Merge branch '17995-filter-by-comparing-attrs'
authorTom Clegg <tom@curii.com>
Fri, 10 Sep 2021 14:25:58 +0000 (10:25 -0400)
committerTom Clegg <tom@curii.com>
Fri, 10 Sep 2021 14:25:58 +0000 (10:25 -0400)
closes #17995

Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curii.com>

doc/api/methods.html.textile.liquid
lib/controller/router/router_test.go
sdk/go/arvados/resource_list.go
sdk/go/arvados/resource_list_test.go
services/api/lib/record_filters.rb
services/api/test/functional/arvados/v1/collections_controller_test.rb

index e051ab66fa7afa18d8e52b09741c292f5e1faa9c..fd529179283f84fbdfe59ad07c8cb95bd9185209 100644 (file)
@@ -136,6 +136,22 @@ table(table table-bordered table-condensed).
 
 Note that exclusion filters @!=@ and @not in@ will return records for which the property is not defined at all.  To restrict filtering to records on which the subproperty is defined, combine with an @exists@ filter.
 
+h4(#filterexpression). Filtering using boolean expressions
+
+In addition to the three-element array form described above, a string containing a boolean expression is also accepted. The following restrictions apply:
+* The expression must contain exactly one operator.
+* The operator must be @=@, @<@, @<=@, @>@, or @>=@.
+* There must be exactly one pair of parentheses, surrounding the entire expression.
+* Each operand must be the name of a numeric attribute like @replication_desired@ (literal values like @3@ and non-numeric attributes like @uuid@ are not accepted).
+* The expression must not contain whitespace other than an ASCII space (newline and tab characters are not accepted).
+
+Examples:
+* @(replication_desired > replication_confirmed)@
+* @(replication_desired = replication_confirmed)@
+
+Both types of filter (boolean expressions and @[attribute, operator, operand]@ filters) can be combined in the same API call. Example:
+* @{"filters": ["(replication_desired > replication_confirmed)", ["replication_desired", "<", 2]]}@
+
 h4. Federated listing
 
 Federated listing forwards a request to multiple clusters and combines the results.  Currently only a very restricted form of the "list" method is supported.
index 0330ec4252c9ad3ee8f461faf9ce7508c17bd3fc..639d2a28b4df5f647da44e9cd946863ff49d6abc 100644 (file)
@@ -47,6 +47,7 @@ func (s *RouterSuite) SetUpTest(c *check.C) {
 func (s *RouterSuite) TestOptions(c *check.C) {
        token := arvadostest.ActiveToken
        for _, trial := range []struct {
+               comment      string // unparsed -- only used to help match test failures to trials
                method       string
                path         string
                header       http.Header
@@ -120,6 +121,32 @@ func (s *RouterSuite) TestOptions(c *check.C) {
                        shouldCall:  "CollectionList",
                        withOptions: arvados.ListOptions{Limit: 123, Offset: 456, IncludeTrash: true, IncludeOldVersions: true},
                },
+               {
+                       comment:     "form-encoded expression filter in query string",
+                       method:      "GET",
+                       path:        "/arvados/v1/collections?filters=[%22(foo<bar)%22]",
+                       header:      http.Header{"Content-Type": {"application/x-www-form-urlencoded"}},
+                       shouldCall:  "CollectionList",
+                       withOptions: arvados.ListOptions{Limit: -1, Filters: []arvados.Filter{{"(foo<bar)", "=", true}}},
+               },
+               {
+                       comment:     "form-encoded expression filter in POST body",
+                       method:      "POST",
+                       path:        "/arvados/v1/collections",
+                       body:        "filters=[\"(foo<bar)\"]&_method=GET",
+                       header:      http.Header{"Content-Type": {"application/x-www-form-urlencoded"}},
+                       shouldCall:  "CollectionList",
+                       withOptions: arvados.ListOptions{Limit: -1, Filters: []arvados.Filter{{"(foo<bar)", "=", true}}},
+               },
+               {
+                       comment:     "json-encoded expression filter in POST body",
+                       method:      "POST",
+                       path:        "/arvados/v1/collections?_method=GET",
+                       body:        `{"filters":["(foo<bar)",["bar","=","baz"]],"limit":2}`,
+                       header:      http.Header{"Content-Type": {"application/json"}},
+                       shouldCall:  "CollectionList",
+                       withOptions: arvados.ListOptions{Limit: 2, Filters: []arvados.Filter{{"(foo<bar)", "=", true}, {"bar", "=", "baz"}}},
+               },
                {
                        method:       "PATCH",
                        path:         "/arvados/v1/collections",
@@ -139,21 +166,23 @@ func (s *RouterSuite) TestOptions(c *check.C) {
                // Reset calls captured in previous trial
                s.stub = arvadostest.APIStub{}
 
-               c.Logf("trial: %#v", trial)
+               c.Logf("trial: %+v", trial)
+               comment := check.Commentf("trial comment: %s", trial.comment)
+
                _, rr, _ := doRequest(c, s.rtr, token, trial.method, trial.path, trial.header, bytes.NewBufferString(trial.body))
                if trial.shouldStatus == 0 {
-                       c.Check(rr.Code, check.Equals, http.StatusOK)
+                       c.Check(rr.Code, check.Equals, http.StatusOK, comment)
                } else {
-                       c.Check(rr.Code, check.Equals, trial.shouldStatus)
+                       c.Check(rr.Code, check.Equals, trial.shouldStatus, comment)
                }
                calls := s.stub.Calls(nil)
                if trial.shouldCall == "" {
-                       c.Check(calls, check.HasLen, 0)
+                       c.Check(calls, check.HasLen, 0, comment)
                } else if len(calls) != 1 {
-                       c.Check(calls, check.HasLen, 1)
+                       c.Check(calls, check.HasLen, 1, comment)
                } else {
-                       c.Check(calls[0].Method, isMethodNamed, trial.shouldCall)
-                       c.Check(calls[0].Options, check.DeepEquals, trial.withOptions)
+                       c.Check(calls[0].Method, isMethodNamed, trial.shouldCall, comment)
+                       c.Check(calls[0].Options, check.DeepEquals, trial.withOptions, comment)
                }
        }
 }
index a5cc7d3b904f271696defd341f48a99dd47b92f8..7f319b41210a6839447056f81ebfbdb52123f0e9 100644 (file)
@@ -37,28 +37,37 @@ func (f *Filter) MarshalJSON() ([]byte, error) {
 
 // UnmarshalJSON decodes a JSON array to a Filter.
 func (f *Filter) UnmarshalJSON(data []byte) error {
-       var elements []interface{}
-       err := json.Unmarshal(data, &elements)
+       var decoded interface{}
+       err := json.Unmarshal(data, &decoded)
        if err != nil {
                return err
        }
-       if len(elements) != 3 {
-               return fmt.Errorf("invalid filter %q: must have 3 elements", data)
-       }
-       attr, ok := elements[0].(string)
-       if !ok {
-               return fmt.Errorf("invalid filter attr %q", elements[0])
-       }
-       op, ok := elements[1].(string)
-       if !ok {
-               return fmt.Errorf("invalid filter operator %q", elements[1])
-       }
-       operand := elements[2]
-       switch operand.(type) {
-       case string, float64, []interface{}, nil, bool:
+       switch decoded := decoded.(type) {
+       case string:
+               // Accept "(foo < bar)" as a more obvious way to spell
+               // ["(foo < bar)","=",true]
+               *f = Filter{decoded, "=", true}
+       case []interface{}:
+               if len(decoded) != 3 {
+                       return fmt.Errorf("invalid filter %q: must have 3 decoded", data)
+               }
+               attr, ok := decoded[0].(string)
+               if !ok {
+                       return fmt.Errorf("invalid filter attr %q", decoded[0])
+               }
+               op, ok := decoded[1].(string)
+               if !ok {
+                       return fmt.Errorf("invalid filter operator %q", decoded[1])
+               }
+               operand := decoded[2]
+               switch operand.(type) {
+               case string, float64, []interface{}, nil, bool:
+               default:
+                       return fmt.Errorf("invalid filter operand %q", decoded[2])
+               }
+               *f = Filter{attr, op, operand}
        default:
-               return fmt.Errorf("invalid filter operand %q", elements[2])
+               return fmt.Errorf("invalid filter: json decoded as %T instead of array or string", decoded)
        }
-       *f = Filter{attr, op, operand}
        return nil
 }
index b36e82c918298fa624cb290b12889cb8da2734c0..e4d89d62ad3a48b8e57c8bc64e016fb07f7f464d 100644 (file)
@@ -5,69 +5,59 @@
 package arvados
 
 import (
-       "bytes"
        "encoding/json"
-       "testing"
        "time"
+
+       check "gopkg.in/check.v1"
 )
 
-func TestMarshalFiltersWithNanoseconds(t *testing.T) {
+var _ = check.Suite(&filterEncodingSuite{})
+
+type filterEncodingSuite struct{}
+
+func (s *filterEncodingSuite) TestMarshalNanoseconds(c *check.C) {
        t0 := time.Now()
        t0str := t0.Format(time.RFC3339Nano)
        buf, err := json.Marshal([]Filter{
                {Attr: "modified_at", Operator: "=", Operand: t0}})
-       if err != nil {
-               t.Fatal(err)
-       }
-       if expect := []byte(`[["modified_at","=","` + t0str + `"]]`); 0 != bytes.Compare(buf, expect) {
-               t.Errorf("Encoded as %q, expected %q", buf, expect)
-       }
+       c.Assert(err, check.IsNil)
+       c.Check(string(buf), check.Equals, `[["modified_at","=","`+t0str+`"]]`)
 }
 
-func TestMarshalFiltersWithNil(t *testing.T) {
+func (s *filterEncodingSuite) TestMarshalNil(c *check.C) {
        buf, err := json.Marshal([]Filter{
                {Attr: "modified_at", Operator: "=", Operand: nil}})
-       if err != nil {
-               t.Fatal(err)
-       }
-       if expect := []byte(`[["modified_at","=",null]]`); 0 != bytes.Compare(buf, expect) {
-               t.Errorf("Encoded as %q, expected %q", buf, expect)
-       }
+       c.Assert(err, check.IsNil)
+       c.Check(string(buf), check.Equals, `[["modified_at","=",null]]`)
 }
 
-func TestUnmarshalFiltersWithNil(t *testing.T) {
+func (s *filterEncodingSuite) TestUnmarshalNil(c *check.C) {
        buf := []byte(`["modified_at","=",null]`)
-       f := &Filter{}
+       var f Filter
        err := f.UnmarshalJSON(buf)
-       if err != nil {
-               t.Fatal(err)
-       }
-       expect := Filter{Attr: "modified_at", Operator: "=", Operand: nil}
-       if f.Attr != expect.Attr || f.Operator != expect.Operator || f.Operand != expect.Operand {
-               t.Errorf("Decoded as %q, expected %q", f, expect)
-       }
+       c.Assert(err, check.IsNil)
+       c.Check(f, check.DeepEquals, Filter{Attr: "modified_at", Operator: "=", Operand: nil})
 }
 
-func TestMarshalFiltersWithBoolean(t *testing.T) {
+func (s *filterEncodingSuite) TestMarshalBoolean(c *check.C) {
        buf, err := json.Marshal([]Filter{
                {Attr: "is_active", Operator: "=", Operand: true}})
-       if err != nil {
-               t.Fatal(err)
-       }
-       if expect := []byte(`[["is_active","=",true]]`); 0 != bytes.Compare(buf, expect) {
-               t.Errorf("Encoded as %q, expected %q", buf, expect)
-       }
+       c.Assert(err, check.IsNil)
+       c.Check(string(buf), check.Equals, `[["is_active","=",true]]`)
 }
 
-func TestUnmarshalFiltersWithBoolean(t *testing.T) {
+func (s *filterEncodingSuite) TestUnmarshalBoolean(c *check.C) {
        buf := []byte(`["is_active","=",true]`)
-       f := &Filter{}
+       var f Filter
+       err := f.UnmarshalJSON(buf)
+       c.Assert(err, check.IsNil)
+       c.Check(f, check.DeepEquals, Filter{Attr: "is_active", Operator: "=", Operand: true})
+}
+
+func (s *filterEncodingSuite) TestUnmarshalBooleanExpression(c *check.C) {
+       buf := []byte(`"(foo < bar)"`)
+       var f Filter
        err := f.UnmarshalJSON(buf)
-       if err != nil {
-               t.Fatal(err)
-       }
-       expect := Filter{Attr: "is_active", Operator: "=", Operand: true}
-       if f.Attr != expect.Attr || f.Operator != expect.Operator || f.Operand != expect.Operand {
-               t.Errorf("Decoded as %q, expected %q", f, expect)
-       }
+       c.Assert(err, check.IsNil)
+       c.Check(f, check.DeepEquals, Filter{Attr: "(foo < bar)", Operator: "=", Operand: true})
 }
index 409e48a6f090a3b348cd5d551bf35a91427e42a9..2f5b67074a9bdf5b24d3689333d17ee6e98e0745 100644 (file)
@@ -142,6 +142,23 @@ module RecordFilters
 
           cond_out << "jsonb_exists(#{attr_table_name}.#{attr}, ?)"
           param_out << operand
+        elsif expr = /^ *\( *(\w+) *(<=?|>=?|=) *(\w+) *\) *$/.match(attr)
+          if operator != '=' || ![true,"true"].index(operand)
+            raise ArgumentError.new("Invalid expression filter '#{attr}': subsequent elements must be [\"=\", true]")
+          end
+          operator = expr[2]
+          attr1, attr2 = expr[1], expr[3]
+          allowed = attr_model_class.searchable_columns(operator)
+          [attr1, attr2].each do |tok|
+            if !allowed.index(tok)
+              raise ArgumentError.new("Invalid attribute in expression: '#{tok}'")
+            end
+            col = attr_model_class.columns.select { |c| c.name == tok }.first
+            if col.type != :integer
+              raise ArgumentError.new("Non-numeric attribute in expression: '#{tok}'")
+            end
+          end
+          cond_out << "#{attr1} #{operator} #{attr2}"
         else
           if !attr_model_class.searchable_columns(operator).index(attr) &&
              !(col.andand.type == :jsonb && ['contains', '=', '<>', '!='].index(operator))
index 6c923ff38d96b8b64e4b8fa8ad83b13b3b29eefc..eac393104cf9b5a63355d131d9b09dbd8ca1a5a1 100644 (file)
@@ -1404,6 +1404,50 @@ EOS
     assert_equal col.version, json_response['version'], 'Trashing a collection should not create a new version'
   end
 
+  [['<', :<],
+   ['<=', :<=],
+   ['>', :>],
+   ['>=', :>=],
+   ['=', :==]].each do |op, rubyop|
+    test "filter collections by replication_desired #{op} replication_confirmed" do
+      authorize_with(:active)
+      get :index, params: {
+            filters: [["(replication_desired #{op} replication_confirmed)", "=", true]],
+          }
+      assert_response :success
+      json_response["items"].each do |c|
+        assert_operator(c["replication_desired"], rubyop, c["replication_confirmed"])
+      end
+    end
+  end
+
+  ["(replication_desired < bogus)",
+   "replication_desired < replication_confirmed",
+   "(replication_desired < replication_confirmed",
+   "(replication_desired ! replication_confirmed)",
+   "(replication_desired <)",
+   "(replication_desired < manifest_text)",
+   "(manifest_text < manifest_text)", # currently only numeric attrs are supported
+   "(replication_desired < 2)", # currently only attrs are supported, not literals
+   "(1 < 2)",
+  ].each do |expr|
+    test "invalid filter expression #{expr}" do
+      authorize_with(:active)
+      get :index, params: {
+            filters: [[expr, "=", true]],
+          }
+      assert_response 422
+    end
+  end
+
+  test "invalid op/arg with filter expression" do
+    authorize_with(:active)
+    get :index, params: {
+          filters: [["replication_desired < replication_confirmed", "!=", false]],
+        }
+    assert_response 422
+  end
+
   ["storage_classes_desired", "storage_classes_confirmed"].each do |attr|
     test "filter collections by #{attr}" do
       authorize_with(:active)