In
e1111cef23 (inline lookup_replace_object() calls, 2011-05-15) a shortcut
for checking the object replacement was added by setting check_replace_refs
to 0 once the replacements were evaluated to not exist. This works fine in
with the assumption of only one repository in existence.
The assumption won't hold true any more when we work on multiple instances
of a repository structs (e.g. one struct per submodule), as the first
repository to be inspected may have no replacements and would set the
global variable. Other repositories would then completely omit their
evaluation of replacements.
This reverts back the meaning of the flag `check_replace_refs` of
"Do we need to check with the lookup table?" to "Do we need to read
the replacement definition?", adding the bypassing logic to
lookup_replace_object after the replacement definition was read.
As with the original patch, delay the renaming of the global variable
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
const char *askpass_program;
const char *excludes_file;
enum auto_crlf auto_crlf = AUTO_CRLF_FALSE;
-int check_replace_refs = 1;
+int check_replace_refs = 1; /* NEEDSWORK: rename to read_replace_refs */
char *git_replace_ref_base;
enum eol core_eol = EOL_UNSET;
int global_conv_flags_eol = CONV_EOL_RNDTRP_WARN;
#include "oidmap.h"
#include "repository.h"
+#include "object-store.h"
struct replace_object {
struct oidmap_entry original;
*/
static inline const struct object_id *lookup_replace_object(const struct object_id *oid)
{
- if (!check_replace_refs)
+ if (!check_replace_refs ||
+ (the_repository->objects->replace_map &&
+ the_repository->objects->replace_map->map.tablesize == 0))
return oid;
return do_lookup_replace_object(oid);
}
oidmap_init(the_repository->objects->replace_map, 0);
for_each_replace_ref(register_replace_ref, NULL);
-
- if (!the_repository->objects->replace_map->map.tablesize)
- check_replace_refs = 0;
}
/* We allow "recursive" replacement. Only within reason, though */