Chess engine internals, search algorithms, and JVM exception debugging — written from code that actually runs.
Same LinkageError family, opposite problem — the class is right there and the JVM refuses to read it. What 61.0 and 65.0 actually mean, the three build topologies that open the gap, and why bumping -target is not a fix.
Read Article
The line in the stack trace is a bystander — whoever allocated last pays for everyone else. Which OOME message you got, three real causes, how to read a dump with dominator trees and retained size, and why -Xmx belongs last.
Read Article
The repeating group of frames is the diagnosis, not the length of the trace. What a thread stack actually holds, why "how deep can I recurse" has no fixed answer, three real causes, and why raising -Xss belongs last.
Read Article
One is a checked exception you asked for, the other is an Error the JVM hands you — and "Could not initialize class" does not mean the class is missing at all. The three loading phases, three real causes, and the fixes with their costs.
Read Article
It has nothing to do with threads, and removing the second-to-last element does not throw at all — it silently skips the last one. The modCount machinery, three real causes, and four fixes with their actual costs.
Read Article
JEP 358 turned the emptiest stack trace in Java into a sentence. The grammar of that sentence, three production causes with the exact message each one throws, and the places it still leaves you stranded.
Read Article
The engine played fine offline, then sat still in its first live game while the clock ran out. The irreversible BOT upgrade, two NDJSON streams, a systemd unit, and eight failures that only appear against real opponents.
Read Article
Move generation, search and evaluation all worked, and Arena still showed the engine as not responding. The handshake, the position and go parsers, the worker thread that keeps stop readable, and the missing flush that lost a game on time.
Read Article
A correct search that only counts material shuffles rooks, refuses to castle and parks knights on the rim. Piece values, 64 numbers per piece type, a midgame-to-endgame taper, a mobility term, and the symmetry test that caught three bugs in one assertion.
Read Article
A correct move generator still plays random chess. So I wrote minimax, added alpha-beta, and counted every node instead of trusting the textbook square root. Move ordering turned out to do almost nothing from the start position and almost everything in a real midgame.
Read Article
Three rules every tutorial hand-waves. They are not hard to understand, they are hard to get right, because none of them live on the board. A stale en passant square, rights that outlived a captured rook, and a clock counting the wrong thing.
Read Article
My engine played legal-looking chess, and legal-looking is not legal. Perft gave me a number that was either right or wrong, and perft divide told me exactly where the en passant, castling, promotion and file-wrapping bugs were hiding.
Read Article
I closed the tab the first time I read about magic bitboards. Months later the same page made sense. The intuition behind occupancy masks and blockers, concise Java code, and every bug that cost me an evening.
Read Article
Every resource jumped straight to magic bitboards. I skipped them and wrote the plain ray-walking generator instead — the direction arrays, the blocker logic, the bugs that ate two weekends, and why correctness beats clever optimisation.
Read Article
A chess board has 64 squares; a Java long has 64 bits. The explanation I wish I'd had — why engines switch from arrays to bitboards, the bitwise tricks, move generation, and the bugs that taught me how they really work.
Read Article
After a stack of Todo apps and inventory systems, I built a chess engine in Java. The design decisions, the minimax and alpha-beta code, the brutal bugs, and what a genuinely hard project actually teaches you.
Read Article