[PATCH 0/1] Add new --index-filter example to git-filter-branch.txt

[PATCH 0/1] Add new --index-filter example to git-filter-branch.txt

[Michael O'Cleirigh]

At work I patched git with the prevously sent --egrep-filter patch that allows
an irregular history to be retained, like where code reorganization occured and
you want to keep the history from the original locations.

The comments I recieved at that time was that my case was an --index-filter
variant.

I've reworked it into an --index-filter using the $GIT_COMMIT variable, tested
that it works, and have added it as an example to the git-filter-branch 
documentation.



Michael O'Cleirigh (1):
  Documentation/git-filter-branch: add --index-filter example

 Documentation/git-filter-branch.txt |   13 +++++++++++++
 1 files changed, 13 insertions(+), 0 deletions(-)

-- 
1.7.2.3

[PATCH 1/1] Documentation/git-filter-branch: add --index-filter example

[Michael O'Cleirigh]

Added an example using git ls-tree on the $GIT_COMMIT and then filtering
what should remain in the history using egrep.

This allows multiple paths accross history to be retained when filtering a branch.
---
 Documentation/git-filter-branch.txt |   13 +++++++++++++
 1 files changed, 13 insertions(+), 0 deletions(-)

diff --git a/Documentation/git-filter-branch.txt b/Documentation/git-filter-branch.txt
index 9dc1f2a..e80bc01 100644
--- a/Documentation/git-filter-branch.txt
+++ b/Documentation/git-filter-branch.txt
@@ -248,6 +248,19 @@ To set a commit (which typically is at the tip of another
 history) to be the parent of the current initial commit, in
 order to paste the other history behind the current history:
 
+-----------------------------------------------------------------------------------------------------------------------------------
+git filter-branch --prune-empty --index-filter 'git ls-tree $GIT_COMMIT | egrep \
+"(top-dir-A|top-dir-B|consolidated-dir-C)" | git mktree | xargs git read-tree -i -m'
+-----------------------------------------------------------------------------------------------------------------------------------
+
+The --subdirectory-filter works great when all commits of interest have
+that path but sometimes you have a history that contains an irregular 
+structure. i.e. that was reorganized at some point and you would like to 
+keep the prior history.
+
+For each commit the top-dir-A, top-dir-B and consolidated-dir-C paths (if they exist)
+are kept (and everything else discarded). 
+
 -------------------------------------------------------------------
 git filter-branch --parent-filter 'sed "s/^\$/-p <graft-id>/"' HEAD
 -------------------------------------------------------------------
-- 
1.7.2.3

Re: [PATCH 1/1] Documentation/git-filter-branch: add --index-filter example

[Johannes Sixt]

I like the example, particularly because it shows that it is not
necessary to start off with index's content in an --index-filter, and
that the index can be replaced with read-tree -i, i.e., GIT_INDEX_FILE
juggling is not necessary for some tasks.

Am 27.04.2011 20:08, schrieb Michael O'Cleirigh:
> Added an example using git ls-tree on the $GIT_COMMIT and then filtering
> what should remain in the history using egrep.
> 
> This allows multiple paths accross history to be retained when filtering a branch.

Sign-off is missing. Please keep lines well below 75 characters. Write
'Add an example...' rather than 'Added an example...'.

> ---
>  Documentation/git-filter-branch.txt |   13 +++++++++++++
>  1 files changed, 13 insertions(+), 0 deletions(-)
> 
> diff --git a/Documentation/git-filter-branch.txt b/Documentation/git-filter-branch.txt
> index 9dc1f2a..e80bc01 100644
> --- a/Documentation/git-filter-branch.txt
> +++ b/Documentation/git-filter-branch.txt
> @@ -248,6 +248,19 @@ To set a commit (which typically is at the tip of another
>  history) to be the parent of the current initial commit, in
>  order to paste the other history behind the current history:
>  
> +-----------------------------------------------------------------------------------------------------------------------------------
> +git filter-branch --prune-empty --index-filter 'git ls-tree $GIT_COMMIT | egrep \
> +"(top-dir-A|top-dir-B|consolidated-dir-C)" | git mktree | xargs git read-tree -i -m'
> +-----------------------------------------------------------------------------------------------------------------------------------
> +
> +The --subdirectory-filter works great when all commits of interest have
> +that path but sometimes you have a history that contains an irregular 
> +structure. i.e. that was reorganized at some point and you would like to 
> +keep the prior history.
> +
> +For each commit the top-dir-A, top-dir-B and consolidated-dir-C paths (if they exist)
> +are kept (and everything else discarded). 
> +
>  -------------------------------------------------------------------
>  git filter-branch --parent-filter 'sed "s/^\$/-p <graft-id>/"' HEAD
>  -------------------------------------------------------------------

This places the new text in the middle of another example, doesn't it?

The prefered style of examples to give motivation first, then the
example command, then a description of command details. I would drop the
sentence "i.e. ... was reoranized... history" because it does not add
value for the reader (I can think of other equally or even more likely
reason why the interesting path structure is irregular).

The example looks better when it is broken into more lines, e.g. like this:

-----------------------------------------------------------
git filter-branch --prune-empty --index-filter '
		git ls-tree $GIT_COMMIT |
		egrep "(top-dir-A|top-dir-B|other-dir-C)" |
 		git mktree | xargs git read-tree -i -m'
-----------------------------------------------------------

And the dash lines are just as long as necessary.

-- Hannes

Re: [PATCH 0/1] Add new --index-filter example to git-filter-branch.txt

[Sverre Rabbelier]

Heya,

On Wed, Apr 27, 2011 at 20:08, Michael O'Cleirigh
<michael.ocleirigh@rivulet.ca> wrote:
> At work I patched git with the prevously sent --egrep-filter patch that allows
> an irregular history to be retained, like where code reorganization occured and
> you want to keep the history from the original locations.
>
> The comments I recieved at that time was that my case was an --index-filter
> variant.
>
> I've reworked it into an --index-filter using the $GIT_COMMIT variable, tested
> that it works, and have added it as an example to the git-filter-branch
> documentation.

Meta nit: for a 1 patch series we would usually put this kind of
introduction after the --- in the actual patch.

-- 
Cheers,

Sverre Rabbelier