Re: [PATCH] Teach git-gui to split hunks

[Junio C Hamano <gitster@pobox.com>]

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

> For this to work, apply has to be called with --unidiff-zero, since
> the new hunks can start or stop with a "-" or "+" line.

You do not have to do "unidiff zero".  Suppose you have this hunk you
need to split.

diff --git a/read-cache.c b/read-cache.c
index 7db5588..4d12073 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -12,8 +12,8 @@
 /* Index extensions.
  *
  * The first letter should be 'A'..'Z' for extensions that are not
- * necessary for a correct operation (i.e. optimization data).
- * When new extensions are added that _needs_ to be understood in
+ * necessary for a correct operation (that is, optimization data).
+ * When new extensions are added that needs to be understood in
  * order to correctly interpret the index file, pick character that
  * is outside the range, to cause the reader to abort.
  */

Think about taking the s/i.e./that is,/ substitution without taking the
other s/_needs_/needs/ substitution.  You do not split the hunk between
two '-' lines, but effectively make it into this hunk instead:

diff --git a/read-cache.c b/read-cache.c
index 7db5588..4d12073 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -12,8 +12,8 @@
 /* Index extensions.
  *
  * The first letter should be 'A'..'Z' for extensions that are not
- * necessary for a correct operation (i.e. optimization data).
+ * necessary for a correct operation (that is, optimization data).
  * When new extensions are added that _needs_ to be understood in
  * order to correctly interpret the index file, pick character that
  * is outside the range, to cause the reader to abort.
  */

That is, , if you want to do finer grained hunk splitting than what "git
add -p" lets you do, you do _not_ let user specify "I want to split the
hunk into two, before this point and after this point".  Instead, let
the user pick zero or more '-' line and zero or more '+' line, and
adjust the context around it.  An unpicked '-' line becomes the common
context, and an unpicked '+' line disappears.  After that, you recount
the diff.  That way, you do not have to do any "unidiff zero" cop-out.

At the same time, you can stash away what was _not_ picked, creating two
variants to be applied on top of the result of applying (or not
applying) the picked patch, if you want to allow "undo".

(variant one: applies after the above is applied)
@@ -12,8 +12,8 @@
 /* Index extensions.
  *
  * The first letter should be 'A'..'Z' for extensions that are not
  * necessary for a correct operation (that is, optimization data).
- * When new extensions are added that _needs_ to be understood in
+ * When new extensions are added that needs to be understood in
  * order to correctly interpret the index file, pick character that
  * is outside the range, to cause the reader to abort.
  */

(variant two: applies if the above is not applied)
@@ -12,8 +12,8 @@
 /* Index extensions.
  *
  * The first letter should be 'A'..'Z' for extensions that are not
  * necessary for a correct operation (i.e. optimization data).
- * When new extensions are added that _needs_ to be understood in
+ * When new extensions are added that needs to be understood in
  * order to correctly interpret the index file, pick character that
  * is outside the range, to cause the reader to abort.
  */