Web lists-archives.com

Re: [PATCH 3/5] match-trees: use hashcpy to splice trees

On Thu, Jan 10, 2019 at 04:25:49AM +0000, brian m. carlson wrote:

> When we're splicing trees, we're writing directly from one location into
> a buffer that is exactly the same size as a tree object. If the current
> hash algorithm is SHA-1, we may not have a full 32 (GIT_MAX_RAWSZ) bytes
> available to write, nor would we want to write that many bytes even if
> we did. In a future commit, we'll split out hashcpy to respect
> the_hash_algo while oidcpy uses GIT_MAX_RAWSZ, so convert the oidcpy to
> a hashcpy so we copy the right number of bytes.
> Signed-off-by: brian m. carlson <sandals@xxxxxxxxxxxxxxxxxxxx>
> ---
>  match-trees.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> diff --git a/match-trees.c b/match-trees.c
> index feca48a5fd..b1fbd022d1 100644
> --- a/match-trees.c
> +++ b/match-trees.c
> @@ -224,7 +224,7 @@ static int splice_tree(const struct object_id *oid1, const char *prefix,
>  	} else {
>  		rewrite_with = oid2;
>  	}
> -	oidcpy(rewrite_here, rewrite_with);
> +	hashcpy(rewrite_here->hash, rewrite_with->hash);

Hrm. Our coccinelle patches will want to convert this back to oidcpy(),
though I see you drop them in the final patch.

However, I wonder if it points to another mismatch. Isn't the point that
we _don't_ actually have "struct object_id"s here? I.e., rewrite_here
and rewrite_with should actually be "const unsigned char *" that we
happen to know are the_hash_algo->raw_sz?

I think the only reason they are "struct object_id" is because that's
what tree_entry_extract() returns. Should we be providing another
function to allow more byte-oriented access?