Bump bundler from 2.2.20 to 2.2.25

[jurre]

https://github.com/rubygems/rubygems/blob/master/bundler/CHANGELOG.md#2225-july-30-2021

Something we might get some support questions on is:
rubygems/rubygems#4647 but overall I think
that's a very good change! The GEM sources not being split has bitten
our users in the past, so happy to see that change.

In a recent version of bundler, passing a major version like
BUNDLER_VERSION=2 bundle --version no longer works.

I suspect this may have happened here: https://github.com/rubygems/rubygems/pull/4750/files

We now need to pass the exact version of bundler we're using, which is a
bit cumbersome as you can see from the diff, and it also means we'll
have to change a few more lines every time we update a version of
bundler.

I don't think the current behavior we were relying on was documented, so
I'm not sure if we can open an issue with the Bundler team.

This seems like the most straight forward fix to me, but I am definitely
open to other suggestions

[brrygrdn]

❓ Should we remove the FAILOVER constant or change it just to equal 1? Or should it always be 1.173 instead? Are there any scenarios where passing the full major version would cause problems?

[jurre]

We use it in two separate places, this detected_bundler_version method is used to gather metrics on the bundler major version, but in bundler_version we now need the full version.

[brrygrdn]

I wonder if we should smooth the process of updating bundler in future by having these values in text files we can read in various places?

It might be worth that or a helper script to do bundler updates, but given we have to run things in different bundler contexts that might get tricky.

[jurre]

I actually started on a script, but the only approach I can come up with is pretty crude and brittle, along the lines of:

uri = URI.parse("https://rubygems.org/api/v1/versions/bundler.json")
response = Net::HTTP.get_response(uri)
body = JSON.parse(response.body)

versions = body.map { |d| Gem::Version.new(d["number"]) }
latest_v2_release = versions.filter { |d| d.canonical_segments.first == 2 }.max
latest_v1_release = versions.filter { |d| d.canonical_segments.first == 1 }.max

dockerfile = File.read("Dockerfile")
dockerfile.gsub!(/\&\& gem install bundler -v 1.+$/, %(&& gem install bundler -v #{latest_v1_release} --no-document \\))
dockerfile.gsub!(/\&\& gem install bundler -v 2.+$/, %(&& gem install bundler -v #{latest_v2_release} --no-document \\))
IO.write("Dockerfile", dockerfile)

I'm not sure if such a script is worth maintaining, as we'll also need to update it as our code changes, I'd expect changes to the script would end up taking roughly as much time as going through the files?

There might be a better approach?

[brrygrdn]

Nice, I agree that making that maintainable and future proof isn't free compared to a find-and-replace. I don't think it should block this change to solve that either, we might not change this often enough it's worth the metaprogramming 🤔