UEFI.pool_allocator.resize() Unreachable Code when length and capacity match on an ArrayList

[dje4321]

Zig Version

0.13.0

Steps to Reproduce and Observed Behavior

Working on a UEFI zig project and keep hitting a random panic when trying to get an ownedSlice from an arraylist. Eventually narrowed it down to only be occurring when capacity and length match on an array list. Manually incrementing the capacity by 1 causes the panic branch to be missed.

The following code will reliably hit the panic branch on my machine. I know its not perfect, pretty, or correct however, it does its job just fine. It could probably be cleaned up by a fair bit but this is what I was able to quickly throw together to get around the compiler optimizing away the code.

    const list = std.ArrayList(*u8);
    var u8_list = list.init(std.os.uefi.pool_allocator);
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(8))))) catch @panic(":(");
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(1))))) catch @panic(":(");
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(2))))) catch @panic(":(");
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(3))))) catch @panic(":(");
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(4))))) catch @panic(":(");
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(5))))) catch @panic(":(");
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(6))))) catch @panic(":(");
    u8_list.append(@constCast(@as(*u8, (@ptrFromInt(7))))) catch @panic(":(");
    // Len & Cap are now both 8

    const s = u8_list.toOwnedSlice() catch @panic("Failure");
    for (s) |t| {
        _ = t;
    }
    @panic("Should have hit unreachable code");

The code panics with the msg reached unreachable code. Digging around the code further revealed it to be caused by the resize method being performed in the toOwnedSlice() call allocator.resize(old_memory, self.items.len)

https://ziglang.org/documentation/master/std/#src/std/array_list.zig

Currently dont have any way to debug zig UEFI applications so this is as far as I was able to take this atm.

Expected Behavior

Standard library should not panic

[truemedian]

This bug probably exhibits itself in a lot of different scenarios. pool_allocator and raw_pool_allocator pass log2_old_ptr_align to mem.alignAllocLen, which expects len_align (which would be 2^log2_old_ptr_align).

Further: neither of these resize functions has any reason to call mem.alignAllocLen because it immediately throws away the result.