Latest endpoint for the Activity API not returning all of the latest new gems

[activefx]

Steps to Reproduce

An API call to Gems.latest

Expected Behavior

Gems.latest is supposed to return the 50 gems most recently added to RubyGems.org for the first time.

Current Behavior

Gems.latest does not appear to return the gems most recently added if a new gem publishes multiple new versions in rapid succession (I believe there may also be other edge cases, such as placeholder gems and beta versions). For example, https://rubygems.org/gems/llama_bot_rails was first published on June 15th, 2025. A call to Gems.latest at this moment returns gems created all the way back to June 11th, 2025, but does not include the llama_bot_rails gem.

Possible Solution

One possible solution would be to use a subquery to find the the first version created before the current version of a Rubygem:

  def self.new_pushed_versions(limit = 5)
    subquery = <<~SQL.squish
      versions.rubygem_id IN (
        SELECT v1.rubygem_id
        FROM versions v1
        WHERE NOT EXISTS (
          SELECT 1
          FROM versions v2
          WHERE v2.rubygem_id = v1.rubygem_id
          AND v2.created_at < v1.created_at
        )
        ORDER BY v1.created_at DESC
        LIMIT :limit
      )
    SQL

    where(subquery, limit: limit)
      .where(latest: true)
      .by_created_at
  end

However, in local testing, this seems slow. Couldn't this method just return Rubygem.order(created_at: :desc).limit(50) instead?

[simi]

Hello @activefx. I think I see your problem.

Gems.latest is using /api/v1/activity/latest.json (https://github.com/rubygems/gems/blob/ec5c7d0dd95d6fa15e4358022c5803c9281ad4bb/lib/gems/v1/client.rb#L248-L258) which uses (as you're mentioning) new_pushed_versions scope.

rubygems.org/app/models/version.rb

Lines 161 to 167 in 3b8fd06

	# This method returns the new versions for brand new rubygems
	def self.new_pushed_versions(limit = 5)
	subquery = <<~SQL.squish
	versions.rubygem_id IN (SELECT versions.rubygem_id FROM versions
	GROUP BY versions.rubygem_id HAVING COUNT(versions.rubygem_id) = 1
	ORDER BY versions.rubygem_id DESC LIMIT :limit)
	SQL

Which for some reason (I'll explain later) filters only gems with only 1 release (version).

I think this check was in there for historical reasons, when it was possible to create Rubygem without actual version or today when you yank all versions (which should be filtered out IMHO). I have checked all gems published in last year and there is no such a case (Rubygem without version).

Today it is not possible to create such a gem, when gem is marked as indexed. So I think we can fix two problems actually in here by using Rubygem.order(created_at: :desc).where(indexed: true) aka Rubygem.with_versions.order(created_at: :desc).

Would that work for you @activefx? I'm just not sure how much of breaking change this is. I do consider this more a fix than breaking change. But we can also make v2 version and keep the original one as is.

[activefx]

Yes, thank you, I think this would work!

[simi]

Yes, thank you, I think this would work!

What do you think about v1 update or new v2 endpoint instead? Would you like to contribute this? I can help if needed.