Re: [PATCH] drop support for "experimental" loose objects

Duy Nguyen [off-list ref] writes:

On Thu, Nov 21, 2013 at 6:48 PM, Jeff King [off-list ref] wrote:

quoted

@@ -1514,14 +1469,6 @@ unsigned long unpack_object_header_buffer(const unsigned char *buf,

 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
 {
-       unsigned long size, used;
-       static const char valid_loose_object_type[8] = {
-               0, /* OBJ_EXT */
-               1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
-               0, /* "delta" and others are invalid in a loose object */
-       };
-       enum object_type type;
-
        /* Get the data stream */
        memset(stream, 0, sizeof(*stream));
        stream->next_in = map;

@@ -1529,27 +1476,6 @@ int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long ma
        stream->next_out = buffer;
        stream->avail_out = bufsiz;

-       if (experimental_loose_object(map)) {

Perhaps keep this..

quoted

-               /*
-                * The old experimental format we no longer produce;
-                * we can still read it.
-                */
-               used = unpack_object_header_buffer(map, mapsize, &type, &size);
-               if (!used || !valid_loose_object_type[type])
-                       return -1;
-               map += used;
-               mapsize -= used;
-
-               /* Set up the stream for the rest.. */
-               stream->next_in = map;
-               stream->avail_in = mapsize;
-               git_inflate_init(stream);
-
-               /* And generate the fake traditional header */
-               stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
-                                                typename(type), size);
-               return 0;

and replace all this with

die("detected an object in obsolete format, please repack the
repository using a version before XXX");

?

Wouldn't that fail to solve the issue of `git fsck` dying on corrupt
data?  experimental_loose_object() would need to be rewritten to be more
conservative in deciding that an object was in the experimental loose
object format.

-Keshav