text2sql-engine 0.1.1__py3-none-any.whl → 0.2.0__py3-none-any.whl

text2sql/__init__.py

@@ -37,7 +37,7 @@ from .types import (
     TokenUsage,
 )
 
-__version__ = "0.1.1"
+__version__ = "0.2.0"
 
 __all__ = [
     "Text2SQL",

text2sql/pipeline/generation.py

@@ -82,6 +82,19 @@ class GenerationStage:
         reduced_schema = full_schema.subset(linked.tables)
         reduced_text = schema_to_prompt(reduced_schema) if reduced_schema.tables else "(none)"
 
+        # Forward Stage 1's reasoning to Stage 2. The linking model already
+        # worked out the logic (ranges, joins, which column means what); the SQL
+        # model should follow it instead of guessing.
+        columns_text = "\n".join(
+            f"- {c.table}.{c.column}" + (f" — {c.reason}" if c.reason else "")
+            for c in linked.columns
+        ) or "(none)"
+        joins_text = "\n".join(
+            f"- {j.left_table}.{j.left_column} = {j.right_table}.{j.right_column}"
+            for j in linked.joins
+        ) or "(none)"
+        rationale_text = linked.rationale.strip() or "(none)"
+
         template = self._prompts.load(stage_cfg.prompt_template)
         system, user = template.render(
             linked_schema=reduced_text,
@@ -89,6 +102,9 @@ class GenerationStage:
             dialect=self._settings.general.sql_dialect,
             entities=", ".join(linked.entities) or "(none)",
             filters="; ".join(linked.filters) or "(none)",
+            columns=columns_text,
+            joins=joins_text,
+            rationale=rationale_text,
         )
 
         attempts = {"n": 0}

text2sql/prompts/generation.txt

@@ -1,12 +1,17 @@
 SYSTEM:
-You are an expert SQL generator. You get a user request and a reduced schema
-(only the relevant tables/columns/joins picked by an upstream linking step).
-Produce one correct SQL query in the target dialect.
+You are an expert SQL generator. You get a user request, a reduced schema (only
+the relevant tables/columns/joins picked by an upstream linking step), and that
+step's analysis. Produce one correct SQL query in the target dialect.
 
 Rules:
 - Use only the tables and columns in the reduced schema.
+- Follow the linking analysis. It already worked out the logic; do not ignore it.
+- Respect the intended comparison semantics from the analysis and filter hints.
+  If a value should be matched by a numeric range or an exact compare, implement
+  it that way. Do not fall back to substring/text (LIKE) matching unless that is
+  clearly what is meant.
+- Use the suggested joins to connect tables. Prefer explicit JOINs.
 - Write idiomatic SQL for the target dialect: {dialect}.
-- Prefer explicit JOINs using the given relations.
 - Do not run the query. Only produce it.
 - Return the answer strictly as the requested JSON object with two fields:
   "sql" (the query) and "explanation" (one or two sentences).
@@ -19,6 +24,15 @@ Reduced schema (relevant subset):
 {linked_schema}
 --------------------------------
 
+Linking analysis (reasoning from the previous step — follow this logic):
+{rationale}
+
+Relevant columns:
+{columns}
+
+Suggested joins:
+{joins}
+
 Detected entities: {entities}
 Filter hints: {filters}
 

text2sql/prompts/generation_raw.txt

@@ -1,12 +1,17 @@
 SYSTEM:
-You are an expert SQL generator. You get a user request and a reduced schema
-(only the relevant tables/columns/joins picked by an upstream linking step).
-Produce one correct SQL query in the target dialect.
+You are an expert SQL generator. You get a user request, a reduced schema (only
+the relevant tables/columns/joins picked by an upstream linking step), and that
+step's analysis. Produce one correct SQL query in the target dialect.
 
 Rules:
 - Use only the tables and columns in the reduced schema.
+- Follow the linking analysis. It already worked out the logic; do not ignore it.
+- Respect the intended comparison semantics from the analysis and filter hints.
+  If a value should be matched by a numeric range or an exact compare, implement
+  it that way. Do not fall back to substring/text (LIKE) matching unless that is
+  clearly what is meant.
+- Use the suggested joins to connect tables. Prefer explicit JOINs.
 - Write idiomatic SQL for the target dialect: {dialect}.
-- Prefer explicit JOINs using the given relations.
 - Do not run the query. Only produce it.
 - Output ONLY the SQL query. No explanation, no comment, no markdown fences.
 
@@ -18,6 +23,15 @@ Reduced schema (relevant subset):
 {linked_schema}
 --------------------------------
 
+Linking analysis (reasoning from the previous step — follow this logic):
+{rationale}
+
+Relevant columns:
+{columns}
+
+Suggested joins:
+{joins}
+
 Detected entities: {entities}
 Filter hints: {filters}
 

{text2sql_engine-0.1.1.dist-info → text2sql_engine-0.2.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: text2sql-engine
-Version: 0.1.1
+Version: 0.2.0
 Summary: Schema-agnostic, config-driven Text-to-SQL library (2-stage LLM pipeline). Produces a SQL string; never connects to or executes against a database.
 Author: Mehmet Kaan Durupunar
 License: MIT
@@ -70,8 +70,9 @@ user request
   reduces the schema to the relevant part and returns structured JSON. Output is
   forced via `response_format` (OpenAI) or tool-calling (Anthropic), with
   strict-JSON parsing and retry as a fallback.
-- **Stage 2 (generation)** takes the request and that reduced subset. It writes
-  SQL in the target dialect. Two output modes (config `output_mode`):
+- **Stage 2 (generation)** takes the request, that reduced subset, and Stage 1's
+  reasoning (rationale, chosen joins, column reasons). It follows that logic and
+  writes SQL in the target dialect. Two output modes (config `output_mode`):
   `structured` forces JSON with `sql` + `explanation`; `raw` asks for plain SQL
   text, for specialized text-to-SQL models.
 - The two stages can use **different providers and models**. For example a cheap

{text2sql_engine-0.1.1.dist-info → text2sql_engine-0.2.0.dist-info}/RECORD

@@ -1,17 +1,17 @@
-text2sql/__init__.py,sha256=oMYRHWUUR6cRjpG0EAMdmDAPF82UCPbMEd4lyRx_4Sc,1442
+text2sql/__init__.py,sha256=dwQ98mfnAHRbUhKmOEnC0BcGK-8Slxosi8hqDgpkheM,1442
 text2sql/config.py,sha256=6snv99rf8jf8NtD3xetSrvIaltbLyq2eoISoMoPadSw,8503
 text2sql/errors.py,sha256=Y5ORy7g_UsFvDJ2KorB2JFvTyOBqmrPv9MWc8Cl9MI8,1052
 text2sql/logging_utils.py,sha256=hiEH1SDqP16kctUj4uPP0tbXdrCO9AlRcdS-mHYDg5s,1633
 text2sql/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 text2sql/types.py,sha256=MxcdHP7Gnftk1yKdcBQKaVVyeqRnBu5kxbhSZ2xJJ80,3532
 text2sql/pipeline/__init__.py,sha256=rjEOi4U9C-SKP15cyKPdvG_EZBOW0VW01LouS3PJVdY,245
-text2sql/pipeline/generation.py,sha256=rXzU6e4LyyPTlHSkqT4IsZA-Xf-DjK3Kp9QAbpA2PJ4,5030
+text2sql/pipeline/generation.py,sha256=M9wehTcMAq1QoK1nnQ4SJd1mszSfXs9E06l26IXpKNw,5750
 text2sql/pipeline/json_utils.py,sha256=92viA3xyw_GMF_nKdjSPgAyQ8iBUZgnyNRAoyWC5ZlE,3145
 text2sql/pipeline/linking.py,sha256=U28LOcWJn3cItwAsv2fG4fYv9KcJ80XmP1Fi8nNx4qc,6966
 text2sql/pipeline/pipeline.py,sha256=TE0cHSEJyeY43P89HsQDbnPGSfgRQmtw7KZAq-Yjj1U,5675
 text2sql/prompts/__init__.py,sha256=HfeflfAGsZl3xHnNrmklTP6K_8EKxX6vWJrayq6QuEQ,3315
-text2sql/prompts/generation.txt,sha256=r4125WCr7gsJIzTa5wZTAmbTyn5WvP3QrfGf4CWNP_w,849
-text2sql/prompts/generation_raw.txt,sha256=wAvUvHd2Qywfy-Poc-7lqW5Gmg1nf3IyFDwtZNBCJls,789
+text2sql/prompts/generation.txt,sha256=th97BZsHFlRtH30OqHj4DY_BooCsa7Zad9EoXC9M_cM,1381
+text2sql/prompts/generation_raw.txt,sha256=EsKLl7WrNWQzcoxAFW9l6yucUtBa7sIimCsvIjRx7ZI,1321
 text2sql/prompts/linking.txt,sha256=nLPSkaq3PiWiJ7wXLeD0BvK_-rR3hhc_tLhZ-daPnGY,889
 text2sql/providers/__init__.py,sha256=R5a-YKtyaUFpkVJGWrmhDGyRHeh84gA3F7nEH57dSGo,345
 text2sql/providers/anthropic_provider.py,sha256=Zhh8zM6EfHqkjTgNmhYKdZLHQ69b0Pu2-5WbCW2rD6s,4342
@@ -23,8 +23,8 @@ text2sql/schema/file_provider.py,sha256=Nv9x2-Eh9mXISk5XvUQaLfagth2s3LbX5F7jWp-o
 text2sql/schema/models.py,sha256=yX5zdKbTuy-g-nkwT0v-0HUTK81jBRoabV-OC-kD-JI,2263
 text2sql/schema/provider.py,sha256=JPMRsjRgEB4RdQR_pFwbP2LwLIXZhQVWSQPoYzaXC7s,798
 text2sql/schema/serialize.py,sha256=_2MeA1IwDJ51-69PXyqXBSFZSyzovbqJkRIMpekkfJA,1833
-text2sql_engine-0.1.1.dist-info/licenses/LICENSE,sha256=qfdfVnpb5NOM3-SK13PnmWagjCLy5sE9ORlgOGXc8hU,1078
-text2sql_engine-0.1.1.dist-info/METADATA,sha256=ixiIwDgwu2F0KvULAh2niQSs-OW_dI7faEMGC4hgfb4,12971
-text2sql_engine-0.1.1.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
-text2sql_engine-0.1.1.dist-info/top_level.txt,sha256=ZONWO1r8wsWpyHhF5qPzAjnsNoCEhwZ2Z-M_V8nKALc,9
-text2sql_engine-0.1.1.dist-info/RECORD,,
+text2sql_engine-0.2.0.dist-info/licenses/LICENSE,sha256=qfdfVnpb5NOM3-SK13PnmWagjCLy5sE9ORlgOGXc8hU,1078
+text2sql_engine-0.2.0.dist-info/METADATA,sha256=Xn9bMUbUlTrirl1JWpZ99bOrfZRZzK2IdDx8bQVmRFw,13061
+text2sql_engine-0.2.0.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
+text2sql_engine-0.2.0.dist-info/top_level.txt,sha256=ZONWO1r8wsWpyHhF5qPzAjnsNoCEhwZ2Z-M_V8nKALc,9
+text2sql_engine-0.2.0.dist-info/RECORD,,