Fix flatware rspec spec/foo sometimes running the entire test suite

lib/flatware/cli.rb

@@ -76,6 +76,10 @@ def try_setpgrp
     def workers
       options[:workers]
     end
+
+    def worker_spawn_count(jobs)
+      [workers, jobs.length].min
+    end
   end
 end
 

lib/flatware/cucumber/cli.rb

@@ -16,24 +16,22 @@ class CLI
       'parallelizes cucumber with custom arguments'
     )
     def cucumber(*args)
-      config = Cucumber.configure args
+      jobs = load_jobs(args)
 
-      ensure_jobs(config)
+      formatter = Flatware::Cucumber::Formatters::Console.new($stdout, $stderr)
 
       Flatware.verbose = options[:log]
-      sink = options['sink-endpoint']
-      Worker.spawn(count: workers, runner: Cucumber, sink: sink)
-      start_sink(
-        jobs: config.jobs,
-        workers: workers,
-        formatter: Flatware::Cucumber::Formatters::Console.new($stdout, $stderr)
-      )
+
+      spawn_count = worker_spawn_count(jobs)
+      Worker.spawn(count: spawn_count, runner: Cucumber, sink: options['sink-endpoint'])
+      start_sink(jobs: jobs, workers: spawn_count, formatter: formatter)
     end
 
     private
 
-    def ensure_jobs(config)
-      return if config.jobs.any?
+    def load_jobs(args)
+      config = Cucumber.configure args
+      return config.jobs if config.jobs.any?
 
       abort(
         format(

lib/flatware/rspec/cli.rb

@@ -23,8 +23,10 @@ def rspec(*rspec_args)
       )
 
       Flatware.verbose = options[:log]
-      Worker.spawn count: workers, runner: RSpec, sink: options['sink-endpoint']
-      start_sink(jobs: jobs, workers: workers, formatter: formatter)
+
+      spawn_count = worker_spawn_count(jobs)
+      Worker.spawn(count: spawn_count, runner: RSpec, sink: options['sink-endpoint'])
+      start_sink(jobs: jobs, workers: spawn_count, formatter: formatter)
     end
   end
 end

lib/flatware/rspec/job_builder.rb

@@ -43,11 +43,17 @@ def jobs
       private
 
       def balance_jobs(bucket_count:, timed_files:, untimed_files:)
-        balance_by(bucket_count, timed_files, &:last)
-          .map { |bucket| bucket.map(&:first) }
-          .zip(
-            round_robin(bucket_count, untimed_files)
-          ).map(&:flatten)
+        timed_groups = balance_by(bucket_count, timed_files, &:last)
+                       .map { |bucket| bucket.map(&:first) }
+        untimed_groups = round_robin(bucket_count, untimed_files)
+
+        # When the files can't be evenly divided between groups, the first groups
+        # in each of timed_groups & untimed_groups will have more files.
+        # By reversing one of them before combining them, we can improve the final distribution.
+        timed_groups
+          .zip(untimed_groups.reverse)
+          .map(&:flatten)
+          .reject(&:empty?)
           .map { |files| Job.new(files, args) }
       end
 

spec/flatware/rspec/job_builder_spec.rb

@@ -16,9 +16,10 @@
 
   let(:persisted_examples) { [] }
   let(:files_to_run) { [] }
+  let(:worker_count) { 2 }
 
   subject do
-    described_class.new([], workers: 2).jobs
+    described_class.new([], workers: worker_count).jobs
   end
 
   context 'when this run includes persisted examples' do
@@ -65,5 +66,26 @@
         )
       end
     end
+
+    context 'and there are more workers than example-files' do
+      let(:files_to_run) do
+        %w[
+          fast_1_spec.rb
+          slow_spec.rb
+          new_1_spec.rb
+        ]
+      end
+      let(:worker_count) { 5 }
+
+      it "doesn't return empty job-groups" do
+        expect(subject).to match_array(
+          [
+            have_attributes(id: include('./fast_1_spec.rb')),
+            have_attributes(id: include('./slow_spec.rb')),
+            have_attributes(id: include('./new_1_spec.rb'))
+          ]
+        )
+      end
+    end
   end
 end