Fix unified diff including up to 2 * context_lines before first hunk when near input start

[mgsloan]

The bug is that when the first hunk start line is > context_lines and <= 2 * context_lines, all the lines before the hunk are included in the context.

This fixes the behavior to be the same as diff -u:

printf "a\nb\nc\nd\ne\nf\ng\nh\ni" > input1
printf "a\nb\nc\nd\nedit\nf\ng\nh\ni" > input2
diff -u input1 input2

@@ -2,7 +2,7 @@
 b
 c
 d
-e
+edit
 f
 g
 h

This came up in Zed when optimizing diffing in a case where the edited lines are known. These edited regions are then provided to imara-diff. Often this works perfectly, but when the first diff is not immediately after the context lines this bug occurs and causes extra prefix context lines to appear.

I realize that only diffing these edited regions will not always match a full input diff, but for this particular use-case it is preferable :)

[Byron]

Thanks a lot!

I could validate that the first commit fails with:

error: expect test failed
   --> src/tests.rs:235:9

You can update all `expect!` tests by running:

    env UPDATE_EXPECT=1 cargo test

To update a single test, place the cursor on `expect` token and use `run` feature of rust-analyzer.

Expect:
----
@@ -2,7 +2,7 @@
 b
 c
 d
-e
+edit
 f
 g
 h

----

Actual:
----
@@ -2,8 +2,8 @@
 a
 b
 c
 d
-e
+edit
 f
 g
 h

----

Diff:
----
@@ -2,78 +2,78 @@
 a
 b
 c
 d
-e
+edit
 f
 g
 h

----

and that it works with the second commit applied.

However, I will leave it to @pascalkuthe to fix this. The version of this in gitoxide was already fixed a while ago I believe.
The problem here remains: the whole library is quite under-tested.

[gcanat]

That's what I was trying to explain here #14 (comment) indeed.
👍

[pascalkuthe]

thanks!