Merge pull request libgit2#3921 from libgit2/cmn/walk-limit-enough

Improve revision walk preparation logic

Author: Edward Thomson

CHANGELOG.md

@@ -15,6 +15,8 @@ v0.24 + 1
 
 * Support for reading and writing git index v4 files
 
+* Improve the performance of the revwalk and bring us closer to git's code.
+
 ### API additions
 
 * You can now get the user-agent used by libgit2 using the

include/git2/revwalk.h

@@ -25,17 +25,15 @@ GIT_BEGIN_DECL
  */
 typedef enum {
 	/**
-	 * Sort the repository contents in no particular ordering;
-	 * this sorting is arbitrary, implementation-specific
-	 * and subject to change at any time.
+	 * Sort the output with the same default time-order method from git.
 	 * This is the default sorting for new walkers.
 	 */
 	GIT_SORT_NONE = 0,
 
 	/**
-	 * Sort the repository contents in topological order
-	 * (parents before children); this sorting mode
-	 * can be combined with time sorting.
+	 * Sort the repository contents in topological order (parents before
+	 * children); this sorting mode can be combined with time sorting to
+	 * produce git's "time-order".
 	 */
 	GIT_SORT_TOPOLOGICAL = 1 << 0,
 

src/commit_list.c

@@ -13,10 +13,15 @@
 
 int git_commit_list_time_cmp(const void *a, const void *b)
 {
-	const git_commit_list_node *commit_a = a;
-	const git_commit_list_node *commit_b = b;
+	int64_t time_a = ((git_commit_list_node *) a)->time;
+	int64_t time_b = ((git_commit_list_node *) b)->time;
 
-	return (commit_a->time < commit_b->time);
+	if (time_a < time_b)
+		return 1;
+	if (time_a > time_b)
+		return -1;
+
+	return 0;
 }
 
 git_commit_list *git_commit_list_insert(git_commit_list_node *item, git_commit_list **list_p)

src/commit_list.h

@@ -28,6 +28,7 @@ typedef struct git_commit_list_node {
 			 uninteresting:1,
 			 topo_delay:1,
 			 parsed:1,
+			 added:1,
 			 flags : FLAG_BITS;
 
 	unsigned short in_degree;

src/pqueue.c

@@ -93,17 +93,23 @@ int git_pqueue_insert(git_pqueue *pq, void *item)
 		(void)git_pqueue_pop(pq);
 	}
 
-	if (!(error = git_vector_insert(pq, item)))
+	if (!(error = git_vector_insert(pq, item)) && pq->_cmp)
 		pqueue_up(pq, pq->length - 1);
 
 	return error;
 }
 
 void *git_pqueue_pop(git_pqueue *pq)
 {
-	void *rval = git_pqueue_get(pq, 0);
+	void *rval;
 
-	if (git_pqueue_size(pq) > 1) {
+	if (!pq->_cmp) {
+		rval = git_vector_last(pq);
+	} else {
+		rval = git_pqueue_get(pq, 0);
+	}
+
+	if (git_pqueue_size(pq) > 1 && pq->_cmp) {
 		/* move last item to top of heap, shrink, and push item down */
 		pq->contents[0] = git_vector_last(pq);
 		git_vector_pop(pq);

src/pqueue.h

@@ -35,6 +35,7 @@ extern int git_pqueue_init(
 #define git_pqueue_clear git_vector_clear
 #define git_pqueue_size  git_vector_length
 #define git_pqueue_get   git_vector_get
+#define git_pqueue_reverse git_vector_reverse
 
 /**
  * Insert a new item into the queue

src/rebase.c

@@ -586,7 +586,7 @@ static int rebase_init_operations(
 		(error = git_revwalk_hide(revwalk, git_annotated_commit_id(upstream))) < 0)
 		goto done;
 
-	git_revwalk_sorting(revwalk, GIT_SORT_REVERSE | GIT_SORT_TIME);
+	git_revwalk_sorting(revwalk, GIT_SORT_REVERSE);
 
 	while ((error = git_revwalk_next(&id, revwalk)) == 0) {
 		if ((error = git_commit_lookup(&commit, repo, &id)) < 0)