From 57ca3512e429f962edc15ea2b24a57f9c778320d Mon Sep 17 00:00:00 2001 From: Ali Cheraghi Date: Tue, 9 Jun 2026 22:37:23 +0330 Subject: [PATCH] std.sort.pdq: fix stack overflow on 32-bit targets our pdq implementation uses an explicit stack of `Range` entries to avoid recursion. The size of this stack was set to `log2(maxInt(usize) + 1)`. on 32-bit targets like `wasm32-freestanding`, this would hit the maximum size, causing OOB. there were two issues: - `max_limit`, which controls how many imbalanced partitions are allowed before the algorithm falls back to heapsort, was computed as `floorPowerOfTwo(n) + 1` instead of `log2(n)` as in the original c++ reference implementation. e.g. for `n = 1_000_000`, this allowed 524289 bad partitions before heapsort would kick in, instead of 19. this meant deeply imbalanced partitions could accumulate stack pushes essentially without limit. However, i didn't observe any meaningful difference in benchmarks. - the worst-case stack depth is bounded by `log2(n) + max_limit`, which approaches `2 * log2(n)`. The reference c++ and go implementation doesn't have this problem because they don't use explicit stack buffer. --- lib/std/sort/pdq.zig | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/lib/std/sort/pdq.zig b/lib/std/sort/pdq.zig index 85d3a62c9c5233441edb013b6f6d611d7159b9b9..5b0da14156ae1365f7a677c84453d0b07d84a774 100644 --- a/lib/std/sort/pdq.zig +++ b/lib/std/sort/pdq.zig @@ -44,12 +44,11 @@ pub fn pdqContext(a: usize, b: usize, context: anytype) void { // slices of up to this length get sorted using insertion sort. const max_insertion = 24; // number of allowed imbalanced partitions before switching to heap sort. - const max_limit = std.math.floorPowerOfTwo(usize, b - a) + 1; + const max_limit = if (b > a) math.log2_int(usize, b - a) else 0; // set upper bound on stack memory usage. const Range = struct { a: usize, b: usize, limit: usize, leftmost: bool }; - const stack_size = math.log2(math.maxInt(usize) + 1); - var stack: [stack_size]Range = undefined; + var stack: [2 * @bitSizeOf(usize)]Range = undefined; var range = Range{ .a = a, .b = b, .limit = max_limit, .leftmost = true }; var top: usize = 0; -- 2.54.0