Leo Famulari writes: > On Tue, Jan 23, 2018 at 07:45:12PM +0000, Christopher Baines wrote: >> >> Leo Famulari writes: >> >> > On Mon, Jan 22, 2018 at 09:08:22PM +0000, Christopher Baines wrote: >> >> * gnu/packages/terminals.scm (go-github.com-howeyc-gopass): New variable. >> > >> >> + (native-inputs >> >> + `(("go-golang.org-x-crypto-ssh-terminal" >> >> + ,go-golang.org-x-crypto-ssh-terminal))) >> > >> > Is it really just a native-input? Or could it need this library later, >> > at run-time? >> >> Well, the package doesn't build without it, and as far as I can tell the >> store output doesn't reference it. I believe that >> go-golang.org-x-crypto-ssh-terminal is a dependency of >> go-github.com-howeyc-gopass, so if you're using >> go-github.com-howeyc-gopass, then you'll need >> go-golang.org-x-crypto-ssh-terminal. Maybe it should be a propagated >> input instead? > > Go looks for dependencies by first looking for their source code (this > is why the go-build-system installs the source code by default). Once it > finds the source code, then it looks up the compiled objects. > > Finally, when building the end-user application (not just some library) > it links everything together in what's basically a static archive. > > If everything seems to work even though x-crypto-ssh-terminal is a > native-input, then I bet that whatever is using howeyc-gopass does not > exercise the part of howeyc-gopass's code using x-crypto-ssh-terminal. > > I think it's more correct here to propagate, unless > x-crypto-ssh-terminal is only used for howeyc-gopass's test suite or > something like that. Ok, I've made it a propagated input, as I think it's more than a build time dependency.