Public Lab Research note


  • 0

Standardizing open source hardware publication practices with contributors.json

by warren |

Alicia Gibb of OSHWA and I (I'm on their board) cooked up the idea, chatting a few months ago, of a "spec" for a file similar to the CONTRIBUTORS or CREDITS text files found in some FOSS projects -- a kind of text file format for describing your project and providing key information. So, a little broader than just listing contributors, but really fulfilling all of the Open Source Hardware Definition. As it'd be machine readable, it could form the basis for a repository -- the URLs to these files in different peoples' online posted docs could be scraped and indexed.

This sample has a lot of ideas, some better than others, but the most minimal example would be really short; just title, version, description, license, contributors, and source_files.

Link to Gist example: https://gist.github.com/jywarren/a4dafb0f02e731eacb14

What do you think? It's a slightly more concrete way to kick off any discussions of repositories, and people could begin creating these "contributors.json" files now, which could help in later attempts to create a more comprehensive directory of open hardware projects.

Basically this is just a rough proposal, but it wouldn't be too hard for Public Lab tool pages to provide such a standardized JSON file, and perhaps others would adopt it as a draft standard?



open-hardware attribution contributing oshwa

11 Comments

I like that its JSON and that there isn't a hard coded standard. It seems more like a guideline or a best practice, but of course deviation is acceptable.

As an example, a contributor might want to list name, but not email. The JSON structure concedes that possibility without requiring it the way XML might, given a proper DTD.

I'm not sure I understand the builds_upon field. If open hardware ends up being like open software, this one field would be some pale comparison of Github's network graph seen on many popular repositories. I just grabbed one of Github's most popular repos this month and here's an example graph: https://github.com/Flipboard/react-canvas/network

A forking, branching, merging network of contributions is best tracked directly by git, mercurial, or something along those lines. Trying to embed that in a human-readable format is going to lead a hairy mess if open hardware becomes anything like open software.


I was a bit inspired by bower's bower.json or npm's package.json formats -- and actually, one thought was that the builds_upon would link to more contributors.json files. And that it helps to show upstream licensing. So maybe more following the "includes" paradigm than mirroring how upstream contributions to the same software project are shown.


I think linking to other files would be alright. In that sense, you are emulating a graph by issuing pointers as opposed to trying to copy-by-value important details like license into your copy of contributors.json. In the git comparison, you're enumerating previous commits or branches but not their contents.

There are some issues here, though. Here are two I came up with by thinking about what git does and how this fails to replicate it.

Cloning a repository maintains a snapshot of the entire history up to that point, but not necessarily after it. If contributors.json is changed upstream, that link doesn't mean the same thing it did when it was created.

Cloning a repository maintains a copy of all commits up to that point, including all user contributions. If the pointer to some upstream contributor.json becomes a broken URL, you lose that all historical context and previous contributors from that section of the tree.


ah, to look to bower and npm again on that point of the linked-to-projects changing, we could link to a specific version of the contributors.json file as they do -- like "link": "https://github.com/jywarren/irkit/contributors.json#0.1.2" or:

"link": "https://github.com/jywarren/irkit",
"version": "v0.1.2",


That covers the first problem but not the second.

So I think you'd need to do one of two things to really preserve the historical contributors:

  • Maintain a complete history of the repository by way of version control and build it into the standards and practices. This more or less guarantees all contributors are tracked, and it does not attempt to turn this tracking into something user friendly which is doomed to fail.

  • Maintain a complete and full copy of every contributors.json, frozen at the time of forking, inside the current contributors.json. This will become very large and ugly, but it's moderately manageable with technology. If one person merged two projects with a common ancestor, then there'd be a lot of redundant information in the full copies of contributors.json of each project found inside that one person's contributors.json.


Timestamping would probably be better than versioning because version numbers don't always get used in proper ways. Think of forking a feature branch that is not yet complete and turning it into a new master in a different repository: version didn't get changed yet because the feature branch wasn't yet merged to master. A timestamp should be solid because all changes tend to be timestamped. git would handle this by tracking the entire commit tree for you by tag or commit hash.

Versions could certainly be added along with timestamps. That would help distinguish different forks from the same project being worked on in parallel, the equivalent to git branches.


Oh here's a crazy idea. You could point at other contributor.json files by reference using a universal identifier, aka UUID. https://en.wikipedia.org/wiki/Universally_unique_identifier

The process of generating a UUID for contributor.json for a given project might be something like this:

  1. the code is updated.
  2. the contributor.json file is updated with any new authors as needed
  3. all files in the project at the current time are hashed using e.g. SHA256 and then all the hashes are XORed (this way order does not matter).
  4. the time itself, in seconds since epoch, is concatenated onto the SHA256.
  5. the remainder of the UUID's bits are randomly generated.
  6. the contributor.json is updated to contain the project's current UUID

A new project would have a UUID of all 0s in the contributor file as an initial seed, but step 6 would replace the 0s.

Now your contributor.json file can maintain a series of UUID pointers to the UUID found in each of the upstream's current contributor.json.

In this way, each change could be tracked by a UUID without copying the repository and is highly likely to be a unique fingerprint for that repository at that time. This is a bit of a one-way mapping, unfortunately. Backing out of it would be difficult. It might require caching this information in a repository or something so that a UUID can be mapped to its originating source.


@warren

What you are talking about reminds me of what Mozilla Science were talking about at the Open Knowledge meetup which Public Labs was also invited to.

I think they are on the same page as you when it comes to standardisation.

@btbonval

My partner uses https://bitbucket.org/.


@amysoyka bitbucket can use git or mercurial, which are both options I noted in a previous comment. GitHub is not the same as git. I actually have one or two projects, using git, that are on both GitHub and BitBucket at the same time :)


I had some ideas (talking with @rjstatic) about how a more Bower- or NPM-style utility could parse such files... these are just roughly sketched out ideas -- say we called it "newt":

  • newt init -- questoinnaire
  • newt register -- tests for presence of req'd docs, clones repos or zips
  • newt compile bom -- aggregate/merge BOMs of nested projects
  • newt compile bom <string> -- aggregate/merge BOMs with links matching provided string like "digikey.com"
  • newt compile price <int> -- calculate unit price for int units
  • newt compile contributors -- compile contributors of nested projects


While not MVP, newt compile price <int> will be so cool in the future when there is adoption of this sort of thing.


You must be logged in to comment.