run repo2docker in CI

[lukasheinrich]

Description

right now we trigger a binder build but i'm not sure, do we get a response / feedback iif the bulid fails? We could use repo2docker to do the binder-like build locally

https://twitter.com/m_deff/status/1277064783662141441

[matthewfeickert]

Right now we trigger a binder build but i'm not sure, do we get a response / feedback if the bulid fails?

No. trigger_binder.sh just curls the endpoint and gives the return code

pyhf/binder/trigger_binder.sh

Lines 6 to 16 in 64dbce0

	curl -L --connect-timeout 10 --max-time 30 "${URL}"
	curl_return=$?
	# Return code 28 is when the --max-time is reached
	if [ "${curl_return}" -eq 0 ] || [ "${curl_return}" -eq 28 ]; then
	if [[ "${curl_return}" -eq 28 ]]; then
	printf "\nBinder build started.\nCheck back soon.\n"
	fi
	else
	return "${curl_return}"
	fi

and as we're using the postBuild Binder config which is basically just installing how we would in CI

pyhf/binder/postBuild

Lines 1 to 2 in 64dbce0

	python -m pip install --upgrade .[complete]
	python -m pip install altair

the repo2docker build on the Binder side shouldn't fail. That being said though, the idea of using repo2docker in CI (to test and prebuild the Docker image for Binder so that work was offloaded from them) were the reasons that I was looking at machine-learning-apps/repo2docker-action (and why there is now a pyhf/binder Docker image). The discussions with Tim and Chris on these Issues:

• machine-learning-apps/repo2docker-action Issue 26
• machine-learning-apps/repo2docker-action Issue 25

make it seem like it might not be a good fit for us as we want to use Binder config files in the top level binder/ dir.

@lukasheinrich did you have additional ideas related to this? Additionally, can you elaborate on what you and @rochaporto were thinking about here?