git rebase + fuzz = possible bad merge

git rebase + fuzz = possible bad merge

[Benny Halevy]

I'm hitting bad merges with (non interactive) git rebase
when a hunk is merged pre-maturely into an inexact match
when there's fuzz.

The following shell script reproduce this with
git version 1.6.2.1:

#!/bin/sh

rm -rf test_file .git
git init

cat >test_file <<EOF
struct one {
	line 1;
	line 2;
	line 3;
	line 4;
	line 5;
	line 6;
	line 7;
};

struct two {
	line 1;
	line 2;
	line 3;
	line 4;
	line 5;
	line 6;
	line 7;
};
EOF

git add test_file
git commit -m test_file

git am <<EOF
From: Benny Halevy <bhalevy@panasas.com>
Subject: change struct two

diff --git a/test_file b/test_file
--- a/test_file
+++ b/test_file
@@ -12,7 +12,7 @@ struct two {
 	line 1;
 	line 2;
 	line 3;
-	line 4;
+	LINE 4;
 	line 5;
 	line 6;
 	line 7;
EOF

git checkout -b test_branch HEAD^
{ for i in {1..10}; do echo fuzz $i; done; echo; cat test_file; } > fuzz_file
mv fuzz_file test_file

git commit -a -m fuzz

git rebase --onto test_branch master^ master

if [ $(awk '/LINE/ {print NR}' test_file) != 26 ]; then
	echo 1>&2 $0: test failed
	exit 1
else
	echo 1>&2 $0: test succeeded
fi

Re: git rebase + fuzz = possible bad merge

[Thomas Rast]

[-- Attachment #1: Type: text/plain, Size: 723 bytes --]

Benny Halevy wrote:
> I'm hitting bad merges with (non interactive) git rebase
> when a hunk is merged pre-maturely into an inexact match
> when there's fuzz.
[...]
> { for i in {1..10}; do echo fuzz $i; done; echo; cat test_file; } > fuzz_file
[...]
> git rebase --onto test_branch master^ master

git-am, and by extension rebase, by default doesn't take history into
account.  It just applies the patches "blindly".  Thus, there's no way
to know which series of 'line N' you really wanted it to go onto.

To avoid this issue, use the -m option to git-rebase so that it uses a
"real" merge.  (You can achieve similar effects for git-am with the -3
option.)

-- 
Thomas Rast
trast@{inf,student}.ethz.ch

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: git rebase + fuzz = possible bad merge

[Benny Halevy]

On Mar. 23, 2009, 15:54 +0200, Thomas Rast <trast@student.ethz.ch> wrote:
> Benny Halevy wrote:
>> I'm hitting bad merges with (non interactive) git rebase
>> when a hunk is merged pre-maturely into an inexact match
>> when there's fuzz.
> [...]
>> { for i in {1..10}; do echo fuzz $i; done; echo; cat test_file; } > fuzz_file
> [...]
>> git rebase --onto test_branch master^ master
> 
> git-am, and by extension rebase, by default doesn't take history into
> account.  It just applies the patches "blindly".  Thus, there's no way
> to know which series of 'line N' you really wanted it to go onto.
> 
> To avoid this issue, use the -m option to git-rebase so that it uses a
> "real" merge.  (You can achieve similar effects for git-am with the -3
> option.)
> 

OK. -m indeed helps and I'm certainly going to adopt it for my rebase scripts.
git rebase -i does too, BTW.

I would expect though that the default mode for automatic rebase would be
the strictest and safest...

Benny

Re: git rebase + fuzz = possible bad merge

[Benny Halevy]

On Mar. 23, 2009, 16:06 +0200, Benny Halevy <bhalevy@panasas.com> wrote:
> On Mar. 23, 2009, 15:54 +0200, Thomas Rast <trast@student.ethz.ch> wrote:
>> Benny Halevy wrote:
>>> I'm hitting bad merges with (non interactive) git rebase
>>> when a hunk is merged pre-maturely into an inexact match
>>> when there's fuzz.
>> [...]
>>> { for i in {1..10}; do echo fuzz $i; done; echo; cat test_file; } > fuzz_file
>> [...]
>>> git rebase --onto test_branch master^ master
>> git-am, and by extension rebase, by default doesn't take history into
>> account.  It just applies the patches "blindly".  ...

git am, in contrast to git rebase, errs on fuzz and you'll need to
apply the patch manually.  It might be annoying, but it'd be safer
if git rebase would either stop on fuzz too or revert to using
merge strategies (same as using git rebase -m) by default.

>> ...  Thus, there's no way
>> to know which series of 'line N' you really wanted it to go onto.

Well, there's the hunk header.

Benny

>>
>> To avoid this issue, use the -m option to git-rebase so that it uses a
>> "real" merge.  (You can achieve similar effects for git-am with the -3
>> option.)
>>
> 
> OK. -m indeed helps and I'm certainly going to adopt it for my rebase scripts.
> git rebase -i does too, BTW.
> 
> I would expect though that the default mode for automatic rebase would be
> the strictest and safest...
> 
> Benny
> --
> To unsubscribe from this list: send the line "unsubscribe git" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html