[Catalog] Add pre- and post-optimizations

This commit introduces pre- and post-optimizations as new `Catalog` Components.

Pre-optimizations are implemented as callback functions that accept the QueryGraph as an argument, prior to its optimization.

Similarly, post-optimizations are defined as callback functions that receive the resulting `Producer` after the optimization process.

Notably, the commit also removes the `plan->minimize_schema()` operation at the end of optimization, registering it as a post-optimization step.

Author: JonasLauermann

include/mutable/catalog/Catalog.hpp

@@ -327,6 +327,12 @@ struct M_EXPORT Catalog
     using TableFactoryDecoratorCallback = std::function<std::unique_ptr<TableFactory>(std::unique_ptr<TableFactory>)>;
     ComponentSet<TableFactoryDecoratorCallback> table_properties_; // stores callback functions that decorate a table with the given decorator
 
+    using PreOptimizationCallback = std::function<void(const QueryGraph &)>;
+    ComponentSet<PreOptimizationCallback> pre_optimizations_;
+
+    using PostOptimizationCallback = std::function<std::unique_ptr<Producer>(std::unique_ptr<Producer>)>;
+    ComponentSet<PostOptimizationCallback> post_optimizations_;
+
     public:
     /*===== Stores ===================================================================================================*/
     /** Registers a new `Store` with the given `name`. */
@@ -592,6 +598,42 @@ struct M_EXPORT Catalog
     auto table_properties_end()    const { return table_properties_.end(); }
     auto table_properties_cbegin() const { return table_properties_begin(); }
     auto table_properties_cend()   const { return table_properties_end(); }
+
+    /*===== Pre-Optimizations ========================================================================================*/
+    /** Registers a new pre-optimization with the given `name`. */
+    void register_pre_optimization(const char *name, PreOptimizationCallback optimization, const char *description = nullptr)
+    {
+        pre_optimizations_.add(
+                pool(name),
+                Component<PreOptimizationCallback>(description, std::make_unique<PreOptimizationCallback>(std::move(optimization)))
+        );
+    }
+
+    auto pre_optimizations()              { return range(pre_optimizations_.begin(), pre_optimizations_.end()); }
+    auto pre_optimizations_begin()        { return pre_optimizations_.begin(); }
+    auto pre_optimizations_end()          { return pre_optimizations_.end(); }
+    auto pre_optimizations_begin()  const { return pre_optimizations_.begin(); }
+    auto pre_optimizations_end()    const { return pre_optimizations_.end(); }
+    auto pre_optimizations_cbegin() const { return pre_optimizations_begin(); }
+    auto pre_optimizations_cend()   const { return pre_optimizations_end(); }
+
+    /*===== Post-Optimizations ========================================================================================*/
+    /** Registers a new post-optimization with the given `name`. */
+    void register_post_optimization(const char *name, PostOptimizationCallback optimization, const char *description = nullptr)
+    {
+        post_optimizations_.add(
+                pool(name),
+                Component<PostOptimizationCallback>(description, std::make_unique<PostOptimizationCallback>(std::move(optimization)))
+        );
+    }
+
+    auto post_optimizations()              { return range(post_optimizations_.begin(), post_optimizations_.end()); }
+    auto post_optimizations_begin()        { return post_optimizations_.begin(); }
+    auto post_optimizations_end()          { return post_optimizations_.end(); }
+    auto post_optimizations_begin()  const { return post_optimizations_.begin(); }
+    auto post_optimizations_end()    const { return post_optimizations_.end(); }
+    auto post_optimizations_cbegin() const { return post_optimizations_begin(); }
+    auto post_optimizations_cend()   const { return post_optimizations_end(); }
 };
 
 }

src/IR/Operator.cpp

@@ -543,3 +543,14 @@ void Operator::minimize_schema()
     SchemaMinimizer M;
     M(*this);
 }
+
+__attribute__((constructor(202)))
+void register_post_optimization()
+{
+    Catalog &C = Catalog::Get();
+
+    C.register_post_optimization("minimize schema", [](std::unique_ptr<Producer> plan) {
+        plan->minimize_schema();
+        return plan;
+    }, "minimizes the schema of an operator tree");
+}

src/IR/Optimizer.cpp

@@ -165,6 +165,10 @@ Optimizer::optimize_with_plantable(const QueryGraph &G) const
         return { std::make_unique<ProjectionOperator>(G.projections()), std::move(plan_table) };
     }
 
+    /*----- Perform pre-optimizations on the QueryGraph. -------------------------------------------------------------*/
+    for (auto &pre_opt : C.pre_optimizations())
+        (*pre_opt.second).operator()(G);
+
     /*----- Initialize plan table and compute plans for data sources. ------------------------------------------------*/
     Producer **source_plans = new Producer*[num_sources];
     for (auto &ds : G.sources()) {
@@ -356,7 +360,10 @@ Optimizer::optimize_with_plantable(const QueryGraph &G) const
         plan = std::move(projection);
     }
 
-    plan->minimize_schema();
+    /*----- Perform post-optimizations on the Operator tree. ---------------------------------------------------------*/
+    for (auto &post_opt : C.post_optimizations())
+        plan = (*post_opt.second).operator()(std::move(plan)); // TODO add plan_table to post-optimizations
+
     delete[] source_plans;
     return { std::move(plan), std::move(plan_table) };
 }