aboutsummaryrefslogtreecommitdiffstats
path: root/remote.h
diff options
context:
space:
mode:
authorJacob Keller <jacob.keller@gmail.com>2025-06-23 16:11:29 -0700
committerJunio C Hamano <gitster@pobox.com>2025-06-23 16:38:55 -0700
commitf62dcc7f30d16af29c0f707005aceb5eb6119279 (patch)
tree7251fb69e14e76d62832974534637b804485847f /remote.h
parent16bd9f20a403117f2e0d9bcda6c6e621d3763e77 (diff)
downloadgit-f62dcc7f30d16af29c0f707005aceb5eb6119279.tar.gz
remote: remove branch->merge_name and fix branch_release()
The branch structure has both branch->merge_name and branch->merge for tracking the merge information. The former is allocated by add_merge() and stores the names read from the configuration file. The latter is allocated by set_merge() which is called by branch_get() when an external caller requests a branch. This leads to the confusing situation where branch->merge_nr tracks both the size of branch->merge (once its allocated) and branch->merge_name. The branch_release() function incorrectly assumes that branch->merge is always set when branch->merge_nr is non-zero, and can potentially crash if read_config() is called without branch_get() being called on every branch. In addition, branch_release() fails to free some of the memory associated with the structure including: * Failure to free the refspec_item containers in branch->merge[i] * Failure to free the strings in branch->merge_name[i] * Failure to free the branch->merge_name parent array. The set_merge() function sets branch->merge_nr to 0 when there is no valid remote_name, to avoid external callers seeing a non-zero merge_nr but a NULL merge array. This results in failure to release most of the merge data as well. These issues could be fixed directly, and indeed I initially proposed such a change at [1] in the past. While this works, there was some confusion during review because of the inconsistencies. Instead, its time to clean up the situation properly. Remove branch->merge_name entirely. Instead, allocate branch->merge earlier within add_merge() instead of within set_merge(). Instead of having set_merge() copy from merge_name[i] to merge[i]->src, just have add_merge() directly initialize merge[i]->src. Modify the add_merge() to call xstrdup() itself, instead of having the caller of add_merge() do so. This makes it more obvious which code owns the memory. Update all callers which use branch->merge_name[i] to use branch->merge[i]->src instead. Add a merge_clear() function which properly releases all of the merge-related memory, and which sets branch->merge_nr to zero. Use this both in branch_release() and in set_merge(), fixing the leak when set_merge() finds no valid remote_name. Add a set_merge variable to the branch structure, which indicates whether set_merge() has been called. This replaces the previous use of a NULL check against the branch->merge array. With these changes, the merge array is always allocated when merge_nr is non-zero. This use of refspec_item to store the names should be safe. External callers should be using branch_get() to obtain a pointer to the branch, which will call set_merge(), and the callers internal to remote.c already handle the partially initialized refpsec_item structure safely. This end result is cleaner, and avoids duplicating the merge names twice. Signed-off-by: Jacob Keller <jacob.keller@gmail.com> Link: [1] https://lore.kernel.org/git/20250617-jk-submodule-helper-use-url-v2-1-04cbb003177d@gmail.com/ Signed-off-by: Junio C Hamano <gitster@pobox.com>
Diffstat (limited to 'remote.h')
-rw-r--r--remote.h4
1 files changed, 2 insertions, 2 deletions
diff --git a/remote.h b/remote.h
index 7e4943ae3a..76d93bf88d 100644
--- a/remote.h
+++ b/remote.h
@@ -315,8 +315,8 @@ struct branch {
char *pushremote_name;
- /* An array of the "merge" lines in the configuration. */
- const char **merge_name;
+ /* True if set_merge() has been called to finalize the merge array */
+ int set_merge;
/**
* An array of the struct refspecs used for the merge lines. That is,