Validate card counts in CalcAllTablesN to prevent segfault on malformed deals

[danielneill]

Summary

calc_all_tables_pbn / CalcAllTablesN can segfault when a batch contains a malformed deal — one whose hands don't each hold 13 cards. The same deal solved alone returns RETURN_CARD_COUNT (-14) cleanly.

Mechanism

A malformed deal passes convert_from_pbn and reaches the multi-threaded solver, which reads out of bounds before the per-board card-count check fires. The crash is intermittent — it depends on the number of tables in the batch and the heap layout (a 32-table batch crashes; the same malformed deal in a 2-table batch returns -14).

Fix

Validate up front in CalcAllTablesN that every hand holds exactly 13 cards (std::popcount over the four suit bitmasks), returning RETURN_CARD_COUNT before any board is dispatched to the solver. This matches the contract the single-board path already enforces.

Scope notes:

• Covers the wrong-count case (the observed crash). Cross-hand duplicate cards (each hand has 13 but a card is repeated) are out of scope.
• The single-board path (CalcDDtableN) is unchanged — it already returns -14 without crashing.

Testing

• New regression test: a malformed deal raises both solo and inside a batch (previously segfaulted the interpreter).
• Verified against a build of this branch: a 32-deal batch that segfaulted (exit 139) now raises -14; valid batches still return tables; all existing //python:all tests pass.

[Copilot]

Pull request overview

This PR prevents intermittent segfaults in batch table calculation (calc_all_tables_pbn / CalcAllTablesN) when a deal is malformed (a hand does not contain 13 cards) by adding an upfront per-hand card-count validation, and adds a Python regression test to ensure the batch path raises an error instead of crashing.

Changes:

• Add an early card-count check in CalcAllTablesN using std::popcount over the suit bitmasks to return RETURN_CARD_COUNT before dispatching work to the multi-threaded solver.
• Add a Python regression test that includes a malformed deal inside a multi-deal batch and asserts that the call raises rather than crashing.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
library/src/calc_tables.cpp	Adds upfront 13-cards-per-hand validation in CalcAllTablesN (requires <bit> for std::popcount).
python/tests/test_calc_tables_regression.py	Adds regression coverage ensuring malformed deals in batch raise cleanly instead of crashing.

[zzcgumn]

Are we sure that this is what we want? I think all hands have the same number of cards and at most 13 cards is the correct criteria as partial deals are allowed.

[tzimnoch]

+1

I also prefer to hide this feature behind a flag. While this isn't enormously expensive, I don't want to take the performance hit to verify something I already guaranteed some other way. This is a speed-critical library for many use cases.