Inconsistent usage of FN_REGISTRY

[denismakogon]

Here's the thing. fn deploy forces developers to set FN_REGISTRY or use --registry, but fn build doesn't do so, internally they use the same functionality.

My question is, should both fn build and fn deploy act the same in terms of relying on FN_REGISTRY?

[carimura]

two options:

1. add FN_REGISTRY requirement to fn build
2. remove FN_REGISTRY requirement from fn deploy

option 2 doesn't seem to make sense because you have to push somewhere.

option 1, what if the "user" of the CLI in this case is only doing a build for a CI system or something and they have a custom push requirement?

[rdallman]

2. is not quite the case, it's possible to do this now if providing 'fully qualified' function names (aka image names), e.g. rdallman/yo.

I do hope we murder the FN_REGISTRY construct altogether https://github.com/fnproject/cli/pull/144, with a pipe wrench, in the ballroom, mustard...

it seems fine to support for build, rather than require it, to be the same as deploy -- but again the idea doesn't work very well for all use cases and should probably get replaced with something that works better for us (it creates an ambiguous grammar with private images, see linked PR)