[Gnu-arch-users] --forward mostly harmless

[David Allouche]

I was writing some test cases for PyArch, when I found myself having
trouble identifying the behaviour enabled by the --forward option to
apply-changeset.

The intended behaviour of the --forward option, according to Tom:

        As with `patch(1)', --forward means "ignore hunks that appear to
        already have been applied".

For some time, I had a gnawing doubt about what the --forward option to
patch really meant. So I looked at the documentation really closely.

        `-N'
        `--forward'
             Ignore patches that `patch' thinks are reversed or already applied.
             See also `-R'.  *Note Reversed Patches::.

Note that the documentation says "ignore patches", not "ignore hunks". 

The "Reversed Patches" page says:

        Often `patch' can guess that the patch is reversed.  If the first
        hunk of a patch fails, `patch' reverses the hunk to see if it can apply
        it that way.  If it can, `patch' asks you if you want to have the `-R'
        option set; if it can't, `patch' continues to apply the patch normally.

In other words, `patch' thinks that a patch is reversed if the _first_
hunk of the patch can only be applied in reverse, and then --forward
tells it to ignore the _whole_ patch.

This behaviour is demonstrated in the attached test case.

I believe we can agree that is not a useful behaviour for tla.

However the test case fails to expose any difference in the behaviour of
tla when --forward is used with a simple changeset which only changes
the contents of a file.

The reason for that surprising result is that tla invokes patch with the
options "[--forward] -f -s --posix" and -f effectively disables 
--forward.

The fact this problem was not found so far strongly suggests that nobody
is actually using --forward.

Since he intended functionality of the --forward option to tla is
currently not implemented by patch, and since the current implementation
of --forward in tla is no-op as far as I can tell, I suggest the
--forward option be removed.

-- 
                                                            -- ddaa

apply-forward.sh
Description: Test cases for patch and tla

method set: names
method set: names
method set: names
~ ~
* produced diff
--- treeA/file.txt      2004-09-13 02:00:09.070591396 +0200
+++ treeC/file.txt      2004-09-13 02:00:09.077590164 +0200
@@ -1,5 +1,5 @@
 1 ---
-2 hello world
+2 Hello, World!
 3 ---
 4 --
 5 -
@@ -7,5 +7,5 @@
 7 -
 8 --
 9 ---
-A yadda yadda
+A Yadda, Yadda!
 B ---
patching file file.txt
Reversed (or previously applied) patch detected!  Skipping patch.
2 out of 2 hunks ignored -- saving rejects to file file.txt.rej
* patch exits with status 0
* produced reject
***************
*** 1,5 ****
  1 ---
- 2 hello world
  3 ---
  4 --
  5 -
--- 1,5 ----
  1 ---
+ 2 Hello, World!
  3 ---
  4 --
  5 -
***************
*** 7,11 ****
  7 -
  8 --
  9 ---
- A yadda yadda
  B ---
--- 7,11 ----
  7 -
  8 --
  9 ---
+ A Yadda, Yadda!
  B ---
~
M  file.txt
** Redundant patch
C   file.txt
* apply-changeset failed
* produced reject
***************
--- 1,5 ****
  1 ---
- 2 hello world
  3 ---
  4 --
  5 -
--- 1,5 ----
  1 ---
+ 2 Hello, World!
  3 ---
  4 --
  5 -
***************
--- 7,11 ****
  7 -
  8 --
  9 ---
- A yadda yadda
  B ---
--- 7,11 ----
  7 -
  8 --
  9 ---
+ A Yadda, Yadda!
  B ---
* no tree changes (normal)
** Only first hunk reversable
C   file.txt
* apply-changeset failed
* produced reject
***************
--- 1,5 ****
  1 ---
- 2 hello world
  3 ---
  4 --
  5 -
--- 1,5 ----
  1 ---
+ 2 Hello, World!
  3 ---
  4 --
  5 -
M  file.txt