perf(codegen): optimize sourcemap builder to reduce allocations (#13670)

Boshen · Boshen · commit b35bf301dd18 · 2025-09-11T05:28:51.000Z
## Summary This PR optimizes the sourcemap builder to reduce memory allocations and improve performance. Profiling showed that `update_generated_line_and_column` and `add_source_mapping` were performance bottlenecks. ### Changes Made **Optimization 4: Optimize Line Offset Table Generation** - Pre-allocate `columns` vector with capacity 256 to avoid frequent reallocations - Replace `columns.clone().into_boxed_slice()` with `std::mem::take(&mut columns).into_boxed_slice()` to avoid unnecessary cloning - Reserve capacity after taking the vector to maintain performance for subsequent lines ### Performance Impact These changes reduce memory allocations when generating sourcemaps, especially for files with Unicode characters. The `clone()` operation was creating unnecessary copies of potentially large vectors on every Unicode line, which is now eliminated. ### Future Optimizations Additional optimizations from the analysis that could be implemented in follow-up PRs: 1. Use SIMD-accelerated line break detection with `memchr` 2. Optimize UTF-16 column calculation to avoid iterator allocation 3. Add fast path for sequential token processing 5. Inline hot functions with `#[inline(always)]` ## Test Plan - [x] All existing tests pass - [x] No functional changes, only performance optimizations - [x] Verified that sourcemap generation still works correctly 🤖 Generated with [Claude Code](https://claude.ai/code)
diff --git a/crates/oxc_codegen/src/sourcemap_builder.rs b/crates/oxc_codegen/src/sourcemap_builder.rs
@@ -308,7 +308,8 @@ impl<'a> SourcemapBuilder<'a> {
         let mut column_offsets = IndexVec::new();
 
         // Used as a buffer to reduce memory reallocations
-        let mut columns = vec![];
+        // Pre-allocate with reasonable capacity to avoid frequent reallocations
+        let mut columns = Vec::with_capacity(256);
 
         // Process content line-by-line.
         // For each line, start by assuming line will be entirely ASCII, and read byte-by-byte.
@@ -389,11 +390,13 @@ impl<'a> SourcemapBuilder<'a> {
                             line_byte_offset += chunk_byte_offset;
 
                             // Record column offsets
+                            // Use mem::take to avoid clone - moves columns out and replaces with empty Vec
                             column_offsets.push(ColumnOffsets {
                                 byte_offset_to_first: byte_offset_from_line_start,
-                                columns: columns.clone().into_boxed_slice(),
+                                columns: std::mem::take(&mut columns).into_boxed_slice(),
                             });
-                            columns.clear();
+                            // Reserve capacity for next line to avoid reallocation
+                            columns.reserve(256);
 
                             // Revert back to outer loop for next line
                             continue 'lines;
