Calculate build sizes relative to merge base

[gaearon]

See #11882 (comment).

[sophiebits]

Can we handle the case where there is no build for a particular SHA? One case this could happen is if someone pushes two commits to master at once and there's no build for the first one, but your PR is based off the first one. Probably uncommon but we should at least handle the error gracefully.

Either by doing nothing or trying to go back and try some previous commits.

[gaearon]

Hmm, we don't want the merge base with master here.

This fails because we force check out PR commit as master on CI (and thus its merge base with master is that commit itself).

We want the branch that the PR was sent against.
I'm not sure how to get that.

[sophiebits]

maybe CIRCLE_COMPARE_URL?

[gaearon]

Seems like that's just a URL so we'd have to parse it.

Also not sure what this means:

Not present for builds that are triggered by GitHub pushes.

Aren't all builds triggered by pushes? (I guess this just means it's PR-only)

[sophiebits]

Yeah, PR vs push.

[gaearon]

Even CIRCLE_COMPARE_URL isn't helpful 😞
It only shows the last commit, e.g. af469c4...27836c7 for the last run.
Not the actual PR's base.

[gaearon]

Looks like @orta also bumped into this. danger/danger#45

[orta]

Yeah, Circle can be real tricky to gather this kind of information - in Danger Ruby we use the Circle API to grab the PR and then use that to pull out the metadata like base/head.

[zpao]

One thing worth noting - if we still aren't doing builds for docs, the merge base might be a commit that we don't have a build for.

Otherwise, I made it possible to jump straight to the commit without needing to look at the commit time. Just /builds/branch/_commits/SHA eg: http://react.zpao.com/builds/master/_commits/247c5241706356033cd5bf1dc31197e058853e13/

[gaearon]

Closing since this is broken.

[sophiebits]

@gaearon We can use GitHub's API as @orta described, right? Maybe we can goodfirsttask it if you don't have time to follow up.

[gaearon]

I think @orta did something in #11865 that fixes this.

[orta]

Yep - that's what it does, uses the PR's base sha 👍 ( sorry for that lack of updates to the PR, I got distracted today for my OSS time with prettier prettier/prettier#3669 )