Go: x/pkgsite: API for pkg.go.dev

[rhcarvalho]

Prior to pkg.go.dev, godoc.org has had a JSON API that can be used to, among other things, discover importers of a given package.

Example: https://api.godoc.org/importers/golang.org/x/net/html

Given that pkg.go.dev does a much better job at tracking importers thanks to Go Modules and the Module Proxy, it would be nice if the community could get access to a public API similar to that of godoc.org.

[julieqiu]

@rhcarvalho - it would be helpful to get a sense of what your current use cases are for api.godoc.org, and feature requests are for an API for pkg.go.dev.

It sounds like getting the importers for a package is one of them. With pkg.go.dev being module aware, what specific information about importers would be useful to surface via an API?

For example:

• importers for a specific version of a package
• only importers of the latest version of a package
• any importer for all versions of a package
• something else?

Additionally, what other information would be useful to you to surface via an API?

/cc @tbpg who has also mentioned wanting an API for pkg.go.dev

[rhcarvalho]

Without much thought, having an API to answer the more specific question "what are the importers of a specific version of a package" would make plausible to derive the answer to the other items in your list.

At the moment I consume the godoc.org API and scrape data from pkg.go.dev to answer the question "who uses my package". As far as I can tell the data in the "Imported By" go.dev tab is unrelated to the version of the package I'm currently browsing.

Here are the other endpoints in the GoDoc API: https://github.com/golang/gddo/blob/7365cb292b8bfd13dfe514a18b207b9cc70e6ceb/gddo-server/main.go#L901-L904

• /search: takes a q parameter. Can be useful to explore known packages/modules.
• /packages: attempts to return all known packages, no pagination. Doesn't seem too usable as is.
• /importers/: useful, hard to compute without central knowledge.
• /imports/: dispensable, easy to compute offline with go list.

So if we need a more specific request, here it is:

api.go.dev MVP

• /importers/{module-or-package}/@v/<version>: returns the list of importers of a given module/package at a given version. URL scheme could be tuned to match the Go Proxy specification (go help goproxy).
• /search/: matches the functionality of https://pkg.go.dev/search?q=hello, returns JSON instead.

[tbpg]

Another endpoint idea (standalone or as part of a set of info returned for a module): available versions of a given module.

[myitcv]

@tbpg

Another endpoint idea (standalone or as part of a set of info returned for a module): available versions of a given module.

Shouldn't people use cmd/go for that? Not least because of the fall-through semantics of the go env GOPROXY` variable. Noting https://github.com/golang/go/issues/37367

[julieqiu]

What information beyond https://proxy.golang.org/<module>/@v/list would be provided by that endpoint?

[tbpg]

Shouldn't people use cmd/go for that? Not least because of the fall-through semantics of the go env GOPROXY` variable. Noting #37367

There are some cases where cmd/go might not be available and a normal web API would be helpful. I think it's reasonable to assume pkg.go.dev will only return the modules/versions it knows about.

What information beyond https://proxy.golang.org/<module>/@v/list would be provided by that endpoint?

None. That works. :)

[myitcv]

There are some cases where cmd/go might not be available and a normal web API would be helpful

I'd be wary of encouraging tool authors to query pkg.go.dev/a proxy directly. Because it could well introduce skew compared to the answers from cmd/go.

[adamdecaf]

In my use-case I don't want to execute cmd/go binaries as I just want to grab some metadata (i.e. detected license of all dependencies) from a dozen modules and organize it on a webpage. If I need local clones of those modules it really complicates this internal tool.

[myitcv]

In https://github.com/golang/go/issues/37952 I raised the question of whether module/package license file information could be surfaced in the output of cmd/go list. Following a conversation on yesterday's golang-tools call, we concluded that doing so would be a bad idea; reasons summarised in https://github.com/golang/go/issues/37952#issuecomment-611527845.

@ianthehat instead suggested exposing the information via a pkg.go.dev API, leveraging the fact that the content and presentation on pkg.go.dev has already jumped through the relevant legal hoops.

This comment is therefore to explicitly request that we include license file information in the API. Thank you!

[tiegz]

👋 Thanks for all the new Go tools like pkg.go.dev, they're super useful. Some input on this topic from the perspective of Libraries.io:

• +1 to a /search/ endpoint, or maybe just adding pagination to this url so it's possible to get a list of all packages: https://index.golang.org/index?since=TIME&limit=LIMIT. Many popular package repositories offer a full listing like this, e.g. https://packagist.org/packages/list.json for Packagist

• What information beyond https://proxy.golang.org//@v/list would be provided by that endpoint?

Currently to get the publish times for all versions, you have to make N+1 requests: one for @v/list and then @v/xxx.info for each version and grab the "Time" value for each one. Assuming it's difficult to change the Go proxy spec, an API endpoint that returns all versions with metadata would be really nice.

• > As far as I can tell the data in the "Imported By" go.dev tab is unrelated to the version of the package I'm currently browsing.

Anyone know if there are plans to fix this?

[jba]

• > As far as I can tell the data in the "Imported By" go.dev tab is unrelated to the version of the package I'm currently browsing.

Anyone know if there are plans to fix this?

No immediate plans. We currently gather that information from import statements in the code, so there is no version information attached. The go.mod file doesn't have all the version information we need.

So we understand it's an approximation and we want to fix it, but it's going to take some time.

[pombredanne]

@julieqiu my use for an API would be to:

1. search and list packages
2. for a package, access all the information as available for a given package in each of the tabs, such as here https://pkg.go.dev/github.com/gin-gonic/gin?tab=licenses ... basically everything that is available as HTML in the frontend https://github.com/golang/pkgsite/tree/master/internal/frontend should be available in a JSON api

[pombredanne]

@julieqiu oh and please do not retire http://api.godoc.org/packages unless there is an alternative!

[eclipseo]

it would be helpful to get a sense of what your current use cases are for api.godoc.org, and feature requests are for an API for pkg.go.dev.

As a downstream package maintainer for Fedora, I'm also interested in an API. We have our own tool, Anitya, to track package releases, but it was not designed to track GIT commits. And many Go packages still don't publish version. So any information about latest published commit, with info like date of the commit, new dependencies, would be very helpful. I would gather data from the API in Python and compare it to the latest version we have for a given package.

I'm also interested in getting the License info, so we could find recursively all the licenses used in a static binary.

[pombredanne]

and BTW my general use case is for https://github.com/nexB/scancode-toolkit and related projects to provide my users with license, origin and dependencies details. And for https://github.com/nexb/vulnerablecode where we provide vulnerabilities details.

[StevenACoffman]

My present use case is discovering major version upgrades. Currently, we have to scrape pkg.go.dev. An alternative to pkg.go.dev API, would be if there were an "official" way (via go list for example) in the future to retrieve the latest major version of a module, we could use it in tools like go-mod-upgrade and icholy/gomajor but that would depend on golang/go#40357 (deprecated notification) and golang/go#40323 (notify about newer major versions). Thanks!

[icholy]

@StevenACoffman I've updated gomajor to find newer versions using the module proxy. The performance is acceptable.

[0x2b3bfa0]

My use case is importing modules/packages to another package manager (Guix) where builds are completely reproducible and monolithic vendoring is strongly discouraged. Having a JSON API would simplify the obtention of metadata like licenses, brief package descriptions, dependencies and hashes.