Re: [PATCH] type_from_string_gently: make sure length matches

Jeff King [off-list ref] writes:

When commit fe8e3b7 refactored type_from_string to allow
input that was not NUL-terminated, it switched to using
strncmp instead of strcmp. But this means we check only the
first "len" bytes of the strings, and ignore any remaining
bytes in the object_type_string. We should make sure that it
is also "len" bytes, or else we would accept "comm" as
"commit", and so forth.

Signed-off-by: Jeff King <redacted>
---
Since the strings we are matching are literals, we could also record
their sizes in the object_type_strings array and check the length first
before even calling strncmp. I doubt this is a performance hot-spot,
though.

You could also potentially just use strlen(object_type_strings[i]), but
I'm not sure if compilers will optimize out the strlen in this case,
since it is in a loop.

That thought crossed my mind while reading your patch.  It could
even make it go faster if we made object_type_strings into an array
of counted strings (i.e. "struct { const char *str; int len; }")
and then took advantage of the fact that we have lengths of both.


 object.c | 24 ++++++++++++++----------
 1 file changed, 14 insertions(+), 10 deletions(-)

diff --git a/object.c b/object.c
index aedac24..51584ea 100644
--- a/object.c
+++ b/object.c

@@ -18,19 +18,22 @@ struct object *get_indexed_object(unsigned int idx)
 	return obj_hash[idx];
 }
 
-static const char *object_type_strings[] = {
-	NULL,		/* OBJ_NONE = 0 */
-	"commit",	/* OBJ_COMMIT = 1 */
-	"tree",		/* OBJ_TREE = 2 */
-	"blob",		/* OBJ_BLOB = 3 */
-	"tag",		/* OBJ_TAG = 4 */
+static struct {
+	const char *str;
+	int len;
+} object_type_name[] = {
+	{ NULL, 0 },      /* OBJ_NONE = 0 */
+	{ "commit", 6 },  /* OBJ_COMMIT = 1 */
+	{ "tree", 4 },	  /* OBJ_TREE = 2 */
+	{ "blob", 4 },    /* OBJ_BLOB = 3 */
+	{ "tag", 3 },     /* OBJ_TAG = 4 */
 };
 
 const char *typename(unsigned int type)
 {
-	if (type >= ARRAY_SIZE(object_type_strings))
+	if (type >= ARRAY_SIZE(object_type_name))
 		return NULL;
-	return object_type_strings[type];
+	return object_type_name[type].str;
 }
 
 int type_from_string_gently(const char *str, ssize_t len, int gentle)

@@ -40,8 +43,9 @@ int type_from_string_gently(const char *str, ssize_t len, int gentle)
 	if (len < 0)
 		len = strlen(str);
 
-	for (i = 1; i < ARRAY_SIZE(object_type_strings); i++)
-		if (!strncmp(str, object_type_strings[i], len))
+	for (i = 1; i < ARRAY_SIZE(object_type_name); i++)
+		if (object_type_name[i].len == len &&
+			    !strncmp(str, object_type_name[i].str, len))
 			return i;
 
 	if (gentle)

On Fri, Apr 17, 2015 at 01:54:27PM -0700, Junio C Hamano wrote:

quoted

Since the strings we are matching are literals, we could also record
their sizes in the object_type_strings array and check the length first
before even calling strncmp. I doubt this is a performance hot-spot,
though.

You could also potentially just use strlen(object_type_strings[i]), but
I'm not sure if compilers will optimize out the strlen in this case,
since it is in a loop.

That thought crossed my mind while reading your patch.  It could
even make it go faster if we made object_type_strings into an array
of counted strings (i.e. "struct { const char *str; int len; }")
and then took advantage of the fact that we have lengths of both.

Right, that was what I meant.

I'd be surprised if it appreciably speeds things up, but I guess it is
not too complicated to do.

+static struct {
+	const char *str;
+	int len;
+} object_type_name[] = {
+	{ NULL, 0 },      /* OBJ_NONE = 0 */
+	{ "commit", 6 },  /* OBJ_COMMIT = 1 */
+	{ "tree", 4 },	  /* OBJ_TREE = 2 */
+	{ "blob", 4 },    /* OBJ_BLOB = 3 */
+	{ "tag", 3 },     /* OBJ_TAG = 4 */
 };

I had envisioned a macro like:

  #define SIZED_STRING(x) { (x), (sizeof(x) - 1) }

though perhaps that is overkill for such a short list (that we don't
even expect to change).

-Peff