Erik Sipsma

> Less commands and flags in the CLI syntax: only what end users need > Allows a [gqurl](https://github.com/hasura/graphqurl)-style client UI, which doesn't require subcommands. > Smaller binary size. Always good,...

I just realized that we may want somewhere to put all the service tty debugger attachment stuff too (and other service socket types in the future). It may make sense...

This is awesome! Will save us so much manual testing time --- > Currently triggered on PRs, but it is quite long (~1h), wdyt ? I'm surprised by this as...

Just some initial unsorted thoughts. --- I think the options you listed are all worth considering, I do have some questions about the current state too though. In this example:...

Sort of late to the party here, but for Go I agree that we should wait for feedback from users after they start using the SDK w/ hidden IDs. Previously,...

Was this encountered only by manually stopping the container or did the container stop for some other reason? We set --restart always so if it's not manually stopped then it...

> I guess the main issue was that dagger.Connect was failing because of this and we didn't get a proper error message about it in the test. Yeah EOF is...

There are utils in buildkit for converting LLB -> graphviz: https://github.com/moby/buildkit/blob/ac4f39aaad34da48855d0beb2785a0666c68b895/cmd/buildctl/debug/dumpllb.go#L83-L104 Maybe useful for us (or maybe just inspirational if not reused directly)

Yeah this another one that's harder than it looks. I completely agree we can't have the shim exit 0 as that'll cache the failure. My immediate thought is to do...

Another implication of the above is that the execop failure is no longer a graphql error like it is today. I think that's fine? But it is different than today....