git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[JGIT PATCH 08/14] Optimize path comparsion within subtrees during TreeWalk

From
Shawn O. Pearce <spearce@spearce.org>
Date
Aug 18, 2008, 23:53 UTC
Message-ID
<1219103602-32222-9-git-send-email-spearce@spearce.org>
In-Reply-To
<1219103602-32222-8-git-send-email-spearce@spearce.org>

If we are comparing two entries whose parents both match the same tree iterator we know the path up through pathOffset must all be identical, as the parents can only match if their paths up to pathOffset were equal and they were both trees.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
---
 .../jgit/treewalk/AbstractTreeIterator.java        |   22 +++++++++++++++++++-
 1 files changed, 21 insertions(+), 1 deletions(-)
diff --git a/org.spearce.jgit/src/org/spearce/jgit/treewalk/AbstractTreeIterator.java b/org.spearce.jgit/src/org/spearce/jgit/treewalk/AbstractTreeIterator.java
index 31257b5..e6aa338 100644
--- a/org.spearce.jgit/src/org/spearce/jgit/treewalk/AbstractTreeIterator.java
+++ b/org.spearce.jgit/src/org/spearce/jgit/treewalk/AbstractTreeIterator.java
@@ -237,7 +237,13 @@ int pathCompare(final AbstractTreeIterator p, final int pMode) {
 		final int bLen = p.pathLen;
 		int cPos;
 
-		for (cPos = 0; cPos < aLen && cPos < bLen; cPos++) {
+		// Its common when we are a subtree for both parents to match;
+		// when this happens everything in path[0..cPos] is known to
+		// be equal and does not require evaluation again.
+		//
+		cPos = alreadyMatch(this, p);
+
+		for (; cPos < aLen && cPos < bLen; cPos++) {
 			final int cmp = (a[cPos] & 0xff) - (b[cPos] & 0xff);
 			if (cmp != 0)
 				return cmp;
@@ -250,6 +256,20 @@ int pathCompare(final AbstractTreeIterator p, final int pMode) {
 		return lastPathChar(mode) - lastPathChar(pMode);
 	}
 
+	private static int alreadyMatch(AbstractTreeIterator a,
+			AbstractTreeIterator b) {
+		for (;;) {
+			final AbstractTreeIterator ap = a.parent;
+			final AbstractTreeIterator bp = b.parent;
+			if (ap == null || bp == null)
+				return 0;
+			if (ap.matches == bp.matches)
+				return a.pathOffset;
+			a = ap;
+			b = bp;
+		}
+	}
+
 	private static int lastPathChar(final int mode) {
 		return FileMode.TREE.equals(mode) ? '/' : '\0';
 	}
-- 
1.6.0.87.g2858d
Previous: Shawn O. PearceNext: Shawn O. Pearce
Message 9 of 20 in “TreeWalk D/F conflict detection”
  1. 00/14 TreeWalk D/F conflict detectionShawn O. Pearce, Aug 18, 2008
  2. 01/14 Detect path names which overflow the name length field in the indexShawn O. Pearce, Aug 18, 2008
  3. 02/14 Fix NB.decodeUInt16 to correctly handle the high byteShawn O. Pearce, Aug 18, 2008
  4. 03/14 Add test cases for NB.encode and NB.decode family of routinesShawn O. Pearce, Aug 18, 2008
  5. 04/14 Fix DirCache's skip over null byte padding when reading a DIRC fileShawn O. Pearce, Aug 18, 2008
  6. 05/14 Fix usage of assertEquals in DirCacheIteratorTestShawn O. Pearce, Aug 18, 2008
  7. 06/14 Refactor AbstractTreeIterator.pathCompare to force another modeShawn O. Pearce, Aug 18, 2008
  8. 07/14 Micro-optimize AbstractTreeIterator.pathCompareShawn O. Pearce, Aug 18, 2008
  9. 08/14 Optimize path comparsion within subtrees during TreeWalkShawn O. Pearce, Aug 18, 2008
  10. 09/14 Refactor AbstractTreeIterator semantics to start on first entryShawn O. Pearce, Aug 18, 2008
  11. 10/14 Make all AbstractTreeIterator implementations bi-directionalShawn O. Pearce, Aug 18, 2008
  12. 11/14 Expose beginning of iterator indication from AbstractTreeIteratorShawn O. Pearce, Aug 18, 2008
  13. 12/14 Allow application code to set ObjectIds in DirCacheEntryShawn O. Pearce, Aug 18, 2008
  14. 13/14 Create NameConflictTreeWalk to transparently detect D/F conflictsShawn O. Pearce, Aug 18, 2008
  15. 14/14 Add test case for NameConflictTreeWalkShawn O. Pearce, Aug 18, 2008
  16. Junio C HamanoAug 19, 2008
  17. Robin RosenbergAug 19, 2008
  18. Shawn O. PearceAug 19, 2008
  19. David WoodhouseAug 19, 2008
  20. Shawn O. PearceAug 19, 2008

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.