Use sorted vectors for DWARF address maps — reduce memory overhead

[lewing]

Problem

wasm-opt uses excessive memory when processing wasm files with DWARF debug symbols, causing OOM on .NET CI Helix agents (dotnet/runtime#125244, dotnet/runtime#125233). The AddrExprMap and FuncAddrMap structures in wasm-debug.cpp use std::unordered_map with ~64 bytes per entry overhead.

Fix

Replace std::unordered_map with a SortedMap (sorted std::vector + binary search). These maps are built once during construction and only used for read-only lookups, making them ideal for this pattern.

Why it works

• std::unordered_map: ~64 bytes/entry (hash buckets, linked list nodes, pointer chasing)
• SortedMap (sorted vector): ~16 bytes/entry (contiguous memory, cache-friendly)
• Binary search O(log n) is faster in practice than hash O(1) due to cache locality on sequential data
• For a wasm binary with 1M expressions, this saves roughly 300-400MB of peak memory

Output is byte-for-byte identical, confirming functional correctness.

This is the first (and simplest) change from #335, split out for easier review.

[Copilot]

Pull request overview

This PR reduces peak memory usage in DWARF debug processing by replacing address-to-expression/function std::unordered_map lookups with a compact sorted-vector map (SortedMap) that is built once and then queried read-only.

Changes:

• Introduces SortedMap<K, V> (sorted std::vector + binary search) for address lookups.
• Replaces AddrExprMap and FuncAddrMap internal std::unordered_map storage with SortedMap.
• Adds up-front reservation and a post-build sort() step for the new maps.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[lewing]

Addressed all three review comments:

2. delimCount under-reservation — Now iterates each DelimiterLocations entry and counts actual non-zero offsets instead of just using delimiterLocations.size().

3. Duplicate keys after sort() — sort() now de-duplicates adjacent equal keys using std::unique (keeps first). This handles legitimate duplicates in FuncAddrMap where funcLocation.start == funcLocation.declarations for some functions.

[Copilot]

Pull request overview

This PR reduces memory usage when processing wasm binaries with DWARF debug info by replacing per-address std::unordered_map lookups with a sorted std::vector-backed map and binary search, targeting OOM issues in large debug symbol workloads.

Changes:

• Introduce a SortedMap<K, V> helper (vector + sort + binary search) for read-only address lookups.
• Replace AddrExprMap and FuncAddrMap internal std::unordered_map structures with SortedMap, including pre-reservation and finalization (sort()).
• Update lookup code paths to use the new find() API and pointer-based returns.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

This PR reduces peak memory usage when processing DWARF debug info by replacing address→node/function hash maps with a sorted-vector-based lookup structure in wasm-debug.cpp, leveraging the fact that these maps are constructed once and then queried read-only during DWARF rewriting.

Changes:

• Introduce a SortedMap (sorted std::vector + binary search) for address-keyed lookups.
• Replace std::unordered_map usages in AddrExprMap and FuncAddrMap with SortedMap.
• Pre-reserve vector capacity and finalize maps via sorting/deduplication after construction.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[lewing]

Both remaining comments addressed in 68b89a6:

• sort() duplicate validation (line 392): sort() now takes assertUniqueValues param (default true). In debug builds, asserts that duplicate keys map to the same value before de-duplicating. AddrExprMap uses the default strict check; FuncAddrMap passes false since contiguous functions legitimately share boundary addresses (func1.end == func2.start).

• AddrExprMap uniqueness (line 478): The strict assertion (assertUniqueValues=true) now validates during sort() that startMap and endMap entries have unique keys (or identical values for any duplicates), preserving the same invariant as the old pre-sort assertions.

[radekdoulik]

Binary search O(log n) is faster in practice than hash O(1) due to cache locality on sequential data

I am curious, do you have some measurements for this? What is a typical size of the data, how many entries?