Re: [PATCH 04/12] btf_encoder: Encode variant parts as union members in BTF

On 7/31/26 12:30 PM, Arnaldo Carvalho de Melo wrote:
> From: Arnaldo Carvalho de Melo <[email protected]>
>
> With the DWARF loader now populating DW_TAG_variant children (previous
> commit), wire them into BTF encoding so Rust discriminated unions
> (Option<T>, Result<T,E>, etc.) are no longer emitted as empty structs.
>
> Two changes:
>
> 1. Struct-to-union promotion: when a DW_TAG_structure_type has
>     variant_parts but no regular data members, encode it as BTF_KIND_UNION
>     instead of BTF_KIND_STRUCT, since the variants overlap at offset 0.
>
> 2. Variant member encoding: after encoding regular data members, iterate
>     the variant_parts and emit each variant as a BTF union field with the
>     variant's name and resolved type reference.
>
> Testing with the sashiko-cli Rust binary (a real-world async HTTP client
> using tokio, hyper, serde, etc.):
>
> Before:
>
>    $ bpftool btf dump file sashiko-cli | grep -c UNION
>    2073
>    $ bpftool btf dump file sashiko-cli | grep 'STRUCT.*vlen=0' | grep -vc 'size=0'
>    24335
>
> After:
>
>    $ bpftool btf dump file sashiko-cli | grep -c UNION
>    25750
>    $ bpftool btf dump file sashiko-cli | grep 'STRUCT.*vlen=0' | grep -vc 'size=0'
>    2236
>
> 22,099 types that were previously encoded as empty structs are now
> properly represented as unions with their variant members:
>
> Before:
>
>    $ bpftool btf dump file code_with_type.o | grep -A1 'Option<u32>'
>    [11] STRUCT 'Option<u32>' size=8 vlen=0
>
> After:
>
>    $ bpftool btf dump file code_with_type.o | grep -A3 'Option<u32>'
>    [11] UNION 'Option<u32>' size=8 vlen=2
>            'None' type_id=9 bits_offset=0
>            'Some' type_id=10 bits_offset=0

FYI, the below is for llvm bpf backend to handle DW_TAG_variant_part
to generate BTF.
   link: https://github.com/llvm/llvm-project/pull/155783
   
For a rust code like:

; Source:
;   #![no_std]
;   #![no_main]
;
;   pub enum MyEnum {
;       First { a: u32, b: i32 },
;       Second(u32),
;   }
;
;   #[unsafe(no_mangle)]
;   pub static X: MyEnum = MyEnum::First { a: 54, b: -23 };
;
;   #[cfg(not(test))]
;   #[panic_handler]
;   fn panic(_info: &core::panic::PanicInfo) -> ! {
;       loop {}
;   }

The BTF encoding:

; CHECK-BTF:      [1] STRUCT 'MyEnum' size=12 vlen=1
; CHECK-BTF-NEXT:         '(anon)' type_id=3 bits_offset=0
; CHECK-BTF-NEXT: [2] INT 'u32' size=4 bits_offset=0 nr_bits=32 encoding=(none)
; CHECK-BTF-NEXT: [3] UNION '(anon)' size=12 vlen=3
; CHECK-BTF-NEXT:         '(anon)' type_id=2 bits_offset=0
; CHECK-BTF-NEXT:         'First' type_id=4 bits_offset=0
; CHECK-BTF-NEXT:         'Second' type_id=6 bits_offset=0
; CHECK-BTF-NEXT: [4] STRUCT 'First' size=12 vlen=2
; CHECK-BTF-NEXT:         'a' type_id=2 bits_offset=32
; CHECK-BTF-NEXT:         'b' type_id=5 bits_offset=64
; CHECK-BTF-NEXT: [5] INT 'i32' size=4 bits_offset=0 nr_bits=32 encoding=SIGNED
; CHECK-BTF-NEXT: [6] STRUCT 'Second' size=12 vlen=1
; CHECK-BTF-NEXT:         '__0' type_id=2 bits_offset=32
; CHECK-BTF-NEXT: [7] VAR 'X' type_id=1, linkage=global
; CHECK-BTF-NEXT: [8] DATASEC '.rodata' size=0 vlen=1
; CHECK-BTF-NEXT:         type_id=7 offset=0 size=12

>
> Assisted-by: Claude:claude-sonnet-4-5
> Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
> ---
>   btf_encoder.c | 37 +++++++++++++++++++++++++++++++++++--
>   1 file changed, 35 insertions(+), 2 deletions(-)
>
> diff --git a/btf_encoder.c b/btf_encoder.c
> index 83ce21186ea7da36..acf0a5ade5e29014 100644
> --- a/btf_encoder.c
> +++ b/btf_encoder.c
> @@ -1855,6 +1855,12 @@ static void dump_invalid_symbol(const char *msg, const char *sym,
>   	fprintf(stderr, "PAHOLE: Error: Use '--btf_encode_force' to ignore such symbols and force emit the btf.\n");
>   }
>   
> +static bool type__has_variant_parts(const struct type *type)
> +{
> +	return !list_empty(&type->variant_parts);
> +}
> +
> +
>   static int32_t btf_encoder__add_struct_type(struct btf_encoder *encoder, struct tag *tag)
>   {
>   	struct type *type = tag__type(tag);
> @@ -1863,8 +1869,18 @@ static int32_t btf_encoder__add_struct_type(struct btf_encoder *encoder, struct
>   	int32_t type_id;
>   	uint8_t kind;
>   
> -	kind = (tag->tag == DW_TAG_union_type) ?
> -		BTF_KIND_UNION : BTF_KIND_STRUCT;
> +	/*
> +	 * Rust discriminated unions (enums) are represented in DWARF as
> +	 * DW_TAG_structure_type with DW_TAG_variant_part children.
> +	 * If the struct has only variant parts and no regular data members,
> +	 * encode it as a BTF union since the variants overlap at offset 0.
> +	 */
> +	if (tag->tag == DW_TAG_union_type)
> +		kind = BTF_KIND_UNION;
> +	else if (type__has_variant_parts(type) && type->nr_members == 0)
> +		kind = BTF_KIND_UNION;
> +	else
> +		kind = BTF_KIND_STRUCT;
>   
>   	type_id = btf_encoder__add_struct(encoder, kind, name, type->size);
>   	if (type_id < 0)
> @@ -1882,6 +1898,23 @@ static int32_t btf_encoder__add_struct_type(struct btf_encoder *encoder, struct
>   			return -1;
>   	}
>   
> +	if (type__has_variant_parts(type) && kind == BTF_KIND_UNION) {
> +		struct variant_part *vpart;
> +
> +		type__for_each_variant_part(type, vpart) {
> +			struct variant *variant;
> +
> +			variant_part__for_each_variant(vpart, variant) {
> +				if (variant->tag.type == 0)
> +					continue;
> +
> +				uint32_t ref_type_id = btf_encoder__tag_type(encoder, variant->tag.type);
> +
> +				if (btf_encoder__add_field(encoder, variant->name, ref_type_id, 0, 0))
> +					return -1;
> +			}
> +		}
> +	}
>   
>   
>   	return type_id;