On Tue, Nov 12, 2019 at 05:57:59AM -0500, Jeff King wrote:
> On Tue, Nov 12, 2019 at 01:13:58PM +0900, Junio C Hamano wrote:
>
> > >> We can't use oid_to_hex() because we don't have a 'struct object_id'
> > >> in the first place, as sha1dc only ever deals with 20 unsigned chars.
> > >
> > > Ah, you're right. I admit I am still getting up to speed on all of the
> > > new hash-agnostic versions of the various functions.
> >
> > Thanks. I've amended this one and the range diff since the pushout
> > yesterday looks like this.
>
> Thanks. This first hunk is what I would have done:
>
> > 1: 8a030f1796 ! 1: 02d21d4117 hex: drop sha1_to_hex()
> > @@ Commit message
> > hex: drop sha1_to_hex()
> >
> > There's only a single caller left of sha1_to_hex(), since everybody now
> > - uses oid_to_hex() instead. This case is in the sha1dc wrapper, where we
> > + uses hash_to_hex() instead. This case is in the sha1dc wrapper, where we
> > print a hex sha1 when we find a collision. This one will always be sha1,
> > - regardless of the current hash algorithm, so we can't use oid_to_hex()
> > + regardless of the current hash algorithm, so we can't use hash_to_hex()
> > here. In practice we'd probably not be running sha1 at all if it isn't
> > the current algorithm, but it's possible we might still occasionally
> > need to compute a sha1 in a post-sha256 world.
>
> This second one is OK, but not entirely necessary:
>
> > @@ cache.h: int hex_to_bytes(unsigned char *binary, const char *hex, size_t len);
> > * buffers, making it safe to make multiple calls for a single statement, like:
> > *
> > - * printf("%s -> %s", sha1_to_hex(one), sha1_to_hex(two));
> > -+ * printf("%s -> %s", oid_to_hex(one), oid_to_hex(two));
> > ++ * printf("%s -> %s", hash_to_hex(one), hash_to_hex(two));
> > */
> > char *hash_to_hex_algop_r(char *buffer, const unsigned char *hash, const struct git_hash_algo *);
> > char *oid_to_hex_r(char *out, const struct object_id *oid);
>
> This one-liner leaves the types of "one" and "two" unspecified. :) So
> it's not wrong to use hash_to_hex(), but maybe it's better to be pushing
> people towards oid_to_hex() as their first choice? It probably doesn't
> matter too much either way.
Yeah, most (over 96%) of the hashes that we print are actually object
ids, so oid_to_hex() is the right function to use most of the time.
And because of this the updated "since everybody uses hash_to_hex()"
in the first hunk sounds a bit wrong, because barely anybody actually
uses hash_to_hex().
