2554
For true distributed development that supports proper merging,
2555
published branches should never be rewritten.
2556
2557
+[[bisect-merges]]
2558
+Why bisecting merge commits can be harder than bisecting linear history
2559
+-----------------------------------------------------------------------
2560
+
2561
+The gitlink:git-bisect[1] command correctly handles history that
2562
+includes merge commits. However, when the commit that it finds is a
2563
+merge commit, the user may need to work harder than usual to figure out
2564
+why that commit introduced a problem.
2565
+
2566
+Imagine this history:
2567
+
2568
+................................................
2569
+ ---Z---o---X---...---o---A---C---D
2570
+ \ /
2571
+ o---o---Y---...---o---B
2572
+................................................
2573
+
2574
+Suppose that on the upper line of development, the meaning of one
2575
+of the functions that exists at Z is changed at commit X. The
2576
+commits from Z leading to A change both the function's
2577
+implementation and all calling sites that exist at Z, as well
2578
+as new calling sites they add, to be consistent. There is no
2579
+bug at A.
2580
+
2581
+Suppose that in the meantime on the lower line of development somebody
2582
+adds a new calling site for that function at commit Y. The
2583
+commits from Z leading to B all assume the old semantics of that
2584
+function and the callers and the callee are consistent with each
2585
+other. There is no bug at B, either.
2586
+
2587
+Suppose further that the two development lines merge cleanly at C,
2588
+so no conflict resolution is required.
2589
+
2590
+Nevertheless, the code at C is broken, because the callers added
2591
+on the lower line of development have not been converted to the new
2592
+semantics introduced on the upper line of development. So if all
2593
+you know is that D is bad, that Z is good, and that
2594
+gitlink:git-bisect[1] identifies C as the culprit, how will you
2595
+figure out that the problem is due to this change in semantics?
2596
+
2597
+When the result of a git-bisect is a non-merge commit, you should
2598
+normally be able to discover the problem by examining just that commit.
2599
+Developers can make this easy by breaking their changes into small
2600
+self-contained commits. That won't help in the case above, however,
2601
+because the problem isn't obvious from examination of any single
2602
+commit; instead, a global view of the development is required. To
2603
+make matters worse, the change in semantics in the problematic
2604
+function may be just one small part of the changes in the upper
2605
+line of development.
2606
+
2607
+On the other hand, if instead of merging at C you had rebased the
2608
+history between Z to B on top of A, you would have gotten this
2609
+linear history:
2610
+
2611
+................................................................
2612
+ ---Z---o---X--...---o---A---o---o---Y*--...---o---B*--D*
2613
+................................................................
2614
+
2615
+Bisecting between Z and D* would hit a single culprit commit Y*,
2616
+and understanding why Y* was broken would probably be easier.
2617
+
2618
+Partly for this reason, many experienced git users, even when
2619
+working on an otherwise merge-heavy project, keep the history
2620
+linear by rebasing against the latest upstream version before
2621
+publishing.
2622
+
2623
[[advanced-branch-management]]
2624
Advanced branch management
2625
==========================