (prev + 0, 71) to (start + 2, 6)
 - Code(Expression(6, Add)) at (prev + 2, 6) to (start + 0, 7)
     = ((((c0 + c1) - c1) - c2) + c3)
 - Code(Expression(5, Add)) at (prev + 2, 5) to (start + 1, 2)
     = (c4 + ((((c0 + c1) - c1) - c2) + c3))
 
 Function name: unicode::サビ
-Raw bytes (9): 0x[01, 01, 00, 01, 01, 1e, 12, 00, 14]
+Raw bytes (9): 0x[01, 01, 00, 01, 01, 1e, 14, 00, 18]
 Number of files: 1
 - file 0 => global file 1
 Number of expressions: 0
 Number of file 0 mappings: 1
-- Code(Counter(0)) at (prev + 30, 18) to (start + 0, 20)
+- Code(Counter(0)) at (prev + 30, 20) to (start + 0, 24)
 
 Function name: unicode::他 (unused)
-Raw bytes (9): 0x[01, 01, 00, 01, 00, 1e, 15, 00, 1f]
+Raw bytes (9): 0x[01, 01, 00, 01, 00, 1e, 19, 00, 25]
 Number of files: 1
 - file 0 => global file 1
 Number of expressions: 0
 Number of file 0 mappings: 1
-- Code(Zero) at (prev + 30, 21) to (start + 0, 31)
+- Code(Zero) at (prev + 30, 25) to (start + 0, 37)
 
 Function name: unicode::申し訳ございません
 Raw bytes (9): 0x[01, 01, 00, 01, 01, 18, 01, 02, 02]

tests/coverage/unicode.coverage

@@ -0,0 +1,40 @@
+   LL|       |// edition: 2021
+   LL|       |// ignore-windows - we can't force `llvm-cov` to use ANSI escapes on Windows
+   LL|       |// llvm-cov-flags: --use-color
+   LL|       |
+   LL|       |// Check that column numbers are denoted in bytes, so that they don't cause
+   LL|       |// `llvm-cov` to fail or emit malformed output.
+   LL|       |//
+   LL|       |// Note that when `llvm-cov` prints ^ arrows on a subsequent line, it simply
+   LL|       |// inserts one space character for each "column", with no understanding of
+   LL|       |// Unicode or character widths. So those arrows will tend to be misaligned
+   LL|       |// for non-ASCII source code, regardless of whether column numbers are code
+   LL|       |// points or bytes.
+   LL|       |
+   LL|      1|fn main() {
+   LL|     33|    for _İ in 'А'..='Я' { /* Я */ }
+                      ^32                ^32
+   LL|       |
+   LL|      1|    if 申し訳ございません() && 申し訳ございません() {
+                                                      ^0
+   LL|      0|        println!("true");
+   LL|      1|    }
+   LL|       |
+   LL|      1|    サビ();
+   LL|      1|}
+   LL|       |
+   LL|      1|fn 申し訳ございません() -> bool {
+   LL|      1|    std::hint::black_box(false)
+   LL|      1|}
+   LL|       |
+   LL|       |macro_rules! macro_that_defines_a_function {
+   LL|       |    (fn $名:ident () $体:tt) => {
+   LL|      1|        fn $名 () $体 fn 他 () {}
+                                      ^0
+   LL|       |    }
+   LL|       |}
+   LL|       |
+   LL|       |macro_that_defines_a_function! {
+   LL|       |    fn サビ() {}
+   LL|       |}
+

tests/coverage/unicode.rs

@@ -1,5 +1,5 @@
 // edition: 2021
-// ignore-mode-coverage-run - `llvm-cov` fails due to incorrectly-split UTF-8
+// ignore-windows - we can't force `llvm-cov` to use ANSI escapes on Windows
 // llvm-cov-flags: --use-color
 
 // Check that column numbers are denoted in bytes, so that they don't cause