fix: five fixes from the python-parity audit (two source-breaking)

CHANGELOG.md

@@ -9,9 +9,30 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- **`generative-deepseek`** — New `GenerativeConfigFactory.Deepseek(...)` collection config and `GenerativeProviderFactory.Deepseek(...)` runtime provider for the `generative-deepseek` module. Requires Weaviate ≥ 1.36.19.
+- **Incremental Backups** — `BackupCreateRequest.IncrementalBaseBackupId` names an existing backup to build on, so files unchanged since that backup are not copied again; `Backup.IncrementalBaseBackupId` surfaces it on `List()` and `GetStatus()`. `Create` throws `WeaviateVersionMismatchException` below 1.37.0, the documented feature floor. The server has accepted the field on create since 1.34.18, but only returns it on read from 1.37.6.
+- **`apiEndpoint` on `Multi2VecGoogle`** — Both `VectorizerFactory.Multi2VecGoogle(...)` overloads take an optional `apiEndpoint`, which selects the Gemini API (`generativelanguage.googleapis.com`) instead of Vertex AI. Both `Multi2VecGoogleGemini(...)` overloads also gained an optional `dimensions`.
+
+### Fixed
+
+- **Backup `Size` Dropped on List** — `BackupClient.List()` discarded the size and incremental base id it had already parsed from the response, so `Backup.Size` was null for every listed backup. Both fields are now mapped.
+- **Aggregate Zeros for Absent Values** — Int, Number, Boolean, Text and Date results, `Count` included, reported `0`, `0.0` or `false` where the server had sent no value at all. Every scalar is now presence-checked and left null when unset, so an empty aggregation is distinguishable from a genuine zero.
+- **`Multi2VecGoogleGemini` Emitted a Non-Existent Module** — The vectorizer declared the module name `multi2vec-google-gemini`, which no Weaviate server provides, so a collection configured with it could never be created. `VectorizerFactory.Multi2VecGoogleGemini(...)` now emits `multi2vec-google` with the Gemini API endpoint.
+
+### Changed
+
+- **`Multi2VecGoogleGemini` is now an `[Obsolete]` shim over `Multi2VecGoogle`** — This breaks at runtime, not only at compile time: the factory now returns a `Multi2VecGoogle`, so `is`/`as`/pattern matches on `Multi2VecGoogleGemini` silently stop matching, and a `switch` with arms for both types no longer compiles (`CS8120: The switch case is unreachable`). Bind the result as `Multi2VecGoogle` or `VectorizerConfig`.
+- **`Multi2VecGoogle.ProjectId` and `.Location` are no longer `required`** — Both getters changed from `string!` to `string?` so the Gemini endpoint, which is scoped to neither a project nor a region, can omit them. Callers with nullable reference types enabled may see CS8600/CS8601.
+- **`Aggregate.Boolean` counts are nullable** — `PercentageTrue`, `PercentageFalse`, `TotalTrue` and `TotalFalse` changed from `double`/`long` to `double?`/`long?`.
+- **`BackupCreateRequest` gained a trailing optional parameter** — Its primary constructor and `Deconstruct` went from six parameters to seven, so six-element positional deconstruction (`var (id, backend, inc, exc, cpu, comp) = req;`) no longer compiles; deconstruct seven elements instead.
+- **Binary compatibility** — The signature changes above are source-compatible for callers using named arguments, but not binary-compatible: assemblies compiled against 1.1.0 throw `MissingMethodException` until recompiled.
+
 ### Removed
 
 - **`ReplicationAsyncConfig.MaxWorkers` and `ReplicationAsyncConfig.AliveNodesCheckingFrequency`** — Both fields have been no-ops on the server since Weaviate 1.37.3 and are now removed from the user-facing model, the OpenAPI spec, and the generated DTO. Existing code that sets these properties will not compile after upgrading; no behavioral change results from the removal.
+- **`Multi2VecGoogleGemini.Model`** — Removed along with the rest of the type's own members; the base `Multi2VecGoogle` declares the same setting as `ModelId`. Migrate `.Model` → `.ModelId`.
 
 ---
 
@@ -67,7 +88,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 #### Vectorizers
 
-- **Audio Field Support** ([#302](https://github.com/weaviate/weaviate-csharp-client/pull/302)): `Multi2VecGoogle` and `Multi2VecGoogleGemini` vectorizers now support audio field configurations with configurable per-field weights.
+- **Audio Field Support** ([#302](https://github.com/weaviate/weaviate-csharp-client/pull/302)): `Multi2VecGoogle` and `Multi2VecGoogleGemini` vectorizers now support audio field configurations with configurable per-field weights. (`Multi2VecGoogleGemini` is deprecated in favour of `Multi2VecGoogle` — see Unreleased.)
 
 #### API Ergonomics
 
@@ -105,7 +126,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Weaviate 1.36 support: HFresh vector index, async replication config, property index deletion
 - Critical fix: gRPC vector serialization no longer doubles dimensions for non-`float[]` vectors
 - Opt-in structured logging via `ILoggerFactory`
-- New vectorizers: `Multi2VecGoogleGemini` and `Multi2MultivecWeaviate`
+- New vectorizers: `Multi2VecGoogleGemini` (never functional — see Unreleased) and `Multi2MultivecWeaviate`
 
 ### Added
 
@@ -119,7 +140,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 #### Vectorizers
 
-- **Multi2VecGoogleGemini** ([#297](https://github.com/weaviate/weaviate-csharp-client/pull/297)): New vectorizer calling the Google Gemini API directly. Supports image, text, and video field weighting. No project ID or location required (unlike the Vertex AI variant). Defaults to `generativelanguage.googleapis.com`.
+- **Multi2VecGoogleGemini** ([#297](https://github.com/weaviate/weaviate-csharp-client/pull/297)): New vectorizer calling the Google Gemini API directly. Supports image, text, and video field weighting. No project ID or location required (unlike the Vertex AI variant). Defaults to `generativelanguage.googleapis.com`. **Correction:** this vectorizer declared the module name `multi2vec-google-gemini`, which no Weaviate server provides, so it could never create a collection; fixed in Unreleased.
 - **Multi2MultivecWeaviate** ([#291](https://github.com/weaviate/weaviate-csharp-client/pull/291)): Support for the `multi2multivec-weaviate` vectorizer, which produces multi-vector embeddings using Weaviate's built-in model.
 - **Cohere Reranker `BaseURL`** ([#287](https://github.com/weaviate/weaviate-csharp-client/pull/287)): Added `BaseURL` property to `RerankerCohereConfig` and a corresponding parameter to `RerankerConfigFactory.Cohere()`, enabling self-hosted or regional Cohere endpoints.
 

ci/docker-compose.yml

@@ -22,7 +22,7 @@ services:
       QUERY_DEFAULTS_LIMIT: 25
       AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED: 'true'
       PERSISTENCE_DATA_PATH: '/var/lib/weaviate'
-      ENABLE_MODULES: text2vec-transformers,text2vec-cohere,backup-filesystem,generative-dummy,generative-anyscale,reranker-dummy,reranker-cohere,text2vec-ollama,generative-ollama
+      ENABLE_MODULES: text2vec-transformers,text2vec-cohere,backup-filesystem,generative-dummy,generative-anyscale,reranker-dummy,reranker-cohere,text2vec-ollama,generative-ollama,multi2vec-google,generative-deepseek
       BACKUP_FILESYSTEM_PATH: "/tmp/backups"
       EXPORT_ENABLED: 'true'
       EXPORT_DEFAULT_PATH: "/tmp/exports"

docs/AGGREGATE_RESULT_ACCESSORS.md

@@ -107,11 +107,11 @@ if (quantityAgg != null)
     Console.WriteLine($"Total quantity: {quantityAgg.Sum}");
 }
 
-// Access boolean aggregation
+// Access boolean aggregation (its members are null when the server did not send them)
 var inStockAgg = result.Boolean("inStock");
-if (inStockAgg != null)
+if (inStockAgg?.PercentageTrue is { } inStockPercentage)
 {
-    Console.WriteLine($"In stock: {inStockAgg.PercentageTrue:P0}");
+    Console.WriteLine($"In stock: {inStockPercentage:P0}");
 }
 
 // Access date aggregation
@@ -157,9 +157,13 @@ if (result.TryGetText("category", out var category))
 }
 
 // Generic TryGet
-if (result.TryGet<Aggregate.Boolean>("inStock", out var stockAgg))
+if (
+    result.TryGet<Aggregate.Boolean>("inStock", out var stockAgg)
+    && stockAgg.TotalTrue is { } inStock
+    && stockAgg.TotalFalse is { } outOfStock
+)
 {
-    Console.WriteLine($"In stock: {stockAgg.TotalTrue}, Out of stock: {stockAgg.TotalFalse}");
+    Console.WriteLine($"In stock: {inStock}, Out of stock: {outOfStock}");
 }
 ```
 
@@ -196,7 +200,11 @@ result.Match("field",
     text: t => Console.WriteLine($"Text: {t.Count} occurrences"),
     integer: i => Console.WriteLine($"Integer: sum={i.Sum}"),
     number: n => Console.WriteLine($"Number: mean={n.Mean}"),
-    boolean: b => Console.WriteLine($"Boolean: {b.PercentageTrue:P0} true"),
+    boolean: b =>
+    {
+        if (b.PercentageTrue is { } percentage)
+            Console.WriteLine($"Boolean: {percentage:P0} true");
+    },
     date: d => Console.WriteLine($"Date range: {d.Minimum} to {d.Maximum}")
 );
 
@@ -205,7 +213,10 @@ var description = result.Match<string>("field",
     text: t => $"Text with {t.TopOccurrences.Count} unique values",
     integer: i => $"Integer ranging from {i.Minimum} to {i.Maximum}",
     number: n => $"Number with mean {n.Mean:F2}",
-    boolean: b => $"Boolean: {b.PercentageTrue:P0} true",
+    boolean: b =>
+        b.PercentageTrue is { } percentage
+            ? $"Boolean: {percentage:P0} true"
+            : "Boolean: percentage not returned",
     date: d => $"Dates from {d.Minimum:d} to {d.Maximum:d}"
 );
 
@@ -223,7 +234,10 @@ foreach (var (name, _) in result.Properties)
         text: t => $"{name}: {t.Count} items, top: {t.TopOccurrences.FirstOrDefault()?.Value}",
         integer: i => $"{name}: range [{i.Minimum}, {i.Maximum}], sum: {i.Sum}",
         number: n => $"{name}: range [{n.Minimum:F2}, {n.Maximum:F2}], mean: {n.Mean:F2}",
-        boolean: b => $"{name}: {b.TotalTrue} true, {b.TotalFalse} false",
+        boolean: b =>
+            b.TotalTrue is { } trueCount && b.TotalFalse is { } falseCount
+                ? $"{name}: {trueCount} true, {falseCount} false"
+                : $"{name}: boolean counts not returned",
         date: d => $"{name}: {d.Minimum:d} to {d.Maximum:d}"
     );
 
@@ -312,10 +326,13 @@ For boolean properties:
 | Property | Type | Description |
 |----------|------|-------------|
 | `Count` | `long?` | Number of values |
-| `TotalTrue` | `long` | Count of true values |
-| `TotalFalse` | `long` | Count of false values |
-| `PercentageTrue` | `double` | Percentage of true values (0-1) |
-| `PercentageFalse` | `double` | Percentage of false values (0-1) |
+| `TotalTrue` | `long?` | Count of true values |
+| `TotalFalse` | `long?` | Count of false values |
+| `PercentageTrue` | `double?` | Percentage of true values (0-1) |
+| `PercentageFalse` | `double?` | Percentage of false values (0-1) |
+
+Every member is nullable: null means the server did not send that value, which is not the same as
+zero. Check before formatting — interpolating a null renders as an empty string.
 
 ### Aggregate.Date
 

src/Weaviate.Client.Tests/Integration/TestCollectionAggregate.cs

@@ -903,4 +903,82 @@ await collectionClient.Data.Insert(
                 break;
         }
     }
+
+    /// <summary>
+    /// Two ways a scalar goes missing, on one collection: a filter that matches nothing, and a
+    /// sub-metric the caller never asked for. Both leave the optional proto field unset, and the
+    /// client must report null rather than the zero of the field's type — a caller cannot tell a
+    /// returned 0 apart from a genuine aggregate of zeros.
+    /// </summary>
+    [Fact]
+    public async Task Test_OverAll_Absent_Metrics_Are_Null_Not_Zero()
+    {
+        var collectionClient = await CollectionFactory(
+            properties: new[]
+            {
+                Property.Text("text"),
+                Property.Int("int"),
+                Property.Number("float"),
+            }
+        );
+
+        await collectionClient.Data.Insert(
+            new
+            {
+                text = "one",
+                @int = 7,
+                @float = 7.5,
+            },
+            cancellationToken: TestContext.Current.CancellationToken
+        );
+
+        // Nothing matches: the metrics were requested, but there is nothing to compute them over.
+        var noMatches = await collectionClient.Aggregate.OverAll(
+            filters: Filter.Property("text").IsEqual("no-such-value"),
+            returnMetrics:
+            [
+                Metrics.ForProperty("int").Integer(maximum: true, mean: true, sum: true),
+                Metrics.ForProperty("float").Number(maximum: true, mean: true, sum: true),
+            ],
+            cancellationToken: TestContext.Current.CancellationToken
+        );
+
+        Assert.Equal(0, noMatches.TotalCount);
+
+        var emptyInteger = Assert.IsType<Aggregate.Integer>(noMatches.Properties["int"]);
+        Assert.Null(emptyInteger.Maximum);
+        Assert.Null(emptyInteger.Mean);
+        Assert.Null(emptyInteger.Sum);
+
+        var emptyNumber = Assert.IsType<Aggregate.Number>(noMatches.Properties["float"]);
+        Assert.Null(emptyNumber.Maximum);
+        Assert.Null(emptyNumber.Mean);
+        Assert.Null(emptyNumber.Sum);
+
+        // Objects do match here, so the nulls come from the metric selection alone.
+        var unrequested = await collectionClient.Aggregate.OverAll(
+            returnMetrics:
+            [
+                Metrics.ForProperty("int").Integer(maximum: true),
+                Metrics.ForProperty("float").Number(mean: true),
+            ],
+            cancellationToken: TestContext.Current.CancellationToken
+        );
+
+        Assert.Equal(1, unrequested.TotalCount);
+
+        var integer = Assert.IsType<Aggregate.Integer>(unrequested.Properties["int"]);
+        Assert.Equal(7, integer.Maximum);
+        Assert.Null(integer.Mean);
+        Assert.Null(integer.Median);
+        Assert.Null(integer.Minimum);
+        Assert.Null(integer.Mode);
+        Assert.Null(integer.Sum);
+
+        var number = Assert.IsType<Aggregate.Number>(unrequested.Properties["float"]);
+        Assert.Equal(7.5, number.Mean);
+        Assert.Null(number.Maximum);
+        Assert.Null(number.Minimum);
+        Assert.Null(number.Sum);
+    }
 }

src/Weaviate.Client.Tests/Integration/TestCollections.cs

@@ -193,6 +193,55 @@ public async Task Collection_Creates_And_Retrieves_Generative_Config()
         Assert.IsType<GenerativeConfig.Custom>(collection.GenerativeConfig);
     }
 
+    /// <summary>
+    /// Tests that a generative-deepseek collection round-trips through a real server. The module
+    /// validates its own settings on create (temperature, both penalties and topP all have
+    /// ranges, and baseURL is parsed), so a misspelt key or a wrong type shows up here rather
+    /// than silently falling back to a default.
+    /// </summary>
+    /// <remarks>
+    /// Integral floats only; the server drops fractional floats for this module on named-vector
+    /// classes. See PR #368. Fractional values are covered by the unit test instead.
+    /// </remarks>
+    [Fact]
+    public async Task Collection_Creates_And_Retrieves_GenerativeDeepseek_Config()
+    {
+        RequireModule("generative-deepseek");
+
+        // Arrange
+        var collectionClient = await CollectionFactory(
+            properties: [Property.Text("Name")],
+            generativeConfig: Configure.Generative.Deepseek(
+                model: "deepseek-chat",
+                temperature: 1,
+                maxTokens: 2048,
+                frequencyPenalty: 0,
+                presencePenalty: 0,
+                topP: 1,
+                baseURL: "https://api.deepseek.com",
+                stop: ["\n\n"]
+            )
+        );
+
+        // Act
+        var collection = await _weaviate
+            .Collections.Use(collectionClient.Name)
+            .Config.Get(TestContext.Current.CancellationToken);
+
+        // Assert
+        Assert.NotNull(collection);
+        var deepseek = Assert.IsType<GenerativeConfig.Deepseek>(collection.GenerativeConfig);
+        Assert.Equal("deepseek-chat", deepseek.Model);
+        Assert.Equal(1, deepseek.Temperature);
+        Assert.Equal(2048, deepseek.MaxTokens);
+        Assert.Equal(0, deepseek.FrequencyPenalty);
+        Assert.Equal(0, deepseek.PresencePenalty);
+        Assert.Equal(1, deepseek.TopP);
+        Assert.Equal("https://api.deepseek.com", deepseek.BaseURL);
+        Assert.NotNull(deepseek.Stop);
+        Assert.Equal(["\n\n"], deepseek.Stop);
+    }
+
     /// <summary>
     /// Tests that test collections export
     /// </summary>

src/Weaviate.Client.Tests/Integration/TestVectorizers.cs

@@ -0,0 +1,95 @@
+namespace Weaviate.Client.Tests.Integration;
+
+using Weaviate.Client.Models;
+
+/// <summary>
+/// The vectorizer module tests class. Covers vectorizer configurations end to end, i.e. that the
+/// module name and settings the client emits are ones a real server accepts.
+/// </summary>
+/// <seealso cref="IntegrationTests"/>
+public class TestVectorizers : IntegrationTests
+{
+    /// <summary>
+    /// Tests that the Google Gemini multimodal factory produces a collection the server accepts.
+    /// </summary>
+    [Fact]
+    public async Task Test_Multi2VecGoogleGemini_Creates_Collection()
+    {
+        RequireModule("multi2vec-google");
+        // The module is on every lane, but its Gemini path (apiEndpoint) only lands in 1.34.20;
+        // older builds still demand the Vertex projectId/location this factory omits.
+        RequireVersion("1.34.20");
+
+        var collection = await CollectionFactory(
+            name: "TestMulti2VecGoogleGemini",
+            properties: [Property.Text("text"), Property.Blob("image")],
+            vectorConfig: Configure.Vector(
+                "default",
+                v =>
+                    v.Multi2VecGoogleGemini(
+                        imageFields: ["image"],
+                        textFields: ["text"],
+                        dimensions: 512
+                    )
+            )
+        );
+
+        var config = await collection.Config.Get(
+            cancellationToken: TestContext.Current.CancellationToken
+        );
+
+        Assert.NotNull(config);
+        var vectorizer = config.VectorConfig["default"].Vectorizer;
+        var google = Assert.IsType<Vectorizer.Multi2VecGoogle>(vectorizer);
+
+        Assert.Equal("multi2vec-palm", google.Identifier);
+        Assert.Equal("generativelanguage.googleapis.com", google.ApiEndpoint);
+        Assert.Equal(512, google.Dimensions);
+        // The factory has no vectorizeClassName: nothing is sent, nothing stored or defaulted back.
+        Assert.Null(google.VectorizeCollectionName);
+        Assert.NotNull(google.ImageFields);
+        Assert.Equal(["image"], google.ImageFields);
+        Assert.NotNull(google.TextFields);
+        Assert.Equal(["text"], google.TextFields);
+        // Vertex-only settings: the Gemini API has neither, and the server must not echo them.
+        Assert.Null(google.ProjectId);
+        Assert.Null(google.Location);
+    }
+
+    /// <summary>
+    /// Tests that the Vertex AI factory creates a collection with project id and location intact.
+    /// </summary>
+    [Fact]
+    public async Task Test_Multi2VecGoogle_Vertex_Creates_Collection()
+    {
+        RequireModule("multi2vec-google");
+
+        var collection = await CollectionFactory(
+            name: "TestMulti2VecGoogleVertex",
+            properties: [Property.Text("text"), Property.Blob("image")],
+            vectorConfig: Configure.Vector(
+                "default",
+                v =>
+                    v.Multi2VecGoogle(
+                        projectId: "my-project",
+                        location: "us-central1",
+                        imageFields: ["image"],
+                        textFields: ["text"]
+                    )
+            )
+        );
+
+        var config = await collection.Config.Get(
+            cancellationToken: TestContext.Current.CancellationToken
+        );
+
+        Assert.NotNull(config);
+        var vectorizer = config.VectorConfig["default"].Vectorizer;
+        var google = Assert.IsType<Vectorizer.Multi2VecGoogle>(vectorizer);
+
+        Assert.Equal("multi2vec-palm", google.Identifier);
+        Assert.Equal("my-project", google.ProjectId);
+        Assert.Equal("us-central1", google.Location);
+        Assert.Null(google.ApiEndpoint);
+    }
+}