2 fixes for 3-way-merge

[Thorium]

Two fixes for 3-way merging, each with a test that fails on master.
Fix 1 makes the line merge more cautious: both-sided additions that used to be merged silently are now conflicts.

1. Blocks added by both sides in the same place were merged twice

A 3-way compare combines two comparisons (BASE–LOCAL, BASE–REMOTE). An inserted block whose first line matches the line after it can sit one line higher or lower, and the two comparisons don't always pick the same position. The two additions then count as unrelated one-sided changes, and
Auto Merge takes both.

LOCAL               BASE            REMOTE
--------            --------        --------
comment 1           comment 1       comment 1
member 1            member 1        member 1
result              result          result

comment x           comment 2       comment x
member x            ...             member x
result                              result

comment y                           comment 2
member y                            ...
other

comment 2
...

Before: two differences (LOCAL only, REMOTE only); Auto Merge writes comment x / member x / result twice, with no conflict.
After: one difference, reported as a conflict (or as "both sides agree" when the two blocks are identical).

Fix: AlignInsertions() in Src/DiffWrapper.cpp moves an inserted block only when it can meet a block inserted by the other side, and only to that position. Differences the two comparisons already agree on are not touched, and blocks that are part of a detected moved block stay where they are.

Test: DiffWrapper.RunFileDiff_3way_BothSidesAddInTheSamePlace (all four diff algorithms, with and without moved-block detection).

2. /noninteractive auto-merge never exits

WinMergeU /noninteractive /minimize /am /o merged.txt LOCAL.txt BASE.txt REMOTE.txt

Before: the process stays open; closing a modified document shows the "save changes?" prompt even in an unattended run. Happens with and without the merge result pane, and for image compare. After: the merged content is written to the /o path and WinMerge exits. Unresolved conflicts in a merge result are written as conflict sections. Without /o, the changes are discarded.

Fix: CMergeDoc::PromptAndSaveIfNeeded(), CMergeDoc::SaveMergeResult() and CImgMergeFrame::PromptAndSaveIfNeeded() no longer prompt when /noninteractive is set.

Test: NonInteractiveTest in Testing/GoogleTest/GUITests (clean merge and merge with a conflict, each with the result pane on and off; image merge).

[sdottaka]

I tried it, but I can't reproduce the duplicated merge. In the merge result pane I get this:

comment 1
member 1
result

<Merge Conflict>
comment 2
...

Could you attach the LOCAL/BASE/REMOTE files that produce the duplicated comment x / member x / result on master? The example in the description looks simplified, and I assume the exact input and diff algorithm matter.

[Thorium]

Sorry, I simplified too much. Attached is a more complete example of where it should occur.

pr3671-repro.zip

Real-world case:

BASE: https://github.com/bryanedds/Nu/blob/0aaf555c4f757b93f664beb0160d9ec92a6352a6/Nu/Nu/World/WorldImNui.fs
LOCAL: https://github.com/bryanedds/Nu/blob/0de5a993bae848587df2ea7bb2b7cd32361205b6/Nu/Nu/World/WorldImNui.fs
REMOTE: https://github.com/bryanedds/Nu/blob/645dd659fd48af545ee4e09592d95de5428a4a1d/Nu/Nu/World/WorldImNui.fs

Both sides add static member doTmxMap; on master the merge result contains it twice, with no conflict. Git reports a conflict there.

[sdottaka]

Thanks for attaching the example. I was able to reproduce the issue.

Since this is related to the comparison algorithm, I'd like to think about it a little more.

[Thorium]

Yes, I tested about 500 merges on random repositories, and only 2 failed, so this isn't a common case. The main problem is that it can produce incorrect behaviour without the user possibly noticing it. Sorry to bundle 2 fixes into one PR; I thought it was the fastest way to get out of beta. Happy to leave these for you.

[Copilot]

🟡 Changes recommended

Freshly built conflict-only results can exit without writing the requested output, and the regression test currently masks this path.

2 open findings

• Check unsaved result state before unattended merge save · New
• Test pure-conflict output preservation during save · New

What changed in this PR

Improves three-way merge correctness and unattended auto-merge shutdown behavior.

Changes:

• Aligns ambiguous insertions from both sides into one conflict.
• Saves unattended text/image merge output without prompting.
• Adds regression and GUI tests.

File	Description
Src/​DiffWrapper.cpp	Aligns equivalent insertion positions.
Src/​MergeDoc.cpp	Handles unattended text saving.
Src/​MergeDocResultPane.cpp	Suppresses unattended conflict prompts.
Src/​ImgMergeFrm.cpp	Handles unattended image saving.
Testing/​GoogleTest/​DiffWrapper/​DiffWrapper_test.cpp	Tests insertion alignment.
Testing/​GoogleTest/​GUITests/​NonInteractiveTest.cpp	Tests unattended merges.
Testing/​GoogleTest/​GUITests/​GUITestUtils.cpp	Adds process execution helper.
Testing/​GoogleTest/​GUITests/​GUITestUtils.h	Declares execution helper.
Testing/​GoogleTest/​GUITests/​GUITests.vcxproj	Registers the new test.
Testing/​GoogleTest/​GUITests/​GUITests.vcxproj.filters	Adds the test to project filters.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[sdottaka]

Thanks for the interesting PR! I'll merge it.

For future PRs, it would be helpful if you could keep unrelated changes in separate PRs.

In particular, PR titles are used directly in the changelog and release notes, so a title like "2 fixes for 3-way merge" doesn't clearly describe the changes, especially when the fixes are unrelated.