Memory layout change for #[repr(C)] struct which contains Option<T>

[aureliar8]

Code

Given this type FwRuleSet

#[repr(C)]
#[derive(Clone, PartialEq, Debug)]
pub struct FwRuleSet {
    pub rules: [Option<FwRule>; MAX_FW_RULES],
}


pub const MAX_FW_RULES: usize = 10;
pub const MAX_PREDICATES_PER_FW_RULE: usize = 2;

#[repr(C)]
#[derive(Default, Debug, PartialEq, Clone)]
pub struct FwRule {
    pub predicates: [Option<FwPredicate>; MAX_PREDICATES_PER_FW_RULE],
}

#[repr(C, u32)]
#[derive(Debug, PartialEq, Clone)]
pub enum FwPredicate {
    // We manually set the enum discriminant to facilitate interop
    InnerDstIpv4(u32) = 1,
    InnerSrcIpv4(u32) = 2,
    InnerDstIpv6(u64) = 3,
    InnerSrcIpv6(u64) = 4,
    InnerIpProto(u8) = 5,
    InnerDstUdp(u32) = 6,
    InnerSrcUdp(u32) = 7,
    InnerDstTcp(u32) = 8,
    InnerSrcTcp(u32) = 9,
    InnerIcmpv4(u32) = 10,
}

The memory representation of the values which contains None changed

let rs = FwRuleSet {
        rules: [const { None }; MAX_FW_RULES],
    };

From

0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0B 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

To

FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
FF FF FF FF 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

In nightly-2026-04-27 and latter

Version it worked on

It most recently worked on: nightly-2026-04-26 or the current stable version

Version with regression

nightly-2026-04-27

More details

I think this is due to #155473
But I'm surprised the memory layout of #[repr(C)] struct can change like that between compiler version

To quickly reproduce you can run the following

git clone git@github.com:aureliar8/memory-layout-bug-reproduction.git
cd memory-layout-bug-reproduction
./run.sh cargo +nightly-2026-04-26 run
./run.sh cargo +nightly-2026-04-27 run
./run.sh cargo  run

@rustbot modify labels: +regression-from-stable-to-nightly -regression-untriaged

[theemathas]

This was changed in #155473 cc @scottmcm

#[repr(C)] on a struct that contains non-#[repr(C)] is not sufficient to guarantee a fixed layout. You need to have #[repr(C)] or something similar on each of the field types recursively too. In particular, Option is not #[repr(C)].

I'd say that your code is relying on something that's never been promised to be stable.

[theemathas]

Actually, I'm not 100% sure this is intentional. My understanding of the linked PR is that Option is supposed to still use None=0 when it can, on a best-effort basis.