Correctly detecting modified paths in merge commits?

2 messages, 2 authors, 2016-06-15 · open the first message on its own page

Correctly detecting modified paths in merge commits?

From: Dun Peal <hidden>
Date: 2016-06-15 22:50:15

Hi.

We wrote a post-receive hook that alerts users (via email) when
specific paths are modified by their peers. The implementation is
pretty simple: whenever a new commit is made, we ask git for the full
list of files modified by that commit:

  git diff --name-only <COMMIT HASH>^!

This works well for regular commits, but breaks for merge commits.

For example, suppose we have the following basic merge scenario:

  B
 / \
A   D
 \ /
  C

Root A was branched to B and C, then merged into commit D.

Problem is, the diff for D^! will include all the changes introduced by C.

One obvious solution is to simply ignore merge commits, by parsing
`git cat-file commit D` and discarding all commits with parent count >
1. But merge commits may actually contain legitimate modified files if
there were any conflict resolutions.

So what's the best solution for this problem, oh wise Git wizards?

Thanks, D.

Re: Correctly detecting modified paths in merge commits?

From: Jakub Narebski <hidden>
Date: 2016-06-15 22:50:15

Dun Peal [off-list ref] writes:
We wrote a post-receive hook that alerts users (via email) when
specific paths are modified by their peers. The implementation is
pretty simple: whenever a new commit is made, we ask git for the full
list of files modified by that commit:

  git diff --name-only <COMMIT HASH>^!

This works well for regular commits, but breaks for merge commits.
Note that <commit>^! is *range* specifier, and 'git diff' really takes
two *endpoints*.  
From git-diff(1) manpage.
    For a more complete list of ways to spell <commit>, see  "SPECIFYING
    REVISIONS" section in gitrevisions(1). However, "diff" is about 
    comparing two _endpoints_, not ranges, and the range notations  
    ("<commit>..<commit>"  and  "<commit>...<commit>")  do not mean a
    range as defined in the "SPECIFYING RANGES" section in gitrevisions(1).

<commit>^1 means include given commit but exclude all of its parents
(see gitrevisions(7)).

For a merge commit r1^! means r1 ^p1 ^p2 (where p1 and p2 are parents
of r1), which for git-diff probably means "git diff p1 r1".
  
 
For example, suppose we have the following basic merge scenario:

  B
 / \
A   D
 \ /
  C

Root A was branched to B and C, then merged into commit D.

Problem is, the diff for D^! will include all the changes introduced by C.
See above.

Try 

  $ git diff-tree --name-only -c <COMMIT HADH>

instead.  '-c' is to show merge commit as combined diff (noting changes
different from both parents).  I'm not sure if this is what you want.

There is alwats '--cc' or '-m' instead of '-c'.
 
-- 
Jakub Narebski
Poland
ShadeHawk on #git
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help