Re: [PATCH 05/20] btf: stop resolving cyclic type references

[Nick Alcock <nix@esperi.org.uk>]

On 30 Oct 2026, Kris Van Hees told this:

> --- a/libdtrace/dt_btf.c
> +++ b/libdtrace/dt_btf.c
> @@ -100,8 +100,14 @@ static btf_type_t *
>  dt_btf_real_type_by_id(const dt_btf_t *btf, int32_t id)
>  {
>  	btf_type_t	*type = dt_btf_type_by_id(btf, id);
> +	uint64_t	remaining = 1 + (uint64_t)btf->type_cnt;
>  
> -	while (type != NULL) {
> +	if (btf->parent != NULL)
> +		remaining += btf->parent->type_cnt;
> +
> +	/* A longer chain must revisit a type, so reject cyclic BTF references. */
> +	while (type != NULL && remaining > 0) {
> +		remaining--;
>  		switch (BTF_INFO_KIND(type->info)) {
>  		case BTF_KIND_CONST:
>  		case BTF_KIND_FUNC:

That's a fun approach. Simple and entirely different from what I'm doing
in a yet-to-be-pushed libctf commit :) (which involves a counter which
starts at 1 and is reduced on every lookup, and whenever it hits 0 an
identity-check is done and the counter is reset with an increased
initial value: a tortoise-and-hare-in-O(1) space-and-time algorithm of
Hellman's. Like yours, it has the advantage of being basically free
unless things go wrong.)

Reviewed-by: Nick Alcock <nix@esperi.org.uk>

-- 
NULL && (void)