From 67b821e925d32bd550d01e01ba201734f36674a3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Motiejus=20Jak=C5=A1tys?= <motiejus@jakstys.lt>
Date: Fri, 20 Feb 2026 08:42:53 +0000
Subject: [PATCH] verbose_air: zero padding in Air.Inst.Data when collecting

Air.Inst.Data is a union; variants smaller than 8 bytes (un_op,
no_op, ty, repeat) leave padding bytes uninitialised.  Zero the
destination buffer and copy only the meaningful bytes per instruction
so that byte-level comparisons in tests are deterministic and
valgrind-clean.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 src/verbose_air.zig | 48 +++++++++++++++++++++++++++++++++++----------
 1 file changed, 38 insertions(+), 10 deletions(-)

diff --git a/src/verbose_air.zig b/src/verbose_air.zig
index 736beaa724..22cd44265c 100644
--- a/src/verbose_air.zig
+++ b/src/verbose_air.zig
@@ -10,6 +10,34 @@ const Compilation = zig_internals.Compilation;
 const Package = zig_internals.Package;
 const Air = zig_internals.Air;
 
+/// Number of meaningful bytes in Air.Inst.Data for a given tag.
+/// Variants smaller than 8 bytes leave padding uninitialised; callers
+/// must only copy this many bytes and zero the rest.
+fn airInstDataSize(tag: Air.Inst.Tag) usize {
+    return switch (tag) {
+        // no_op: 0 meaningful bytes
+        .ret_addr, .frame_addr => 0,
+        // un_op: 4 meaningful bytes (single Ref / u32)
+        .sqrt, .sin, .cos, .tan, .exp, .exp2,
+        .log, .log2, .log10,
+        .floor, .ceil, .round, .trunc_float,
+        .neg, .neg_optimized,
+        .is_null, .is_non_null, .is_null_ptr, .is_non_null_ptr,
+        .is_err, .is_non_err, .is_err_ptr, .is_non_err_ptr,
+        .ret, .ret_safe, .ret_load,
+        .is_named_enum_value, .tag_name, .error_name,
+        .cmp_lt_errors_len,
+        .c_va_end,
+        => 4,
+        // ty: 4 meaningful bytes (single Type / u32)
+        .alloc, .ret_ptr, .c_va_start => 4,
+        // repeat: 4 meaningful bytes (single Index / u32)
+        .repeat => 4,
+        // All other variants use the full 8 bytes.
+        else => 8,
+    };
+}
+
 /// Matches C `Air` struct layout (air.h).
 const CAir = extern struct {
     inst_len: u32,
@@ -64,20 +92,20 @@ const AirCollector = struct {
             break :blk dst.ptr;
         } else null;
 
-        // Copy datas (8 bytes per instruction)
+        // Copy datas (8 bytes per instruction).
+        // Air.Inst.Data is a union; variants smaller than 8 bytes
+        // (un_op, no_op, ty, repeat) leave padding bytes uninitialised.
+        // Zero the buffer first, then copy only the meaningful bytes
+        // per instruction so that padding is deterministically zero.
         const datas_byte_len = inst_len * 8;
         const datas_copy: ?[*]u8 = if (inst_len > 0) blk: {
             const dst = try gpa.alloc(u8, datas_byte_len);
+            @memset(dst, 0);
             const zig_datas = air.instructions.items(.data);
-            if (@sizeOf(Air.Inst.Data) == 8) {
-                const src = @as([*]const u8, @ptrCast(zig_datas.ptr))[0..datas_byte_len];
-                @memcpy(dst, src);
-            } else {
-                // Safety build: @sizeOf(Data) may be > 8, copy first 8 bytes per element
-                for (zig_datas, 0..) |*d, i| {
-                    const src = @as(*const [8]u8, @ptrCast(d));
-                    @memcpy(dst[i * 8 ..][0..8], src);
-                }
+            for (0..inst_len) |i| {
+                const src = @as(*const [8]u8, @ptrCast(&zig_datas[i]));
+                const n = airInstDataSize(zig_tags[i]);
+                @memcpy(dst[i * 8 ..][0..n], src[0..n]);
             }
             break :blk dst.ptr;
         } else null;