Interesting. This script modifies the commit timestamp until the commit object hashes to a SHA-1 that starts with a prefix specified on the command-line. It seems like it was used on the repo itself with prefixes of 0001, 0002, 0003.
$ ./beautify_git_hash.py 0001
I wonder if this somehow increases the probability of a hash collision, if only ever so slightly? In any case, it's a neat hack!
The script prints a suggested "git commit --amend" command so don't run the suggestion if you've already pushed the head of your branch.
There is another downside to keep in mind. git commands accept the shortest unambiguous SHA-1 prefix instead of requiring you to specify the entire thing. Usually the first 7 characters or so will do, even on fairly large projects.
Using this script too often may force you to specify more characters to get past the prefix being identical on too many commits.
Exactly! That's exactly what I was thinking reading this... it's a bad thing, not a feature. Why on earth would anyone do this to their repo willingly? This is visual aesthetics run amok, in the hands of someone who doesn't understand what the point of that hash value was. Ouch.
SHA-1 has a digest size of 160 bits. The first two bytes of these beautified hashes are predetermined, which lowers the size to 144 bits. For a more than 50% chance of a collision, more than 4 sextillion hashes[1] would have to be generated, compared to 1 septillion hashes[2] for the normal digest. A difference of an order of a magnitude, yet absolutely nothing to be worried about. Collisions are vastly more likely to be caused by a stray cosmic ray than actual chance.
This is only true if you use the same prefix for all commits. However, the script takes the prefix as argument, which allows you to use prefixed like 0001, 0002, etc.
Apart from that, this is a toy project and by no means meant seriously.
While it does increase the chances of a hash collision, note that no-one has ever been able to find a SHA-1 collision, out of all the projects that use it everywhere.
If you find a SHA-1 collision, by any method, you will become very, very famous. It will probably be worth having a broken git repository.
The script is able to set 5 digits, not 4. The prefixes in the examples are actually 0001a, 0002a, etc. I added the trailing "a" to make the numbers more readable. Otherwise you would get hashes like 00015... which would look awkward. So I'm using the "a" as a kind of separator.
Also, the script refuses to change timestamps by more than 30 minutes. So changing only the committer timestamp would allow you to set only 2-3 hash digits. That's why I'm also changing the author timestamp, which leads to enough possible combinations to set about 5 hash digits. (Unless you're unlucky, because there's always a rest possibility for failure in this kind of algorithms.)
BTW, this program has been inspired by BitCoin's proof of work concept.
While the beauty may be in the eye of the beholder, for X11 based systems this is actually a step back.
I used to be able to highlight and paste the hashes, now the hash string is "hidden" inside a button where text link would do. Sure, as an ugly, kludgy workaround there is the flash-based "copy to clipboard" button, but that only works with ctrl-v pasting, not the middle mouse button. Good luck pasting that into an xterm. So it's either copy-type or click on the link and find the sha in text form on the resulting page.
I found myself using git log a lot more frequently during the past few days.
like everyone else said, the link's not about the UI. But on that subject, autocutsel (http://www.nongnu.org/autocutsel/) can keep all of your X11 clipboards synchronized, so that anything copied to the clipboard can be MMB pasted, and anything selected with the LMB can be Ctrl-V pasted.
Righty-o, it's purely for fun. @vog: merge my pull request, man. There's a post-commit hook in there. I think we finally solved the SHA-1 vs r#### argument ;-)
A is commit number one. D is commit number three. If B is two, then what is C? Assuming you can come up with a consistent view when all the information about the branch topology is available, what about when people have disconnected branches? How do you avoid duplicate generation numbers? If you can't avoid duplicate generation numbers, what's the point of having them? "Fixed in commit 2." Great, which commit 2 did you mean?
You have duplicates, B and C would have the same generation number. It just represents length of the longest path from the root commit. This is already calculated internally to optimize several operations, so they were considering making it part of the commit data.
It often won't uniquely identify commits, but revision numbers in SVN are also sometimes used imprecisely ("oh, it was around R1500"), so I think it would have value.
Yeah, but you know people want this so they can write things in their bug tracker like "fixed in r1232" instead of "fixed in 2c4d96d1dc2d12afd486561329a9a7b74d4ae110".
Actually if you follow the thread(s) relating to this, a less invasive change to speed up "git tag --contains" was adopted (see git.git@ffc4b8012), so "soon" probably isn't the right word.
26 comments
[ 0.24 ms ] story [ 69.6 ms ] threadThe script prints a suggested "git commit --amend" command so don't run the suggestion if you've already pushed the head of your branch.
Using this script too often may force you to specify more characters to get past the prefix being identical on too many commits.
Apart from that, of course you should use sensible prefixes like "0001", "0002", etc. and not use the same prefix for all commits.
[1] 2^(144/2) ~= 7e21
[2] 2^(160/2) ~= 1e24
Apart from that, this is a toy project and by no means meant seriously.
If you find a SHA-1 collision, by any method, you will become very, very famous. It will probably be worth having a broken git repository.
Some small corrections:
The script is able to set 5 digits, not 4. The prefixes in the examples are actually 0001a, 0002a, etc. I added the trailing "a" to make the numbers more readable. Otherwise you would get hashes like 00015... which would look awkward. So I'm using the "a" as a kind of separator.
Also, the script refuses to change timestamps by more than 30 minutes. So changing only the committer timestamp would allow you to set only 2-3 hash digits. That's why I'm also changing the author timestamp, which leads to enough possible combinations to set about 5 hash digits. (Unless you're unlucky, because there's always a rest possibility for failure in this kind of algorithms.)
BTW, this program has been inspired by BitCoin's proof of work concept.
I used to be able to highlight and paste the hashes, now the hash string is "hidden" inside a button where text link would do. Sure, as an ugly, kludgy workaround there is the flash-based "copy to clipboard" button, but that only works with ctrl-v pasting, not the middle mouse button. Good luck pasting that into an xterm. So it's either copy-type or click on the link and find the sha in text form on the resulting page.
I found myself using git log a lot more frequently during the past few days.
https://github.com/davvid/beautify_git_hash/commits/master
https://github.com/vog/beautify_git_hash/pull/1
Git already has a mechanism for naming commits, anyway. They are called "tags".
edit: jacknagel points out that they actually didn't go with this after all. Insted: https://github.com/git/git/commit/ffc4b8012d9a4f92ef238ff72c...
It often won't uniquely identify commits, but revision numbers in SVN are also sometimes used imprecisely ("oh, it was around R1500"), so I think it would have value.
https://github.com/stephenh/git-central/blob/master/server/p...
(It makes a tag-per-commit; I'm not saying that's necessarily a good idea, but if you really want revision numbers...)