‹ BackHN Continuity

Thread

Git 3.0's upcoming SHA-256 default will be a costly mistake

570 points · 536 comments · chmaynard

  1. nicoburns · · focus · HN ↗
    From what I'd read, SHA256 in git is showing every sign of being another IPv6. In particular:

    - It's implemented in a non-backwards-compatible way

    - The benefits over the older model are a bit nebulous

    - There's a large amount of tooling that needs to catch up, and little sign that there is movement there

    1. sltkr · · focus · HN ↗
      The difference with IPv6 adoption is that the internet relies heavily on network effects: so long as some hosts only have an IPv4 address, you need an IPv4 address for full connectivity, but then if everyone has an IPv4 address anyway, there is no immediate need to migrate to IPv6.

      (Yes us Hacker News users have plenty of use cases for IPv6, like self-hosting and peer-to-peer networking and so on; we are not the average user.)

      This effect doesn't exist for the Git migration. Each repo can be updated independently; it doesn't affect users of other repositories, and most likely, the majority of devs will work on some SHA-1 repos and some SHA-256 repos with no issue.

      If anything, I would compare it with the Python 2 to Python 3 migration, which was also painful, but succeeded eventually (despite being much less necessary in the first place).

      1. metalliqaz · · focus · HN ↗
        changing repos to the new IDs would break any existing links to content on the pre-migration repos.
        1. AndrewDucker · · focus · HN ↗
          Unless you link using tags.
          1. crote · · focus · HN ↗
            Which is considered a Really Bad Idea because tags aren't immutable, so there's absolutely zero guarantee that it'll point to the same commit a few months from now.

            The GitHub Actions ecosystem found out the hard way, through some rather high-profile compromises. They hotfixed it by adding "immutable tags" to their platform, and are now working on adding a lockfile to... easily reference a commit hash.

            1. PunchyHamster · · focus · HN ↗
              well if you use a knife to stab your fingers that's not a knife's fault

              repo can also rewrite existing commit and you again won't be able to retrieve it so switching to commit IDs only lowers the level of failure somewhat

              1. computerfriend · · focus · HN ↗
                Having a fetch break is better than having a fetch pull down malware.
                1. PunchyHamster · · focus · HN ↗
                  which I already said

                  > lowers the level of failure somewhat

                  congratulations on lack of ability to read with understanding

                  1. computerfriend · · focus · HN ↗
                    > congratulations on lack of ability to read with understanding

                    This seems weirdly aggressive and not nice.

      2. charlieyu1 · · focus · HN ↗
        I just started learning IPv6 with AWS since they charge $0.005/hr per IPv4. Maybe it will be more expensive in the future and eventually it will be the new default.
      3. kllrnohj · · focus · HN ↗
        > Yes us Hacker News users have plenty of use cases for IPv6, like self-hosting

        Funnily enough, self hosting is why I can't use IPv6. I want vlan isolation, but only get a /64 from my ISP.

        Fortunately the lack of IPv6 also isn't a meaningful loss anyway so whatever

        1. Dylan16807 · · focus · HN ↗
          What kind of isolation are you looking for? Vlans might still be possible.
        2. someonebaggy · · focus · HN ↗
          FWIW there's nothing about the addresses themselves that stops you subdividing a /64, but it depends on your router.

          You could put your server at x::1 and static-route that address as a /128 on your router, if it supports it. The reverse route might be a bit tricky but putting ::0 on the router and telling the server it's a /127 should work. Anything outside of the /127 (so, all the randomly generated addresses on your home network) would go back through the router.

          Now if that /64 is also changing every day, then it's a problem and idk what you'd do.

      4. Black616Angel · · focus · HN ↗
        You are wrong with your own examples.

        Github is THE main platform for git. If github doesn't upgrade (and their code has been shit and hard to fix/update before) then the shift will not happen. Because yes, you can upgrade your repo independently, but if there is nowhere to push, no one will do it.

        IPv6 is (also because of github) a great example for this. You can easily have an IPv6 address next to your IPv4 address, but a lot of websites (e.g. github) don't have that. Why would a normal company use IPv6 if even the bastion of nerds doesn't use it?

        And to Python 2's "eventual migration" I can unhappily tell you, that my company (recently) bought an actively developed tool, that still uses Python 2.

        1. globular-toast · · focus · HN ↗
          I don't think GitHub is quite as important as that, outside of some specific projects that made bad/lazy decisions (thinking Golang here). If they didn't support 256 I think a lot of enterprises and open source stuff would simply jump ship to GitLab or elsewhere. In the enterprise world it only takes one person to write a security document banning sha1 for this to happen. For that reason, GitHub will support sha256.
          1. radicalcentrist · · focus · HN ↗
            I think you would be very unpleasantly surprised to see how many organizations deeply rely on GitHub. Think about how many open-source projects you've seen that have willingly tied themselves to GH-specific features, and then extrapolate to thousands of companies with masses of spaghetti code that are hopelessly reliant on their internal GH Enterprise features.
        2. someonebaggy · · focus · HN ↗
          GitHub is one platform that can make a centralised decision. What it does, goes. If only IPv6 had such an entity, we'd have it by now.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.