reduce the size of commit info file

[onur-ozkan]

As we already have the complete version of the commit hash, we don't need to keep its short version in the commit file and memory (during bootstrap).

[rustbot]

r? @Mark-Simulacrum

rustbot has assigned @Mark-Simulacrum.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[Kobzol]

Note that this is not actually 100% equivalent to rev-parse --short, because that will return a unique prefix with at least 9 characters (https://git-scm.com/docs/git-rev-parse#Documentation/git-rev-parse.txt---shortltlengthgt). Not sure if we ever hit the case where a 9 character SHA prefix wouldn't be unique though.

[cuviper]

I happen to have an old blog post exploring git abbreviations:
https://blog.cuviper.com/2013/11/10/how-short-can-git-abbreviate/
(Maybe I should rewrite that in Rust!)

That script on the Rust repo gives me:

2581360 objects
 4: 2581360 / 65536
 5: 2360776 / 738816
 6: 368483 / 179538
 7: 24480 / 12224
 8: 1488 / 744
 9: 102 / 51
10: 6 / 3
11: 0 / 0
130c811393499854d9ed66edb4060797ea3a8b81 blob 11235
130c8113935c7402fc5fc1eab0d778c54c1aa302 tree 142
31c7d4e8041d065214c8098090b9c7d34d2c5268 tree 236
31c7d4e804ee40192f72a79237ddd6636c251575 blob 7811
afb64da8b61738966ee1171e5b2f96c8e004c054 blob 140593
afb64da8b6f34d1df1ecc1eb4ae289e115cb9dfe blob 8706

If you drop the initial --objects in the script, you'll get only commits:

259050 objects
 4: 254083 / 59335
 5: 57089 / 27377
 6: 3847 / 1919
 7: 210 / 105
 8: 8 / 4
 9: 0 / 0
60cc6b931996b65a053699973bb08487e31bf9f0 commit 1084
60cc6b93f6490eab960e90664215c989515d2f06 commit 1504
dcbb27aa60f5105a49d5e416490d6549385c325d commit 310
dcbb27aab60eac01a666a2d12a64683a8f237e8b commit 707
eb4860c77c62450f721cdefab22a2bde3ae62ec0 commit 994
eb4860c7e1e4109d94e84adc2794f720120604e3 commit 1647
faa280ceb9cd440f462349cd5bbaaff83c269213 commit 250
faa280cee69a34462e43b1a5df7c06faf97e56be commit 287

So 9 characters are currently enough to uniquely identify commits, at least.

[onur-ozkan]

Note that this is not actually 100% equivalent to rev-parse --short, because that will return a unique prefix with at least 9 characters (https://git-scm.com/docs/git-rev-parse#Documentation/git-rev-parse.txt---shortltlengthgt). Not sure if we ever hit the case where a 9 character SHA prefix wouldn't be unique though.

I did not know that, thanks!

So 9 characters are currently enough to uniquely identify commits, at least.

Taking the 10 characters should be fairly enough seems like.

[Mark-Simulacrum]

I'm not terribly happy with just truncating to 10 characters, but I guess it's true that we likely have considerable headroom, at least without a malicious attacker.

What do we use the short format for today? Just displaying a change log? Or are there other places it shows up, such as rustc -V output or similar?

[onur-ozkan]

What do we use the short format for today? Just displaying a change log?

It's widely used for many purposes (e.g., it's passed as CFG_SHORT_COMMIT_HASH env to cargo).