Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
Codex Review: Here are some suggestions.
Reply with @codex fix comments to fix any unresolved comments.
About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you open a pull request for review, mark a draft as ready, or comment "@codex review". If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex fix this CI failure" or "@codex address that feedback".
src/position.cpp
Outdated
| b |= attacks_bb(~c, move_pt, s, occupied) & attacks_bb(~c, move_pt, s, occupied & ~janggiCannons) & pieces(c, JANGGI_CANNON); | ||
| else | ||
| b |= attacks_bb(~c, move_pt, s, occupied) & pieces(c, pt); | ||
|
|
||
| b |= check_attacks_bb(~c, move_pt, s, occupied) & pieces(c, pt); |
There was a problem hiding this comment.
[P1] Check-only attacks ignore blocking for asymmetrical riders
The new check_attacks_bb() call is added unconditionally after the asymmetrical-rider branch in attackers_to. For pieces whose check-only movement is asymmetric (e.g., a horse-style piece whose leg squares must be empty), the routine now evaluates check-only attacks using check_attacks_bb(~c, move_pt, s, occupied) without the per-attacker validation performed in the ASYMMETRICAL_RIDERS branch above. This means a piece can be considered to give check through a blocked leg, because the blocking is tested relative to the target square instead of the attacker. Any variant that defines a check-only horse-like move will produce false positive checks and illegal evasions.
Useful? React with 👍 / 👎.
| { | ||
| PieceType pt = pop_lsb(ps); | ||
| PieceType movePt = pt == KING ? king_type() : pt; | ||
| si->checkSquares[pt] = ksq != SQ_NONE ? attacks_bb(~sideToMove, movePt, ksq, pieces()) : Bitboard(0); | ||
| si->checkSquares[pt] = ksq != SQ_NONE | ||
| ? (attacks_bb(~sideToMove, movePt, ksq, pieces()) | ||
| | check_attacks_bb(~sideToMove, movePt, ksq, pieces())) | ||
| : Bitboard(0); | ||
| // Collect special piece types that require slower check and evasion detection | ||
| if (AttackRiderTypes[movePt] & NON_SLIDING_RIDERS) | ||
| si->nonSlidingRiders |= pieces(pt); |
There was a problem hiding this comment.
[P1] Non-sliding check-only pieces omitted from slow-check bookkeeping
set_check_info still populates nonSlidingRiders using AttackRiderTypes only. If a piece has no capture moves and only check-only leaping moves (e.g., a horse that can only give check), AttackRiderTypes stays zero while CheckRiderTypes contains the non-sliding rider bits. Such pieces are therefore excluded from nonSlidingRiders, and the later check-handling code treats their checks as if they were sliding and blockable. In positions where a check is delivered by a check-only non-sliding piece, legal move generation can incorrectly accept blocks instead of enforcing capture or king moves. The bookkeeping should also consider check-only rider types.
Useful? React with 👍 / 👎.
|
@codex fix comments |
|
Summary
Testing
|
1 participant
Summary
kqualifierTesting
cd src && make -j2 ARCH=x86-64 buildhttps://chatgpt.com/codex/tasks/task_e_68d12f3fd62c83229eb36b0733dee539