text stringlengths 0 1.49k |
|---|
**Erik St. Martin:** What is this Go thing? |
**Sam Boyer:** It's weird, I hear there's rodents involved... I don't know much more than that. I've been interested in things related to package management for a long time. One of my earlier things in my open source world was I was the lead engineer who built the Git hosting platform that's still used by the Drupal pr... |
A few years ago - I think actually it was around the first GopherCon, 2014... I know [Matt Butcher](https://twitter.com/technosophos) and [Matt Farina](https://twitter.com/mattfarina) - all three of us were co-workers back in 2010 or so, and they're of course the authors of Glide now. At the first GopherCon we were sor... |
So they started to work on Glide; I didn't really come in until later... But there have been all these different projects all over the place from a bunch of different people that tackle at least part of the general problem. |
For me it was end of December 2015 that I started getting -- I can't remember what the precipitating event was, but I remember being really annoyed (that much is clear in my mind) with some tool that I was using. It may have been Glide, maybe it was GoDep - it doesn't matter. But I decided "Okay, I'm gonna try to descr... |
I'm proud to say that I stole five years of productivity from the world, because that's how many hours it takes for the number of people who read the whole thing to read the whole thing. \[laughter\] It's my favorite way to think about it. |
\[03:48\] This is the thing, too... This is a complex social problem as much as it is a complex technical problem, and I think it's fair to say that there was a lot of acrimony over this problem a year ago. The approach that I decided to take was I was gonna write this essay and then try to crank up this library that i... |
I feel like I had an [MVP](https://en.wikipedia.org/wiki/Minimum_viable_product) with GPS around GopherCon last year. When we tried to wrap this up, the shortest summary from there is that [Peter Bourgon](https://twitter.com/peterbourgon) decided to convene this package management committee, which started meeting Septe... |
Around October last year we started actually implementing the dep tool on top of GPS and then we released in early January. Now we're on the mad dash towards making this thing real and official, that is integrating it into the Go toolchain. |
**Erik St. Martin:** Let's talk about that a little bit, too... GPS, specifically - you had built a library around doing the package solving to be used potentially by multiple tools... |
**Sam Boyer:** Yeah. |
**Erik St. Martin:** Let's talk a little bit about that and kind of what role that plays in it, and the motivation behind having a library instead of one standard canonical tool that everybody uses. |
**Sam Boyer:** Right. So at least half of the motivation there really was social and not technical, at least at the outset. Again, I didn't wanna just ship yet another tool. I have the [XKCD](https://www.xkcd.com/) situation in mind - there are 14 competing standards. I know what I'll do - "I'll write another standard"... |
So instead I wanted to sort of push this down into a library, with the idea that if I could solve the problem in a general way where different tools could use it, then we would end up bringing the tools closer together. It will be a way to heal some of the divide in the tooling. |
After having done that, I'm really glad in retrospect, because it turned into a really interesting problem - how do you actually pull out the pieces of dependency management and create interfaces around them... And what are the pieces of dependency management? |
In many ways, it sort of flowed directly from writing this giant, long article where I tried to describe the design of a good system that does this into actually doing it in an abstracted way. |
**Carlisia Thompson:** One thing that I noticed when I was going over the GPS library - which stands for Go Packaging Solver - is that the language, just like you explained, was meant to be a foundation for the tools to build on top. It seems that that was the motivation. Now, going forward with GoDep, is that still a ... |
**Sam Boyer:** Yes, yes. That's a key part of the question, that I should address. We are very much hoping to have that become official. To be very clear, it's not like the dep tool has been blessed by the Go team at this point - this is still experimental - this is by no means guaranteed to happen, but we're on a good... |
**Erik St. Martin:** \[07:58\] And the committee that you were a part of in developing this pretty much had representation from the majority of the tools that existed out there, so it should in theory solve all of the use cases that each of the individual authors had for their own tools... |
**Sam Boyer:** I think so. We had the core committee that I listed the members of already, and then we also had the advisory committee. I'm gonna feel terrible if I forget somebody, but I believe that was Daniel Theophanes, [Dave Cheney](https://twitter.com/davecheney), Matt Farina, [Steve Francia](https://twitter.com/... |
There were a couple tools that we've missed perhaps, but yeah, we've tried very hard to make sure that we're covering all the use cases. There are a couple things, like right now we don't have the ability to support local mirroring in a way that [Glide](https://github.com/Masterminds/glide) does, but these things are o... |
**Brian Ketelsen:** So there won't be anything preventing Go users from continuing to do their own dependency management... |
**Sam Boyer:** There won't. We can't and shouldn't coerce people out of that; my experience is that doesn't tend to work too well. It's fair to say that most of the way we're approaching this is twofold. One is let's build all this community consensus momentum behind one tool, so that we can sort of avoid the pointless... |
**Brian Ketelsen:** Perfect. I don't ask because I want to use a different tool, I just ask because several people out there are probably already upset. "They're shoving this dep tool down my throat...!" |
**Sam Boyer:** Exactly, that's not a situation I wanna create. To me, that's just pointless acrimony and an argument we don't need to have if we make something which is good. |
**Brian Ketelsen:** Awesome. |
**Erik St. Martin:** The interesting problem with the multiple tools though, rather than having a ubiquitous way of doing this stuff is that when you have dependencies that have dependencies and all of them use some different vendoring tool it gets really complicated to be able to flatten the dependency tree and be abl... |
**Sam Boyer:** This is actually one of the earliest design choice that I made in GPS - GPS is a concept of both the manifest and a lock; this sort of two-file system... I could go into describing into what's all entailed there, but the brief version of it is manifests essentially describe constraints, and manifests onl... |
So GPS is built around the idea that the implementing tool passes in an implementation of an analyzer, and it really only has one method, which is `getManifestAndLock`. So each different tool that implements it could write an analyzer that just learns how to read its own manifest and lock file. But it's also totally po... |
What this means is that we could have a system - and I'll get to whether it's a good idea or not in a second - where we teach dep essentially to read and translate the metadata files from existing tools and do that all on the fly, transparently, inside of a solving process, and use that in order to inform solving decis... |
GoDep, which only records basically commit id SHA1's there's even a way to use that as advisory information, but not an actual constraint. Say, "Let's prefer to use this version of this crazy five-deep transitive dependency. But if we can't work it all out using that version, then it's alright, we'll move over to somet... |
\[12:07\] The question though, like I said, is whether this is a good idea or not. It was clearly a good idea for that when -- and there's an outstanding branch to convert Glide using GPS, and it's clearly a good idea in the Glide case, because Glide was never going to become the ubiquitous tool; it was going to exist ... |
However, with dep we're in a different situation. If we put that support in, then we've created a situation where we're actually kind of encouraging people to continue using the existing tools, which isn't necessarily what we wanna do. The drawback is if we don't put it in, then we don't really have any of that metadat... |
**Erik St. Martin:** I was going to point out that that kind of requires that all the tools at least recognize [semver](https://semver.org/) or something like that, because if they're just generically tracking dependencies and not the specific versions of those dependencies it gets hard to solve it above, right? Unless... |
**Sam Boyer:** Yeah... So there's a whole universe of different kinds of versions out there. You can have semver and you can have branches and you can have revisions, and you can have tags, and some of this is intersecting, some of it is sibling, some of it is not... So everything that you could do on a current tool wh... |
So it's possible for us to get all that information in. However, solving is pretty much a useless process unless there is some sort of range specified. If there's something that says, "This is the one and only version that this can work with", there's no real solving to do there. It's "This works, or we fail." So yes, ... |
However, the plus side of this is I sort of wrote it understanding that there are a lot of different tradeoffs here and it's hard to know from the outset which one of them is the right way to go. I optimized for making it easy to change the way that we do that. We can write the implementation to suck in code from diffe... |
**Erik St. Martin:** Brian, were you going to ask something? |
**Brian Ketelsen:** I was going to ask roughly the same question, but with a little bit more of a spin to "If we could steer the Go community in one direction, would we ask everybody to start tagging and using semver?" What's the thing that would make the most sense to the community as a whole? How can we encourage tha... |
**Sam Boyer:** \[15:53\]Yes, start tagging and using semver, that would be great. Dave Cheney put a [post](https://dave.cheney.net/2016/06/24/gophers-please-tag-your-releases) out about that a year ago or so, and of course there's been an argument for that for a while... But yes, start tagging your stuff with semver no... |
**Brian Ketelsen:** I've seen a lot of tools on GitHub, like Ruby-based tools or even Node or JavaScript-based tools that do automatic tagging and GitHub releases and such. Do you think that we could encourage adoption if that were built into a Go type command, like `go tag` and `go release` and `go increment version`? |
**Sam Boyer:** Yes. Right, someone already wrote -- we put a call out asking for something like this as part of the blog post that we wrote for the Advent series last December. Someone wrote something which I am embarrassed to say that I have not actually had time to look at yet. But yes, having this as a part of the t... |
Right now a release is you tag your thing and you push it, for whatever that means in the version control system that you're using. In the future, if we end up going the direction of having some sort of central registry - something analogous to npm or Crates or all the different ones that are out there - then the `rele... |
The real value now actually is the static analysis that can tell you which semver version you should bump. That's the thing that helps, because it makes sure that we're using semver correctly. |
**Brian Ketelsen:** This is completely off-topic and on-topic at the same time, so forgive me for my random ADD thing, but one of the interesting notes that we had today for news items, new projects that came out was the UpSpin... |
**Sam Boyer:** Yeah, [UpSpin](https://upspin.io/). |
**Brian Ketelsen:** Yeah, Andrew and Rob's distributed storage thingy... And it occurred to me that using that content-addressable distributed file store like that would be a really interesting way to tag and have a very large-scaled distributed repository of stuff. One of the things that everybody hates about GitHub -... |
**Sam Boyer:** No, totally... Yes, and actually I would say one of the immediate comments that I got a year ago after I published the article was people from IPFS saying, "Hey, we could host this on [IPFS](https://ipfs.io/). Wouldn't it be great?" And on the other side, I can't remember who it was, but someone also sai... |
But yes, I do think this is an interesting thing. In fact, I actually joked - to backchannel to Andrew - that "Hey, maybe we could use UpSpin to do this exact thing that you're saying. It's interesting to explore, maybe we can do it. |
**Brian Ketelsen:** Interesting. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.