The pull request hack is fucking magic | Hacker News Reader

The pull request hack is fucking magic

shkspr.mobi

169 pointsedent55 comments

I can't remember the name of the projects but a "bad guy" befriended the guy who looked after a small bit of code that happened to be a down stream dependency of a crypto wallet. After helpfully contributing code and back and forth friendly chat, the maintainer gave the bad guy commit access.

Edit here it is: https://www.theserverside.com/feature/Recent-open-source-fla...

In this case there isn’t a lot you can do. If a bad actor is going to spend a lot of time gaining your trust and doing good before they do bad how do you vet that? You either have to maintain full control forever or you have accept this possibility. The OP does mention that this might be a bad idea if your project has any level of security.

> how do you vet that

One of the maxims of security is that a sufficiently determined and resourceful attacker will always win. The defender's job is to disincentivize the attacker.

However, I think for a sufficiently high-impact project no-one should have commit access, every commit should be reviewed by quorum. Even so, you still run into "Underhanded C"-style stuff (and disinterested reviewers), and you still need to vet the quorum.

Most project dependencies management process is flawed.

We shouldn't blindly upgrade modules/deps minor or patch versions to latest with a single command. SemVer encourages this, but we should be more careful.

The software supply chain for a lot of OSS is flawed as well. We need software signatures, and less central artifact registries that can be compromised easily.

I don't have a perfect solution for all these problems. Rewriting code instead of relying on external deps, vendoring, automated vulnerability scans, higher scrutiny for all external code, and tighter processes for code management.

I somewhat agree on principle, but at the same time, it is hard enough keeping ur software patched and up to date already, adding more friction to that process may cause more damage than it prevents.

You might prevent malicious new code at the cost of running old vulnerable code longer.

> The Pull Request Hack probably isn't suitable if you're running big projects. And it is almost certainly a stupid idea if you write code which is actively used by multiple downstream projects. And if your stuff has even the slightest chance of compromising security then you're better off sticking to trusted members.

The code (event-stream, a popular NPM package which was targeted because it's used by a crypto wallet) fails all 3 criteria.

Only one... It is really tiny and has nothing to do with security. (As in, not directly... Actually everything can impact security if it's used wrong so that point is moot).

It does have multiple downstream users. But I will bet 99% of projects where contributors submit good PRs, and maintainers have not enough time so they give commit access, have multiple downstream users.

Have never understood the desire to hoard commit rights, half the point of RCS is rollback, and as the article says, once intent is established through effort ("proof of work") there is little risk in assuming good faith until demonstrated otherwise.

Commit rights are an asymmetrical reward for someone showing interest in a piece of code - cheap to offer and incredibly valuable to receive. They also immediately reduce the long term burden and "ownership" of the code, and for the trusted committer, massively reduce friction and any sense of gatekeeping in contributing to the project. I've done this on at least 4 projects, 3 of them were eventually transferred to the new long term maintainer and 1 became an org of its own, with two maintainers having full admin

I don’t think “hoarding” is a useful framing.

Instead, the observation is that granting others access to projects has a time dimension: you need to not only find them trustworthy now, but also reasonably believe that they will remain trustworthy. That last part is much harder, and includes things that are partially outside of the other person’s control (like account takeovers, being targeted by foreign and domestic law enforcement, etc.).

I generally think projects should accept more maintainers; however, your point about asymmetry goes both ways.

In my experience commit rights also decrease the quality of the code. Nearly everyone sucks. I've had to rewrite FOSS to get rid of the ugly. Ideally I could have kept a better eye on the commits but time doesn't always allow.

I'd probably "hoard" rights. Which is to say, any FOSS project that i am invested in, i am going to fight hard to ensure someone does not have the ability to make my life more difficult. I'd love to give out commit rights, but if it's some random good quality PR there's a very large chance that i now have to maintain that code - as well as any future code they merge if given rights.

I hand out code rights freely to those who seem similarly invested. As they can then share the burden for each new issue popping up, etc. However writing the code (the PR) is usually not the hard part in a FOSS project. The fortitude to maintain it is. As the weeks, months and years go by - who will be around?

"Commit rights" here seem pretty off-topic to me, honestly. This is FOSS. Assuming a permissive license, you can fork it and publish your own freely. It's not an insult to me for you to do that.

But yea, if i'm looking for an out - i'll hand them out to anyone interested. If however i'm needing to maintain it for my own purpose, then my time and will are the resources i'm most concerned about. A single PR is not a the only variable in the equation there, imo.

Exactly 0 people are maintainers of projects they don't have commit access to. Even if they dip after 6 month it's far better to have the option to remove any of the 6 months worth of commits in the current codebase than to hypothesis about the state of a future codebase/environment 6 months ago. This becomes even easier with good commit messages. If somebody is a user and decides to write a pull request, they are already invested anything more is virtue signaling, no amount of which will guarantee that they will be there in years, which is an unreasonable stipulation for a project that could be obsolete yesterday.

When they dip 6 months time, you’re maintaining whatever strange crap they let into your project.

If they offer to maintain it, I’d sooner point my docs to suggest using their fork instead of my repo. That makes them the maintainer without me offering to support what they do

removing code that is not dead, but because it is bad quality is about the hardest thing to do. because if i had time to clean up the code, i could have done it when the pull request was made. so i prefer to gatekeep and not let in bad code to begin with. if it is an application i depend on, then i need to trust the the code to be good enough. mistakes are ok, i make them too. but giving someone commit rights and then finding out 6 months later that almost everything they committed was not good sounds like a bad idea.

i also believe that code should be reviewed by a second person, if possible. so if i have a one-man project and there is a second contributor, i would review their pull-requests and commit them. if they in turn are able and willing to review my commits, i might give them commit rights as long as we both can agree to review each others commits. in that case, the other person will only commit code i have written, and i will commit code they have written. (or however we set up the process. either way i will still have seen all the code going into the project.)

but i agree with the article. it really depends on the nature of the project. i do have some projects where i am just messing around, creating something useful, but not critical. having others directly commit to that would probably not be a big problem, even if the code they write is messy, as long as it works, it'll be ok.

alternatively, if i do have the time to review code, i can also do that after it has been committed. if it's bad, i can clean it up then. and if there is to much cleanup work happening i can revoke commit rights and i'll only have to remove the most recent commits that i don't have the time to clean up.

> If somebody is a user and decides to write a pull request, they are already invested anything more is virtue signaling, no amount of which will guarantee that they will be there in years, which is an unreasonable stipulation for a project that could be obsolete yesterday.

I'll believe that when they're replying to issues, labeling bug reports, etc. Hell, or even making PRs that maintain the code base! That would be an instant commit access. Writing new code is easy. Maintaining is tough, boring, etc.

Again, if i even get a gut feeling that this is going to be work - i'd rather just fork my own damn project and give them my old one. I FOSS projects to let others use my work if it's of value to them.

It's a thin line between having fun side projects and work. I think you misunderstand the type of projects i FOSS. If your arguments were common i'd be more inclined to just make projects private, as i'm that averse to wanting more work in life.

I really, really like to keep the fun in my projects.

replying to issues, labeling bug reports, etc

i do that on projects that i am interested in. because it sometimes is a good way to get into a project and show that i care about the project as a whole, not just getting my patch accepted. and why should i not care? i am using it myself.

to call that virtue signaling as the grand parent comment does, is frankly insulting.

however, caring about a project doesn't guarantee anything.

a while ago i found a nice project that solved a problem i had. it had some bugs that i fixed to make it work for me. i shared the issues and fixes with the project, expecting to do more work later. but then my situation changed and this project is no longer a priority. i hope to get back to the it, but i can't say when that will be. i still care about the project, i just don't have time for it now.

> there is little risk in assuming good faith until demonstrated otherwise.

The risk is that a malicious user fakes good commits, earns commit rights, and then pushes malware that is then installed by unsuspecting downstream users before the code is found and reverted.

This can be partially mitigated by making sure they don’t have permission to create a release, but it would still be possible to compromise developers working off the main branch.

Really depends on the dynamics of the project, and its life cycle. If you're in a well-established project with norms, conventions, long-time users, existing long-time developers, stable features and cadence, and have a lot of code to maintain -- I think being conservative with who you give commit rights to is fairly important. The project is going to have expectations around conventions, maintenance, what quality bar to meet, etc and breaking from those isn't always a good thing.

I do think being more liberal with commit bits helps a lot more when you have a project you want to grow, that is new, and you have a lot of code that needs to be written by a lot of people, and ideas to explore, and things that even need rewriting. You need the code written, people to use the software, so you know where to further drive improvements. You need people willing to dogfood and rewrite things and fix the thing they just broke.

I wouldn't give commit rights to anybody in that case but, like. If you have a solid history of +5 patches, you have some clear FOSS experience of your own, clearly have skill, and it's early on in the life of the project? I just need the help? Yeah, you could make worse decisions.

I have been on both very established projects of 20+ years where commit rights came months after writing patches, and newer projects where they came days after writing patches because the project was (is) still growing. Ultimately, I think it comes down to having an idea of where you want things to go, and picking out the people who seem to share that idea.

Game theory around this is fairly robust for me. If someone is willing to spend their time to put together anything with good intentions, then I feel like an initial offering of trust is an excellent response.

Sadly, as with all things nice, if the practice became widespread enough, it would cease to work for abundance of malicious actors.

Just about any code project in any real programming language targeting any not super locked-down platform has a chance of compromising security, so that’s not a great criterion. The project given as an example (collection of super small SVG icons) is an outlier, and even there it’s possible to smuggle a short embedded script into a <1KB SVG, e.g. you can read/manipulate DOM (if the <svg> is directly in DOM as opposed to in a separate document) and use fetch() without problem in an embedded script tag.

I’ve assigned strangers as maintainers on moderately popular projects (>1k GitHub stars) and the results have been satisfactory so far. Of course there’s a possibility that one day I’ll end up on HN front page having to explain why my project is shipping malware, but that chance seems really slim.

> Several years ago, I came across Felix Geisendörfer's Pull Request Hack. The premise is simple - if people are making decent Pull Requests to your project then you should give them commit access.

This is not a "hack", this is precisely how free and open source software projects have operated since before GitHub existed. It is sad how much of that culture has been lost and is reinventing the wheel.

The more radical version is that you automatically merge every PR that passes the tests and give the committer commit access. But it seems strange to see the default social practice of giving commit access to those who make decent PRs described as a "hack."

Um, how exactly do you mean that? The pre-GitHub equivalent would be giving people commit access to the CVS/Subversion repository. This generally did not happen lightly either in most projects.

> The pre-GitHub equivalent would be giving people commit access to the CVS/Subversion repository.

Correct.

> This generally did not happen lightly either in most projects.

Maybe not for the super big or central projects. But for most in that long tail, that's been my experience, at least in Debian. For a long time, there have been tons of orphaned projects or projects actively looking for new maintainers.

I think the “hack” language is correct because it’s a rethinking of GitHub at some level.

More importantly though I think you’re missing something very big about open source projects — the transition from “small project with single maintainer” to “group ownership” has always been a difficult one; that’s not unique to GitHub.

I can't agree from my own memory of the time. From how jealously most projects guarded commit access to CVS, you'd think the commit bit was a key to their personal homes and an invitation to sleep with their wives. Many (most?) "open source" projects in the early days had a private CVS that outsiders couldn't even read; the original model everyone followed was just source code dumps on FTP alongside binary releases. Maybe we're talking about different time periods.

> Maybe we're talking about different time periods.

Perhaps, or different ecosystems. Most of my experience is in Debian.

I attended a talk on Perl6 by https://en.wikipedia.org/wiki/Audrey_Tang once.

Very nicely she gave me a commit bit for Perl6. Sorry Audrey, I never used it!

It's a bit surprising to me that this is seen as a "hack". My projects have always had limited enough reach that I probably would've readily given commit access to anyone who cared enough to submit a PR.

An even more radical version of this idea was popularized by the late Pieter Hintjens —an amazing guy— in his post (and later book) “Confessions of a Necromancer”: http://hintjens.com/blog:125

One of the most inspiring engineering biographies I’ve ever read.

I applied this on my latest open source app, and I added two contributors, and so far it has worked fine!

However I only added them after the second or third pr, just to know if they are interested enough, and I always monitor the repository so if for some reason someone adds malicious code I will notice (and revert back).

It only works for certain kinds of projects. It makes sense for the icons repo, the contributions are minimal in variance.

In the case of large complex codebase though, the change is something that you will have to maintain going forward. The committer will have attained their goal of introducing a change they wanted and may not be around to deal with issues arising from that new feature. That's been my observation so far across many repos. Having a maintainer that chooses to accept or deny something is vital imo.

A project that I took over did this (before I took over), it's something the previous maintained did out of desperation as he didn't really want to maintain it any more, but also had trouble finding a new maintainer.

I have since revoked most people's access because:

- Most people were inactive and never contributed anything beyond the initial PR in terms of code or triaging issues.

- One person made a horrible PR that ended up having to be reverted (happened before I become active).

- One person, a Red Hat employee with no involvement for over a year, pretty much told someone to fuck off after asking a question on the issue tracker and closed the issue. The question was fine and this is very much NOT how I want to run projects (I have since removed his access).

In short, I'm not a fan, especially due that last issue: people being inactive is fine, a bad PR could be caught before release, but someone making "official" rude comments and closing issues: can't really do much about that

I created https://www.codeshelter.co/ years ago based on basically this premise (put your project in the community, any vetted member of the community can be automatically approved to become a contributor).

Unfortunately, there was much more demand than supply, so it never took off, but I still think it's a good model.

The Python/Django ecosystem has a similar concept with Jazzband – https://jazzband.co/

I've always found their projects to be high quality, well maintained, with a great community, and I quickly got to the point where I wouldn't think twice about adding a Jazzband dependency.

The model can work, and when it does it's great for the wider community.

> It sounds mad, I know.

No, it sounds like it usually works, even if not necessarily right after the first contribution. I've got commit access to many repositories just by contributing things to them (and sometimes it happened already after I was done with stuff I wanted to do there and never actually used that access).

It doesn't make sense if you have enough responsive maintainers, but many projects don't.

Incorporates one of the more astounding insights I’ve had over the years: complexity can often be reduced through trust.

'scuse me?? Are you allowed to talk that way on the internet??

This is a post about another post:

https://felixge.de/2013/03/11/the-pull-request-hack/

The tl;dr is up top:

    Whenever somebody sends you a pull request, give them commit access to your project.

which was also discussed on hacker news: https://news.ycombinator.com/item?id=5357417

Off-topic, but does anyone else love the design of that blog?

I think it's a pleasant departure from the norm, but I'm not sure I'd use it myself.

It has got some neobrutalism vibes

Thank you :-)

I designed it myself on top of a bare-bones WordPress theme.

The source code is at https://gitlab.com/edent/blog-theme/

Pull Requests - and commit rights - welcome!

It was so refreshing, cheers :)

Developers discover community management - news at 11.

Attackers would use it to gain access to inject malware if the project was at all popular. The Internet is a dark forest.

This only works as a skill filter. It doesn’t work in an environment with malicious actors.

If only the author had something to say about this problem...

> The Pull Request Hack probably isn't suitable if you're running big projects.

> And it is almost certainly a stupid idea if you write code which is actively used by multiple downstream projects.

> And if your stuff has even the slightest chance of compromising security then you're better off sticking to trusted members.

He did, but the problem is that any package anyone uses for anything could compromise security. So what could you use this with?

This sort of openness is an elegant weapon from a more civilized age.

These days the net is like a zombie movie where if you leave a crack in the boards covering the windows the arms will reach in.

>So what could you use this with?

Dead projects and projects that are so young they are more likely to be abandoned than finished.

It doesn't work for projects that already have enough vetted members.

Ok, lets consider the authors example of SuperTinyIcons (https://github.com/edent/SuperTinyIcons). What are some possible ways you could send a PR, gain commit rights and then use it for compromising security?

Javascript in the SVGs?

For what benefit? Infecting a few dozen probably fairly small sites while ruining the credibility of the contributor’s account? It’s extremely obvious and would pretty quickly get caught, even if the article author doesn’t keep too close a tab on the repo.

Doesn’t seem worth it to me.

Security and usability are always in conflict. Any time you take a hard line on security, you’ll find yourself crossing that line to get work done.

Oh, the language. You wouldn't hear it from a docker! Fecking this, fecking that...

Ah, you would. They use very bad language.

Effin' this and effin' that...

Oh, it was much worse than that, Ted, she was saying fu...

Now, Dougal!

Reply on news.ycombinator.com