> Le 12 sept. 2025 à 18:57, Elijah Newren <newren@xxxxxxxxx> a écrit :
>
> On Fri, Sep 12, 2025 at 6:38 AM Guo Tingsheng <CoriCraft16@xxxxxxxxxxx> wrote:
>>
>> Hello Git developers,
>>
>> I would like to report a potential issue in Git's merge behavior, where semantically independent changes at the same position are reported as a conflict, even though they could be merged automatically.
>>
>> Environment:
>> - git version: 2.43.0
>> - OS: Ubuntu 24.04 LTS
>>
>> Steps to reproduce:
>> 1. Start with a file containing only:
>>
>> public class Calculator {
>> }
>>
>> 2. On branch A, add a new method `add`:
>>
>> @@ -1,1 +1,3 @@
>> public class Calculator {
>> + public static double add(double a, double b) {
>> + return a + b;
>> + }
>>
>> 3. On branch B, add a new method `subtract`:
>>
>> @@ -1,1 +1,3 @@
>> public class Calculator {
>> + public static double subtract(double a, double b) {
>> + return a - b;
>> + }
>>
>> 4. Merge branch A and branch B.
>>
>> Expected result:
>> - The merge should succeed automatically, producing a file that contains both methods (order does not matter).
>> For example:
>>
>> public class Calculator {
>> public static double add(double a, double b) {
>> return a + b;
>> }
>> public static double subtract(double a, double b) {
>> return a - b;
>> }
>> }
>>
>> Actual result:
>> - Git reports a conflict and aborts the merge, requiring manual conflict resolution.
>>
>> Additional information:
>> - Although the two changes occur at the same location in the file, they are independent additions with no semantic overlap.
>> - It would be desirable for Git's merge algorithm to automatically combine such changes, as the final merged state is deterministic and conflict-free.
>> - This limitation may impact developer productivity in real-world projects where multiple contributors extend the same class or configuration file independently.
>
> Thanks for the report, but how is Git supposed to know that they are
> independent changes with no semantic overlap? Git doesn't understand
> the semantics of the files it tracks. It has no idea whether the
> things being added are functions, or statements/expressions within a
> function, or college essays, or hand-written recipes, or data dumps,
> or anything else. It would need to know those semantics, which would
> probably require something on the level of use of AI to determine that
> the changes are independent non-competing additions with no semantic
> overlap and that you want to keep both.
Sorry, can’t resist: we’ve been writing semantic analysis code for decades. It wouldn’t take AI to get this right.
What’s hard about the situation is that it doesn’t generalize easily: a semantic analyzer that can resolve conflicts for one file, project, language, etc., is not necessarily of any help to its brethren.
Still, if you write a lot of Java or similar code and have this problem frequently, building a merge strategy option or mergetool you can use to resolve this easily seems worthwhile. (Not something mainline Git needs, of course, for all the reasons pointed out here.)
