PG UX polish: dialect-aware AI prompts, button visibility, label cleanup

Four issues surfaced in actual PG dashboard use:

1. EXPLAIN false-positive on Rails query annotations
   SQL like `SELECT COUNT(*) FROM "taggings" /*action='table_sizes',
   ...*/` was rejected as "truncated" because the closing `*/` ends
   in `/` and tripped the trailing-operator check. looks_complete?
   now strips block comments before inspecting the trailing token.

2. AI assistant generates MySQL backticks on a PG database
   `Suggestion`, `RewriteQuery`, `Optimization`, `IndexAdvisor`,
   `IndexPlanner`, `MigrationRisk`, `SchemaReview`, `WorkloadDigest`,
   and `PatternGrouper` all hardcoded "MySQL" in their system
   prompts; Suggestion also told the model to "use backticks for
   table and column names." On PG this produced unparseable SQL.

   New Ai::DialectHints module exposes name_for(conn) and
   identifier_quoting_rule(conn). Every prompt now interpolates the
   dialect name; Suggestion's quoting rule swaps to double quotes
   on PG. IndexPlanner's prompt also specifies dialect-appropriate
   DROP INDEX syntax.

3. MySQL-only AI buttons visible on PG dashboard
   "Why is it slow?", "Anomaly Detection", "Variable Review",
   "Connection Advisor", "InnoDB Health" all require MySQL-only data
   sources. The buttons were rendered, clicking them returned a
   friendly "MySQL/MariaDB-only" error — but the noise was the user
   experience. capability?(:mysql_only_ai) now returns false on PG,
   and the Server tab gates those five buttons on it. The controller-
   side friendly-error guard stays as a defensive backstop.

4. Misleading server labels on PG
   - "InnoDB Buffer Pool" card → renamed "Buffer Cache" (works for
     both InnoDB and shared_buffers).
   - PG ServerOverview was reporting tmp_disk_pct = 100% whenever
     temp_files > 0, flagging a red badge on every dashboard. PG's
     temp_files is a single count with no in-memory denominator, so
     it now reports 0% and shows N / N for the count.
   - Was mapping xact_rollback to "Aborted Clients" which is a
     different concept; PG's connections block now reports 0 for
     aborted_clients/aborted_connects until a proper PG-side mapping
     exists.

https://claude.ai/code/session_01Xy96nK3Ron2NqukBiSgtzE

Author: claude

app/controllers/concerns/mysql_genius/shared_view_helpers.rb

@@ -45,12 +45,32 @@ def render_partial(name)
       view_context.render(partial: "mysql_genius/queries/#{name}")
     end
 
-    # Capability flag for shared templates. The Rails adapter always
-    # reports every capability as present because it owns all routes
-    # (including the Redis-backed slow_queries / anomaly_detection /
-    # root_cause features).
-    def capability?(_name)
-      true
+    # Capability flag for shared templates. Used to hide AI feature buttons
+    # whose underlying service has no equivalent on the connected dialect
+    # (e.g. InnoDB Health, Variable Review, Connection Advisor, Root Cause
+    # Analysis, and Anomaly Detection all read MySQL-specific server state
+    # via SHOW commands / performance_schema).
+    #
+    # All other capabilities default to true — the Rails adapter owns every
+    # route, including Redis-backed slow_queries.
+    def capability?(name)
+      case name
+      when :mysql_only_ai
+        !connected_to_postgresql?
+      else
+        true
+      end
+    end
+
+    private
+
+    def connected_to_postgresql?
+      MysqlGenius::Core::Connection::ActiveRecordAdapter
+        .new(ActiveRecord::Base.connection)
+        .server_version
+        .postgresql?
+    rescue StandardError
+      false
     end
   end
 end

lib/mysql_genius/core.rb

@@ -42,6 +42,7 @@ def views_path
 require "mysql_genius/core/connection/fake_adapter"
 require "mysql_genius/core/ai/config"
 require "mysql_genius/core/ai/client"
+require "mysql_genius/core/ai/dialect_hints"
 require "mysql_genius/core/ai/suggestion"
 require "mysql_genius/core/ai/optimization"
 require "mysql_genius/core/ai/schema_context_builder"

lib/mysql_genius/core/ai/dialect_hints.rb

@@ -0,0 +1,35 @@
+# frozen_string_literal: true
+
+module MysqlGenius
+  module Core
+    module Ai
+      # Snippets injected into AI system prompts so the model generates SQL in
+      # the right dialect. Without these, prompts hardcode "MySQL" and tell
+      # the model to use backticks — which produces broken SQL on PostgreSQL
+      # (PG uses double quotes for identifiers and has no backtick syntax).
+      module DialectHints
+        extend self
+
+        # Display name suitable for "You are a SQL assistant for a #{name_for}
+        # database." prompts.
+        def name_for(connection)
+          case connection.server_version.dialect
+          when :postgresql then "PostgreSQL"
+          else "MySQL/MariaDB"
+          end
+        end
+
+        # Identifier-quoting rule string for inclusion in a numbered Rules
+        # list in a prompt. PG uses double quotes; MySQL/MariaDB uses
+        # backticks.
+        def identifier_quoting_rule(connection)
+          if connection.server_version.postgresql?
+            %(Use double quotes ("col_name") for case-sensitive identifiers; otherwise leave them bare.)
+          else
+            "Use backticks (`col_name`) for table and column names."
+          end
+        end
+      end
+    end
+  end
+end

lib/mysql_genius/core/ai/index_advisor.rb

@@ -26,7 +26,7 @@ def call(sql, explain_rows)
 
         def system_prompt
           <<~PROMPT
-            You are a MySQL index advisor. Given a query, its EXPLAIN output, and current index/cardinality information, suggest optimal indexes. Consider:
+            You are a #{DialectHints.name_for(@connection)} index advisor. Given a query, its EXPLAIN output, and current index/cardinality information, suggest optimal indexes. Consider:
             - Composite index column ordering (most selective first, or matching query order)
             - Covering indexes to avoid table lookups
             - Partial indexes for long string columns

lib/mysql_genius/core/ai/index_planner.rb

@@ -75,14 +75,14 @@ def user_prompt(schema, unused, duplicates, index_map)
 
         def system_prompt
           <<~PROMPT
-            You are a MySQL index consolidation planner. Given schema information, unused indexes, duplicate indexes, and current index listings, produce a consolidated optimization plan. For each recommendation:
-            - DROP redundant or unused indexes (with exact ALTER TABLE ... DROP INDEX statements)
+            You are a #{DialectHints.name_for(@connection)} index consolidation planner. Given schema information, unused indexes, duplicate indexes, and current index listings, produce a consolidated optimization plan. For each recommendation:
+            - DROP redundant or unused indexes using #{DialectHints.name_for(@connection)} syntax (e.g. `ALTER TABLE t DROP INDEX i;` on MySQL/MariaDB, `DROP INDEX "i";` on PostgreSQL)
             - MERGE overlapping indexes into composites where beneficial
             - KEEP indexes that are actively used and well-structured
             - ADD new composite indexes where query patterns suggest benefit
             - Provide rationale for each change and estimated impact on read/write performance
             #{@config.domain_context}
-            Respond with JSON: {"plan": "markdown with specific ALTER TABLE / DROP INDEX / CREATE INDEX statements, rationale for each change, and estimated impact"}
+            Respond with JSON: {"plan": "markdown with dialect-appropriate DROP INDEX / CREATE INDEX statements, rationale for each change, and estimated impact"}
           PROMPT
         end
       end

lib/mysql_genius/core/ai/migration_risk.rb

@@ -33,7 +33,7 @@ def extract_table_names(migration_sql)
 
         def system_prompt
           <<~PROMPT
-            You are a MySQL migration risk assessor. Given a Rails migration or DDL, evaluate:
+            You are a #{DialectHints.name_for(@connection)} migration risk assessor. Given a Rails migration or DDL, evaluate:
             1. Will this lock the table? For how long given the row count?
             2. Is this safe to run during traffic, or does it need a maintenance window?
             3. Should pt-online-schema-change or gh-ost be used instead?

lib/mysql_genius/core/ai/optimization.rb

@@ -36,7 +36,7 @@ def call(sql, explain_rows, allowed_tables)
 
         def system_prompt(schema_description)
           <<~PROMPT
-            You are a MySQL query optimization expert. Given a SQL query and its EXPLAIN output, analyze the query execution plan and provide actionable optimization suggestions.
+            You are a #{DialectHints.name_for(@connection)} query optimization expert. Given a SQL query and its EXPLAIN output, analyze the query execution plan and provide actionable optimization suggestions.
 
             Available schema:
             #{schema_description}

lib/mysql_genius/core/ai/pattern_grouper.rb

@@ -50,7 +50,7 @@ def extract_tables(stats)
 
         def system_prompt
           prompt = <<~PROMPT
-            You are a MySQL performance analyst specializing in root-cause analysis.
+            You are a #{DialectHints.name_for(@connection)} performance analyst specializing in root-cause analysis.
           PROMPT
 
           if @config.domain_context && !@config.domain_context.empty?

lib/mysql_genius/core/ai/rewrite_query.rb

@@ -27,7 +27,7 @@ def call(sql)
 
         def system_prompt(schema)
           <<~PROMPT
-            You are a MySQL query rewrite expert. Analyze the SQL for anti-patterns and suggest a rewritten version. Look for:
+            You are a #{DialectHints.name_for(@connection)} query rewrite expert. Analyze the SQL for anti-patterns and suggest a rewritten version. Look for:
             - SELECT * when specific columns would suffice
             - Correlated subqueries that could be JOINs
             - OR conditions preventing index use (suggest UNION ALL)

lib/mysql_genius/core/ai/schema_review.rb

@@ -27,7 +27,7 @@ def call(table)
 
         def system_prompt
           <<~PROMPT
-            You are a MySQL schema reviewer. Analyze the following schema and identify anti-patterns and improvement opportunities. Look for:
+            You are a #{DialectHints.name_for(@connection)} schema reviewer. Analyze the following schema and identify anti-patterns and improvement opportunities. Look for:
             - Inappropriate column types (VARCHAR(255) for short values, TEXT where VARCHAR suffices, INT for booleans)
             - Missing indexes on foreign key columns or frequently filtered columns
             - Missing NOT NULL constraints where NULLs are unlikely