Add fallback for applications that bypass RubyGem to load dependencies

[arielvalentin]

The GitHub monolith bypasses RubyGems to load dependencies and as a result is unable to install any auto-instrumentation gems that use Gem.loaded_specs to determine compatibility.

@ericmustin implemented several fixes in ddtrace to mitigate this issue. We discussed introducing a helper method but then also pointed out that there are some cases where gem authors do not include a VERSION constant or follow a consistent pattern to namespace the constant.

We may want to consider adding a helper similar to what @ericmustin described in one of the PRs:

# This allows instrumentation authors to provide a default value
Datadog::Utils::Gem.loaded_specs_version('dogstatsd-ruby') { Datadog::Statsd::VERSION }

That would make it so instrumentation authors could do something like this:

module OpenTelemetry
  module Instrumentation
    module MyGem
      class Instrumentation < OpenTelemetry::Instrumentation::Base
        compatible do
          loaded_specs_version("my-gem") { MyGem::VERSION } <= "2.0.1"
        end
      end
    end
  end
end

Longer term I propose that the Instrumentation::Base DSL could possibly include a declaration that uses RubyGems pessimistic version constraint e.g.:

module OpenTelemetry
  module Instrumentation
    module MyGem
      class Instrumentation < OpenTelemetry::Instrumentation::Base
        supported_versions '>= 2.2.0', '< 2.3.0'
      end
    end
  end
end

# When the version is located in an submodule:  
module OpenTelemetry
  module Instrumentation
    module MyGem
      class Instrumentation < OpenTelemetry::Instrumentation::Base
        supported_versions '>= 2.2.0', '< 2.3.0', "MyGem::ChildModule::GrandChild::VERSION"
      end
    end
  end
end

Related

• Slack Conversation
• Fix in ddtrace
• Fix for Capybara @seejohnrun @joelhawksley

cc: @SophieDeBenedetto @wogri

[arielvalentin]

Instead of falling back programmatically we will rely on gemspec declarations for default compatibility checks:

#1016 (comment)