gtfs_df 0.9.2 → 0.9.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 01e5ed5d1ce0ac7df81fe984b7bc8ca7716f0010672b7c3738d336ed1fbeaee0
-  data.tar.gz: 6f3a21037faaa76aadf9747bba7df94d7fdac9bb57e0b458f5f17696a7158d67
+  metadata.gz: 9c0746b1937afcfb7000425a59976b3eba8662778437ddf3adc02ec7729c3a49
+  data.tar.gz: e3a553aef868b4c29e06731f0a5e3984efc067a096001e90150743a80f3d45bc
 SHA512:
-  metadata.gz: 1a50d3af43551dda4a03a829ca1d4fd21e323c771181ee2238f9bcea41f4cb0e883d1d543bc26024d2c31d39456212e85bcd6beb6c2935ae98993f34cfd8816c
-  data.tar.gz: 7e1ef2009d75b6dc3f0b82d90183cb048a6a15db74ac7c4076e9acd264fed013a6920e65c95021b59bdfb263fbd4c014b9764b22f396f4957fcec0c7820cea87
+  metadata.gz: 0fccae16bb46da6db651da04ea2591fa17e991219d4d8eb9be3a9444ed0b1b190b315f9223b738f399f3b6fade41c2c2e1081d1aad921a39483426d5b86b1aca
+  data.tar.gz: b98eccb0861c46d8510eaaea6a97c48d6b1fa3687d1c35d445edbf19c349a6c03d45c3a2654da20ab577aa083003eff5335a8fcad6ee0d9f6f7f86a78c2ed33b

data/.conform.yaml

@@ -24,3 +24,4 @@ policies:
                   - docs
                   - ci
                   - qol
+                  - test

data/CHANGELOG.md

@@ -1,8 +1,23 @@
-## [0.9.2] - 2026-02-21
+## [0.9.3] - 2026-02-27
+
+### 🐛 Bug Fixes
+
+- Allow multiple filters
+- Refactor prune to keep caledar_dates-only dependencies
 
 ### 📚 Documentation
 
 - Add Brooke to the list of authors
+
+### 🧪 Testing
+
+- Ensure we don't drop trips and routes
+
+### ⚙️ Miscellaneous Tasks
+
+- Bump version to 0.9.2
+- Avoid converting series into arrays
+- Simplify trip pool reduction
 ## [0.9.1] - 2026-02-17
 
 ### 🐛 Bug Fixes

data/examples/split-by-agency/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: ../..
   specs:
-    gtfs_df (0.9.2)
+    gtfs_df (0.9.3)
       networkx (~> 0.4)
       polars-df (~> 0.22, < 0.24)
       rubyzip (>= 3.0, < 4.0)

data/lib/gtfs_df/feed.rb

@@ -121,19 +121,19 @@ module GtfsDf
         # Trips are the atomic unit of GTFS, we will generate a new view
         # based on the set of trips that would be included for each invidual filter
         # and cascade changes from this view in order to retain referential integrity
-        trip_ids = nil
+        trip_ids = Polars::Series.new.alias("trip_id")
 
         view.each do |file, filters|
           new_filtered = filter!(file, filters, filtered.dup)
-          trip_ids = if trip_ids.nil?
+          trip_ids = if trip_ids.empty?
             new_filtered["trips"]["trip_id"]
           else
-            trip_ids & new_filtered["trips"]["trip_id"]
+            trip_ids.filter(trip_ids.is_in(new_filtered["trips"]["trip_id"]))
           end
         end
 
         if trip_ids
-          filtered = filter!("trips", {"trip_id" => trip_ids.to_a}, filtered.dup)
+          filtered = filter!("trips", {"trip_id" => trip_ids.implode}, filtered.dup)
         end
       end
 
@@ -181,7 +181,7 @@ module GtfsDf
         df = filtered[file]
 
         filters.each do |col, val|
-          df = if val.is_a?(Array)
+          df = if val.is_a?(Polars::Series) || val.is_a?(Array)
             df.filter(Polars.col(col).is_in(val))
           elsif val.respond_to?(:call)
             df.filter(val.call(Polars.col(col)))
@@ -200,9 +200,16 @@ module GtfsDf
 
     # Traverses the graph to prune unreferenced entities from child dataframes
     # based on parent relationships. See GtfsDf::Graph::STOP_NODES
+    #
+    # The trips table has multiple parents (calendar, calendar_dates, routes,
+    # stop_times). We accumulate valid values from all of them and keep rows
+    # that match any parent, so trips referenced only via calendar_dates are
+    # not dropped when another edge is processed first.
     def prune!(root, filtered, filter_only_children: false)
       seen_edges = Set.new
       rerooted_graph = Graph.build(bidirectional: !filter_only_children)
+      accumulated_service_ids = Polars::Series.new("service_id", dtype: Polars::String)
+      trips_base_df = nil
 
       queue = [root]
 
@@ -245,37 +252,46 @@ module GtfsDf
           attrs[:dependencies].each do |dep|
             parent_col = dep[parent_node_id]
             child_col = dep[child_node_id]
-            allow_null = !!dep[:allow_null]
+            allow_null_flag = !!dep[:allow_null]
 
             next unless parent_col && child_col &&
               parent_df.columns.include?(parent_col) && child_df.columns.include?(child_col)
 
             # Get valid values from parent
-            valid_values = parent_df[parent_col].to_a.uniq.compact
-
-            # Annoying special case to make sure that if we have a calendar with exceptions,
-            # the calendar_dates file doesn't end up pruning other files
-            if parent_node_id == "calendar_dates" && parent_col == "service_id" &&
-                filtered["calendar"]
-              valid_values = (valid_values + calendar["service_id"].to_a).uniq
-            end
-
-            # Filter child to only include rows that reference valid parent values
-            before = child_df.height
-            filter = Polars.col(child_col).is_in(valid_values)
-            if allow_null
-              filter = (filter | Polars.col(child_col).is_null)
-            end
-            child_df = child_df.filter(filter)
-            changed = child_df.height < before
-
-            # If we removed a part of the child_df earlier, concat it back on
-            if saved_vals
-              child_df = Polars.concat([child_df, saved_vals], how: "vertical")
-            end
-
-            if changed
-              filtered[child_node.fetch(:file)] = child_df
+            valid_values = parent_df[parent_col].drop_nulls.unique
+
+            if child_node_id == "trips" && (parent_node_id == "calendar" || parent_node_id == "calendar_dates")
+              # Calendar + calendar_dates both define service for the same trips, so we want
+              # union semantics across those two parents (a trip is valid if it appears in
+              # either).
+              #
+              # Here we accumulate valid service_ids across calendar/calendar_dates, but only
+              # within the pool of trips that are already reachable from structural parents.
+              accumulated_service_ids = Polars.concat([accumulated_service_ids, valid_values]).unique
+
+              # Determine the base pool of trips:
+              # - If we've already restricted trips via structural parents (routes,
+              #   stop_times, shapes, etc), use that as the base.
+              # - Otherwise, like when filtering directly on trips, use the current
+              #   trips dataframe.
+              trips_base_df ||= filtered[child_node.fetch(:file)]
+              next unless trips_base_df && trips_base_df.height > 0
+
+              filtered[child_node.fetch(:file)] = trips_base_df.filter(
+                Polars.col("service_id").is_in(accumulated_service_ids.implode)
+              )
+            else
+              # Original single-edge logic for all other nodes
+              before = child_df.height
+
+              cond = Polars.col(child_col).is_in(valid_values.implode)
+              cond = (cond | Polars.col(child_col).is_null) if allow_null_flag
+              child_df = child_df.filter(cond)
+
+              if child_df.height < before
+                child_df = Polars.concat([child_df, saved_vals], how: "vertical") if saved_vals
+                filtered[child_node.fetch(:file)] = child_df
+              end
             end
           end
         end

data/lib/gtfs_df/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module GtfsDf
-  VERSION = "0.9.2"
+  VERSION = "0.9.3"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: gtfs_df
 version: !ruby/object:Gem::Version
-  version: 0.9.2
+  version: 0.9.3
 platform: ruby
 authors:
 - David Mejorado