rowproof 0.1.0__tar.gz → 0.1.2__tar.gz

{rowproof-0.1.0 → rowproof-0.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rowproof
-Version: 0.1.0
+Version: 0.1.2
 Summary: Prove a database migration or replication didn't lose, add or change a single row.
 License: Apache-2.0
 Requires-Python: >=3.11
@@ -87,6 +87,18 @@ public.orders  postgres:postgres/public.orders  ->  clickhouse:default/orders
 | ClickHouse | `clickhouse-connect` | |
 | Snowflake | `snowflake-connector-python` | Optional extra: `rowproof[snowflake]` |
 
+### Snowflake: known limitations
+
+Snowflake support works, but it is the least battle-tested of the three engines. It has been tested against a single trial account with small tables; Postgres and ClickHouse have had far more. Know these before you rely on it:
+
+- **Login is username and password only.** The connection string has no place for a warehouse, role, SSO, MFA, or key-pair login. Accounts that require SSO or key-pair authentication won't work yet, and queries run on the user's *default* warehouse (a user with none set will fail).
+- **Column names are case-sensitive.** Snowflake stores unquoted names in upper case, so `--key id` won't match a column stored as `ID`. Pass `--key ID`, or create the table with quoted lower-case names. The error message tells you when this is the problem.
+- **Primary keys are often not declared** on Snowflake tables (Snowflake doesn't enforce them). If auto-detection finds none, pass `--key`.
+- **JSON columns are compared as plain text.** `VARIANT`, `ARRAY` and `OBJECT` use rule UNK-1: two JSON values that mean the same thing but are formatted differently will show as different, and a JSON `null` is not distinguished from an SQL `NULL`.
+- **Special characters in passwords must be percent-encoded** in the connection string (`@` becomes `%40`).
+- **It uses your warehouse's credits.** Every query runs on your Snowflake warehouse. Start with a small table or `--sample`.
+- **No Snowflake benchmark yet.** The 100M-row benchmark below is Postgres to ClickHouse only.
+
 ## Supported types (canonical comparison rules)
 
 Every cross-engine type difference is governed by one of these published rules, and every difference rowproof reports cites the rule ID that applied — never a vague "values differ."

{rowproof-0.1.0 → rowproof-0.1.2}/README.md

@@ -65,6 +65,18 @@ public.orders  postgres:postgres/public.orders  ->  clickhouse:default/orders
 | ClickHouse | `clickhouse-connect` | |
 | Snowflake | `snowflake-connector-python` | Optional extra: `rowproof[snowflake]` |
 
+### Snowflake: known limitations
+
+Snowflake support works, but it is the least battle-tested of the three engines. It has been tested against a single trial account with small tables; Postgres and ClickHouse have had far more. Know these before you rely on it:
+
+- **Login is username and password only.** The connection string has no place for a warehouse, role, SSO, MFA, or key-pair login. Accounts that require SSO or key-pair authentication won't work yet, and queries run on the user's *default* warehouse (a user with none set will fail).
+- **Column names are case-sensitive.** Snowflake stores unquoted names in upper case, so `--key id` won't match a column stored as `ID`. Pass `--key ID`, or create the table with quoted lower-case names. The error message tells you when this is the problem.
+- **Primary keys are often not declared** on Snowflake tables (Snowflake doesn't enforce them). If auto-detection finds none, pass `--key`.
+- **JSON columns are compared as plain text.** `VARIANT`, `ARRAY` and `OBJECT` use rule UNK-1: two JSON values that mean the same thing but are formatted differently will show as different, and a JSON `null` is not distinguished from an SQL `NULL`.
+- **Special characters in passwords must be percent-encoded** in the connection string (`@` becomes `%40`).
+- **It uses your warehouse's credits.** Every query runs on your Snowflake warehouse. Start with a small table or `--sample`.
+- **No Snowflake benchmark yet.** The 100M-row benchmark below is Postgres to ClickHouse only.
+
 ## Supported types (canonical comparison rules)
 
 Every cross-engine type difference is governed by one of these published rules, and every difference rowproof reports cites the rule ID that applied — never a vague "values differ."

{rowproof-0.1.0 → rowproof-0.1.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "rowproof"
-version = "0.1.0"
+version = "0.1.2"
 description = "Prove a database migration or replication didn't lose, add or change a single row."
 readme = "README.md"
 requires-python = ">=3.11"

rowproof-0.1.2/rowproof/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.1.2"

{rowproof-0.1.0 → rowproof-0.1.2}/rowproof/cli/main.py

@@ -135,23 +135,26 @@ def _parse_column_map(value: str | None) -> dict | None:
     return result or None
 
 
+def _hook_sql_log(connector, sql_log: list[str] | None) -> None:
+    """spec §8.2: JSON output carries every generated SQL statement too —
+    reuse the exact same on_query hook --verbose already installs
+    (chaining onto it, so --verbose logging to stderr keeps working
+    unchanged) rather than a second SQL-capturing mechanism. `list.append`
+    is atomic in CPython, so pooled worker threads can share one log.
+    """
+    if sql_log is None or not hasattr(connector, "on_query"):
+        return
+    previous = connector.on_query
+    connector.on_query = lambda sql, _prev=previous: (_prev(sql) if _prev else None, sql_log.append(sql))
+
+
 def _connect_pair(args, sql_log: list[str] | None = None):
     src_spec = parse_source_spec(args.source)
     tgt_spec = parse_source_spec(args.target)
     source = _make_connector(src_spec.engine, args.verbose)
     target = _make_connector(tgt_spec.engine, args.verbose)
-    if sql_log is not None:
-        for connector in (source, target):
-            if hasattr(connector, "on_query"):
-                previous = connector.on_query
-                # spec §8.2: JSON output carries every generated SQL
-                # statement too — reuse the exact same on_query hook
-                # --verbose already installs (chaining onto it, so
-                # --verbose logging to stderr keeps working unchanged)
-                # rather than a second SQL-capturing mechanism.
-                connector.on_query = (
-                    lambda sql, _prev=previous: (_prev(sql) if _prev else None, sql_log.append(sql))
-                )
+    _hook_sql_log(source, sql_log)
+    _hook_sql_log(target, sql_log)
     _connect_with_retry(source, src_spec.connect_dsn)
     _connect_with_retry(target, tgt_spec.connect_dsn)
 
@@ -168,6 +171,12 @@ def _connect_pair(args, sql_log: list[str] | None = None):
     if threads > 1:
         source_pool = [_make_connector(src_spec.engine, args.verbose) for _ in range(threads)]
         target_pool = [_make_connector(tgt_spec.engine, args.verbose) for _ in range(threads)]
+        # The pooled connectors run the per-segment hash queries -- the
+        # actual proof of a match -- so they must feed the same SQL log as
+        # the main connectors, or the JSON/HTML "Reproduce" output would
+        # silently omit exactly the queries a reviewer wants to re-run.
+        for connector in (*source_pool, *target_pool):
+            _hook_sql_log(connector, sql_log)
         for connector in source_pool:
             _connect_with_retry(connector, src_spec.connect_dsn)
         for connector in target_pool:

{rowproof-0.1.0 → rowproof-0.1.2}/rowproof/core/errors.py

@@ -20,6 +20,32 @@ class NoPrimaryKeyError(RowProofError):
         self.table_name = table_name
 
 
+class TableNotFoundError(RowProofError):
+    def __init__(self, table_name: str):
+        super().__init__(
+            f"Table '{table_name}' not found, or it has no visible columns "
+            "(check the name, the schema, and that this login can read it)."
+        )
+        self.table_name = table_name
+
+
+class KeyColumnNotFoundError(RowProofError):
+    def __init__(self, table_name: str, column: str, available: list[str]):
+        hint = ""
+        by_lower = {c.lower(): c for c in available}
+        if column.lower() in by_lower and by_lower[column.lower()] != column:
+            hint = (
+                f" Did you mean '{by_lower[column.lower()]}'? Column names are "
+                "case-sensitive here (Snowflake stores unquoted names in upper case)."
+            )
+        super().__init__(
+            f"Key column '{column}' not found in '{table_name}'.{hint} "
+            f"Available columns: {', '.join(available)}."
+        )
+        self.table_name = table_name
+        self.column = column
+
+
 class NonUniqueKeyError(RowProofError):
     def __init__(self, table_name: str, key_columns: list[str], duplicate_example: tuple):
         example = ", ".join(f"{c}={v!r}" for c, v in zip(key_columns, duplicate_example))

{rowproof-0.1.0 → rowproof-0.1.2}/rowproof/core/hashdiff.py

@@ -19,7 +19,12 @@ from dataclasses import dataclass
 
 from rowproof.connectors.base import Connector
 from rowproof.core.column_matching import match_columns
-from rowproof.core.errors import NonUniqueKeyError, NoPrimaryKeyError
+from rowproof.core.errors import (
+    KeyColumnNotFoundError,
+    NonUniqueKeyError,
+    NoPrimaryKeyError,
+    TableNotFoundError,
+)
 from rowproof.core.models import (
     Algorithm,
     Column,
@@ -105,7 +110,12 @@ def _build_plan(
     where: str | None = None,
 ) -> _Plan:
     cols_by_name = _resolve_columns(connector, table)
+    if not cols_by_name:
+        raise TableNotFoundError(table.qualified_name)
     resolved_key = _resolve_key_columns(connector, table, key_columns)
+    for key_col in resolved_key:
+        if key_col not in cols_by_name:
+            raise KeyColumnNotFoundError(table.qualified_name, key_col, list(cols_by_name))
 
     if columns:
         selected = list(columns)

{rowproof-0.1.0 → rowproof-0.1.2}/rowproof.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rowproof
-Version: 0.1.0
+Version: 0.1.2
 Summary: Prove a database migration or replication didn't lose, add or change a single row.
 License: Apache-2.0
 Requires-Python: >=3.11
@@ -87,6 +87,18 @@ public.orders  postgres:postgres/public.orders  ->  clickhouse:default/orders
 | ClickHouse | `clickhouse-connect` | |
 | Snowflake | `snowflake-connector-python` | Optional extra: `rowproof[snowflake]` |
 
+### Snowflake: known limitations
+
+Snowflake support works, but it is the least battle-tested of the three engines. It has been tested against a single trial account with small tables; Postgres and ClickHouse have had far more. Know these before you rely on it:
+
+- **Login is username and password only.** The connection string has no place for a warehouse, role, SSO, MFA, or key-pair login. Accounts that require SSO or key-pair authentication won't work yet, and queries run on the user's *default* warehouse (a user with none set will fail).
+- **Column names are case-sensitive.** Snowflake stores unquoted names in upper case, so `--key id` won't match a column stored as `ID`. Pass `--key ID`, or create the table with quoted lower-case names. The error message tells you when this is the problem.
+- **Primary keys are often not declared** on Snowflake tables (Snowflake doesn't enforce them). If auto-detection finds none, pass `--key`.
+- **JSON columns are compared as plain text.** `VARIANT`, `ARRAY` and `OBJECT` use rule UNK-1: two JSON values that mean the same thing but are formatted differently will show as different, and a JSON `null` is not distinguished from an SQL `NULL`.
+- **Special characters in passwords must be percent-encoded** in the connection string (`@` becomes `%40`).
+- **It uses your warehouse's credits.** Every query runs on your Snowflake warehouse. Start with a small table or `--sample`.
+- **No Snowflake benchmark yet.** The 100M-row benchmark below is Postgres to ClickHouse only.
+
 ## Supported types (canonical comparison rules)
 
 Every cross-engine type difference is governed by one of these published rules, and every difference rowproof reports cites the rule ID that applied — never a vague "values differ."

rowproof-0.1.0/rowproof/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.1.0"