copies: move from a copy on branchpoint to a copy on write approach
authorPierre-Yves David <pierre-yves.david@octobus.net>
Tue, 15 Oct 2019 18:23:34 +0200
changeset 43300 ffd04bc9f57d
parent 43299 83bb1e89ab9b
child 43301 90213d027154
copies: move from a copy on branchpoint to a copy on write approach Before this changes, any branch points results in a copy of the dictionary containing the copy information. This can be very costly for branchy history with few rename information. Instead, we take a "copy on write" approach. Copying the input data only when we are about to update them. In practice we where already doing the copying in half of these case (because `_chain` makes a copy), so we don't add a significant cost here even in the linear case. However the speed up in branchy case is very significant. Here are some timing on the pypy repository. revision: large amount; added files: large amount; rename small amount; c3b14617fbd7 9ba6ab77fd29 before: ! wall 1.399863 comb 1.400000 user 1.370000 sys 0.030000 (median of 10) after: ! wall 0.766453 comb 0.770000 user 0.750000 sys 0.020000 (median of 11) revision: large amount; added files: small amount; rename small amount; c3b14617fbd7 f650a9b140d2 before: ! wall 1.876748 comb 1.890000 user 1.870000 sys 0.020000 (median of 10) after: ! wall 1.167223 comb 1.170000 user 1.150000 sys 0.020000 (median of 10) revision: large amount; added files: large amount; rename large amount; 08ea3258278e d9fa043f30c0 before: ! wall 0.242457 comb 0.240000 user 0.240000 sys 0.000000 (median of 39) after: ! wall 0.211476 comb 0.210000 user 0.210000 sys 0.000000 (median of 45) revision: small amount; added files: large amount; rename large amount; df6f7a526b60 a83dc6a2d56f before: ! wall 0.013193 comb 0.020000 user 0.020000 sys 0.000000 (median of 224) after: ! wall 0.013290 comb 0.010000 user 0.010000 sys 0.000000 (median of 222) revision: small amount; added files: large amount; rename small amount; 4aa4e1f8e19a 169138063d63 before: ! wall 0.001673 comb 0.000000 user 0.000000 sys 0.000000 (median of 1000) after: ! wall 0.001677 comb 0.000000 user 0.000000 sys 0.000000 (median of 1000) revision: small amount; added files: small amount; rename small amount; 4bc173b045a6 964879152e2e before: ! wall 0.000119 comb 0.000000 user 0.000000 sys 0.000000 (median of 8023) after: ! wall 0.000119 comb 0.000000 user 0.000000 sys 0.000000 (median of 7997) revision: medium amount; added files: large amount; rename medium amount; c95f1ced15f2 2c68e87c3efe before: ! wall 0.201898 comb 0.210000 user 0.200000 sys 0.010000 (median of 48) after: ! wall 0.167415 comb 0.170000 user 0.160000 sys 0.010000 (median of 58) revision: medium amount; added files: medium amount; rename small amount; d343da0c55a8 d7746d32bf9d before: ! wall 0.036820 comb 0.040000 user 0.040000 sys 0.000000 (median of 100) after: ! wall 0.035797 comb 0.040000 user 0.040000 sys 0.000000 (median of 100) The extra cost in the linear case can be reclaimed later with some extra logic. Differential Revision: https://phab.mercurial-scm.org/D7124
mercurial/copies.py
--- a/mercurial/copies.py	Sat Oct 12 18:35:14 2019 +0200
+++ b/mercurial/copies.py	Tue Oct 15 18:23:34 2019 +0200
@@ -270,15 +270,19 @@
                 childcopies = {
                     dst: src for dst, src in childcopies.items() if match(dst)
                 }
-            # Copy the dict only if later iterations will also need it
-            if i != len(children[r]) - 1:
-                newcopies = copies.copy()
-            else:
-                newcopies = copies
+            newcopies = copies
             if childcopies:
                 newcopies = _chain(newcopies, childcopies)
+                # _chain makes a copies, we can avoid doing so in some
+                # simple/linear cases.
+                assert newcopies is not copies
             for f in removed:
                 if f in newcopies:
+                    if newcopies is copies:
+                        # copy on write to avoid affecting potential other
+                        # branches.  when there are no other branches, this
+                        # could be avoided.
+                        newcopies = copies.copy()
                     del newcopies[f]
             othercopies = all_copies.get(c)
             if othercopies is None: