We have a library with several versions which can be used in parallel.
Each version is in its own branch because a lot of code is shared but some things need to be different.
Developers should be working on different branches in parallel and merge forth and back as needed.
I don't know how to do this because with merge some changes cannot be preserved when merging back.
Here's a simple example, expressed in shell script form:
#!/bin/bash
rm -rf test
mkdir test
cd test
git init .
echo keep a diff > test.cpp
echo merge >> test.cpp
git add .
git commit -a -m a1
git branch -m a
git branch b
sed -i -e 's/merge/merge a1/g' test.cpp
git commit -a -m a2
git checkout b
sed -i -e 's/keep a diff/keep b diff/g' test.cpp
git commit -a -m b1
git merge --no-ff a
# manual resolve
echo keep b diff > test.cpp
echo merge a1 >> test.cpp
git commit -a -m "m1a->b"
git checkout a
echo a3 >> test.cpp
git commit -a -m a3
git merge --no-ff -m "m1b->a" b # hoping to get a conflict on the first line of test.cpp
print_file() {
git checkout $1
echo $1:
cat test.cpp
echo $1 end
echo
}
echo
print_file a
print_file b
It can be run online.
Its graph looks like this:
a1---a2--a3-mb
\ \ /
b-b1-mf--
I'm basically trying to preserve the first line with its differences per branch and merge all other lines. The problem occurs when merging back where the first line is overwritten.
I was hoping to have the following state in the two branches:
branch a, file test.cpp:
keep a diff
merge a1
a3
branch b, file test.cpp:
keep b diff
merge a1
but instead, for branch a, i get
keep b diff
merge a1
a3
I tried all different merge strategies but it seems there's always a fast forward possible, e.g. no conflicts are present.
Is there a way do this? I can use a manual merge tool which ignores the common ancestor but this is just a simple example and when merging many commits at once i'd like to preserve them.
