[cirwin]

Primary contributor to the feature here.

We went back on forth on this a lot, but it boiled down to wanting only one dependency graph per module instead of two. This simplifies things like security scanners, and other workflows that analyze your dependencies.

A `// tool` comment would be a nice addition, it's probably not impossible to add, but the code is quite fiddly.

Luckily for library authors, although it does impact version selection for projects who use your module; those projects do not get `// indirect` lines in their go.mod because those packages are not required when building their module.

[syvolt]

Thank you for working on it. It is a nice feature and still better than alternatives.

I'm not a library author and I try to be careful about what dependencies I introduce to my projects (including indirect dependencies). On one project, switching to `go tool` makes my go.mod go from 93 lines to 247 (excluding the tools themselves) - this makes it infeasible to manually review.

If I'm only using a single feature of a multi-purpose tool for example, does it matter to me that some unrelated dependency of theirs has a security issue?

[wakawaka28]

>If I'm only using a single feature of a multi-purpose tool for example, does it matter to me that some unrelated dependency of theirs has a security issue?

How is anyone supposed to know whether there's an issue or not? To simplify things, if you use the tool and the dependency belongs to the tool, then the issue can affect you. Anything more advanced than that requires analyzing the code.

[syvolt]

What if I'm already using techniques, such as sandboxing, to prevent the tools from doing anything unexpected? Why bring this entire mess of indirect dependencies into my project if I'm just using a tool to occasionally analyze my binary's output size? Or a tool to lint my protobuf files?

[wakawaka28]

If it's a build dependency, then you have to have it. If you don't like the size of the tool then take it up with the authors. I'm not a Go programmer by the way, this is all just obvious to me.

[syvolt]

The functionality we're discussing can be used for tools that are not build dependencies. They may be important for your project and worth having contributors be on the same version but not part of the build.

It will still add the dependencies of those tools as indirect dependencies to your go.mod file, that is what's being discussed.

[wakawaka28]

If you use the tool to develop your project then it is basically a build dependency. That is a sweeping generalization, but it's essentially correct in most cases.

[verdverm]

In addition, a good dependency security scanning tool can analyze reachability to answer this question for you

[wakawaka28]

That is a bit much to ask for IMO. In any case, the project may not be aware of how any given developer will use the tool. So who is to say that if you change the order of two parameters to the tool, the tool might not take a different path and proceed to hack your computer? You really don't want any of this problem. What you should ask for is for the tools' dependencies to be listed separately, and for each tool to follow the Unix philosophy of "do one thing well."

[verdverm]

> for each tool to ... "do one thing well."

There is a lot of merit to this statement, as applied to `go tool` usage and to security scanning. Just went through a big security vendor analysis and POCs. In the middle I saw Filippo Valsorda post [1] about false positives from the one stop shops, while govulncheck (language specific) did not have them. At the same time, there was one vendor who did not false positive with the reachability checks on vulns. While not always as good, one-stop-shops also add value by removing a lot of similar / duplicated work. Tradeoffs and such...

[1] https://bsky.app/profile/filippo.abyssdomain.expert/post/3ld...

[wakawaka28]

The similar/duplicated stuff can be rolled into libraries. Just don't make the libraries too big lol. I suspect there's less duplicated stuff than you think. Most of it would be stuff related to parsing files and command parameters, I guess.

[syvolt]

Reachability analysis on a tool that could be called by something outside of the project? We're talking about tools here after all - anything that can run `go tool` in that directory can call it. The go.mod tool entry could just be being used for versioning.

[verdverm]

I'm speaking of tools and processes independent of this "go tool" stuff that we already use in our CI pipelines

Big fan of Dagger over this go tool thing

I generally loath the use of comments for things other than comments

[arccy]

it probably doesn't, and good vulnerability scanners like govulncheck from the go team won't complain about them, because they're unreachable from your source code.

now, do you care about some development tool you're running locally has a security issue? if yes, you needed to update anyway, if not, nothing changes.

[eadmund]

> it boiled down to wanting only one dependency graph per module instead of two

Did you consider having tool be an alias for indirect? That would have kept a single dependency graph per module, while still enabling one reading one’s go.mod by hand rather than using ‘go mod’ to know where each dependency came from and why?

I know, a random drive-by forum post is not the same as a technical design …

[caust1c]

Having not looked at it deeply yet, why require building every time it's invoked? Is the idea to get it working then add build caching later? Seems like a pretty big drawback (bigger than the go.mod pollution, for me). Github runners are sllooooow so build times matter to me.

[cirwin]

`go tool` doesn't require a rebuild, but it does checking that the tool is up-to-date (which requires doing at least a bit of work).

This is one of the main advantages of using `go tool` over the "hope that contributors to have the right version installed" approach. As the version of the tool required by the project evolves, it continues to work.

Interestingly, when I was first working on the proposal, `go run` deliberately did not cache the built binary. That meant that `go tool` was much faster because it only had to do the check instead of re-running the `link` step. In Go 1.24 that was changed (both to support `go tool`, but also for some other work they are planning) so this advantage of `go tool` is not needed anymore.

[caust1c]

Thanks for the explanation and contribution! Very much appreciated :-)