just curious: what influences a commit hash?

Hi,

being new to git I did some experiments with commits looking at the hashes. What I observed:
* The same commit (same file, same committer, same message) into different empty repositories (git init) gives different hashes. So I assume that also the time of the commit influences the hash. Is this intended? For what reason?
* Having created two repositories exactly the same way (the history is the same except for the commit times and hashes) I applied the same patch (using git am) and again I got different hashes for these commits. So in some way also the repository/branch influences the hash of a commit!?

From reading the Git user's manual, chapter 10, object storage format, I was not expecting this. Can someone explain or give a link to a more detailed description?

thank you,

Wolfgang

On Thu, 2009-03-05 at 07:36 +0100, stoecher@gmx.at wrote:

This should be expected if the initial hashes in the history are different. The hash of a commit is based also on the hashes of all parent commits - in this way git 'protects' the repository history by guaranteeing that if two objects have the same hash, they will come from the same history. So the second issue is a consequence of the first, though I am not certain why the first occurs (if the file contents and size are the same, I would expect the hash for the blob/tree to be the same - maybe due to git's special handling of initial commits?)

Hi,

On Thu, Mar 05, 2009 at 02:25:39AM -0500, Matt Enright wrote:

Yes, commit time and commit date influence the hash.

But the hashes for the corresponding trees should be the same. Check the output of git rev-parse $commit^{tree}.

If you want to reproduce the exact same commit, you need to set the env variables GIT_AUTHOR_DATE and GIT_COMMITTER_DATE. (Not sure, but GIT_AUTHOR_DATE might be handled by git am.)

Best regards Uwe

stoecher@gmx.at writes:

You should distinguish "commit objects" from "tree objects". You can see the "commit" object with, for example:

$ git cat-file -p HEAD tree 4b825dc642cb6eb9a060e54bf8d69288fbee4904 author Matthieu Moy <Matthieu.Moy@imag.fr> 1236249249 +0100 committer Matthieu Moy <Matthieu.Moy@imag.fr> 1236249249 +0100

foo

See: it contains a reference to the tree, possibly references to the
parents, and a timestamp. So, hashing it takes the timestamp into
account. A consequence of this is that you can not change the
timestamp for a commit without changing the "revision identifier"
(i.e. the hash of the commit object).

But the empty tree object is deterministically 4b825dc642cb6eb9a060e54bf8d69288fbee4904.