x/pkgsite: linking to module source code

[jba]

This issue tracks requests to provide source links to code-hosting sites.

We won't be pursuing a new go-source meta tag (#39559). Instead, we will add code to the internal/source package as needed.

Use this issue to request that we add a site. Your request can take a few different forms. In order of preference:

1. Write a CL. Examples: https://golang.org/cl/245379, https://golang.org/cl/245380. If the site requires more than just adding a pattern, describe your changes here first.

2. Provide the pattern and templates that should be added to the patterns variable in source.go, or describe the linking scheme if it doesn't fit our patterns.

3. Provide a link to a repo on the site. We'll determine the patterns.

No need to file a separate issue for your site; just comment here. An exception is the existing issue for launchpad.net: #39627.

[myitcv]

We won't be pursuing a new go-source meta tag (#39559)

Please can you provide some details on why? Because in implementing #39559 others (i.e. people/tools/whatever other than pkgsite) can benefit from the surfaced information.

The two don't appear to be mutually exclusive either; i.e. we can pursue this approach in the short term with #39559 for the longer term?

[mvdan]

I agree with Paul; if I host a personal source code hosting site, I shouldn't need to add it to a global list of well-known code hosting sites. The HTTP meta tags should be plenty for the URLs to describe what the setup looks like.

[dmitshur]

I think there is a benefit to using this approach as a next step to support linking to module source code.

Iterating on this approach in an internal package can help inform the design of an HTTP meta tag, if that were to be pursued in the future. If something about the internal API turns out to work poorly, or an opportunity to improve or simplify appears, code in the internal package can be refactored atomically. Once the internal package stabilizes, someone can decide it'd be worth creating an equivalent HTTP meta tag design that websites can choose to implement instead, and the existing entries in the internal package can serve as test cases.

[mvdan]

Right, agreed that keeping a list is good in the short term. Even if a meta tag was accepted and announced today, it could take months or years for it to see any widespread use.

What I think we were aiming for is to have a tracking issue for what the long-term solution to the problem would be, so that anyone running their own git server can follow that issue and trust that it won't be closed until there's a proper solution. I think we can all agree that a very very long list of sites can be a decent solution for most people in the short term, but it will never be enough for the practically endless amount of source code hosting sites out there.

All this to say - @jba has reopened the meta tag issue, so that's a good start :) I only ask that, if the proposal gets rejected, we open a new issue about the problem statement, which is "how do we support any code hosting site without hard-coding it into one big global list".

[jba]

We identified a few important problems with the go-source-v2 idea, most notably:

• The only guarantee for a module is a zip file. There may not even be a publicly browseable source code repository available.
• The extra meta tag is only accessible if you go back to the original redirect page. It's not proxyable like all the other module metadata, nor is it preserved if the origin goes away. (Part of why it's not proxyable is that it's only a godoc.org add-on, not something the Go toolchain has ever defined.)
• The rules around {revision} are very specific to Git repos and bake in details about pseudo-versions that may change in the future.

This makes us pretty confident that go-source-v2 as proposed in the other issue is not the right long-term solution. It's a little bit more module-friendly in that it knows what a version is, but it's not module-friendly enough. More thought is clearly needed.

And even if we added that go-source-v2 support today, we'd need every go get redirector to be updated before any links would start working. That's a lot to ask for a design that we're not even sure is right, especially when a better design might not require any changes at all. The right answer might be to display the code directly from the zip files, or it might be to put some info in the zip file that helps find a source display, or it might be something else entirely. We don't know.

For now, instead of defining a new tag that will require widespread adoption but still not be completely right, it seems best to get the most common sites working by making changes to pkg.go.dev directly, and then revisit the topic when we've had more time to think about the right path forward.

[dmitshur]

1. Write a CL. Examples: https://golang.org/cl/245379, https://golang.org/cl/245380. If the site requires more than just adding a pattern, describe your changes here first.

I tried to create a CL that would add support for packages hosted on dmitri.shuralyov.com, a custom server that hosts some Go packages, but it didn't seem to be possible to do by just adding a pattern. So, as you asked, describing it here.

The server itself doesn't support displaying source code in the browser, so it defers to another website that can display source code of Go packages. It was possible to implement this with the following go-source meta tag, using the dmitri.shuralyov.com/gpu/mtl package as an example:

$ curl -s 'https://dmitri.shuralyov.com/gpu/mtl?go-get=1' | grep 'go-source'
<meta name="go-source" content="dmitri.shuralyov.com/gpu/mtl https://dmitri.shuralyov.com/gpu/mtl/... https://gotools.org/dmitri.shuralyov.com/gpu/mtl https://gotools.org/dmitri.shuralyov.com/gpu/mtl#{file}-L{line}">

The go-source tag includes 4 parts (as documented): the import path corresponding to the repository root, the repository's home page, a URL pointing to the directory corresponding to the Go package, and a URL template for a specific line in a specific file. This makes it work on godoc.org (e.g., see https://godoc.org/dmitri.shuralyov.com/gpu/mtl).

I learned there are differences in how the templates work in pkgsite, which contributed to me not seeing a way this package could be made to work without additional modifications:

1. There's no pattern for the repo/module home page. It's always just {repo} and isn't configurable, so I couldn't find a way to add the "/..." suffix as done in the go-source meta tag.
2. The {repo} substitution unconditionally includes a "https://" prefix immediately followed by the repo root, and I didn't see a way to use a different domain to display the source code.
3. The {file} substitution is not just the file name (as it is on godoc.org), but the full path to file from the module root. I didn't see a way to split apart just the name of the file, needed for my template. Similarly, I didn't see a way to compute the import path of Go package whose file (or directory) is to be viewed, as that requires combining {repo} with just the directory of {file}.
4. On godoc.org, the go-source meta tag applies to an individual package, so it's possible for a custom server to provide templates that work for each package. The templates in pkgsite seem to apply to a repository (or module), so I couldn't work around limitations of templates on the side of the server. (This isn't a problem if there's an equivalent way of doing what I needed via the template.)

I've sent a DNS/DNR CL 246239 to show my local prototype in case it's easier to look at some code to understand some details and/or give feedback, but I certainly don't expect it to be merged as is.

I'm not sure if it's in scope of pkgsite to support what's needed to make that package work, or how to best do it, but I wanted to share what I learned from my experience. If it is in scope, please let me know what you think is a good next step. Thank you.

[jba]

@dmitshur We definitely want to make your site work. Your CL looks like it's on the right track. We can continue the discussion there.

[gopherbot]

Change https://golang.org/cl/246239 mentions this issue: internal/source: add support for dmitri.shuralyov.com/... packages