Lila: Singleton WASMX engine not properly reinitialized when switching study chapters

[mmixu]

I don't believe the cloud eval is correct after 2. Nf3 https://lichess.org/cMuPJorY/white#2

[mmixu]

Another weird one: https://lichess.org/PNIPbK7v#2
"2. d4?? Blunder Ne5 was the best. 2. Ne5 Nxe5 3. c4 Nxc4 5. Qc2 Nxd2 5. Qxh7 Rxh7 6. Kxd2 Rxh2"

Is the analysis recommending antichess moves in crazyhouse?
Antichess position after 2. Ne5: https://i.imgur.com/ZVVKBV9.png
ZH cloud and analysis: https://i.imgur.com/LFriYaZ.png

[ijhchess]

It seems the engine is messed up for every variant. After 1.Nf3 in atomic, the engine suggest c5?? which instantly loses to Ne5...

[niklasf]

@ijhchess Cannot reproduce. Can you please make a new issue with more details?

@mmixu ~Interestingly lichess-adele provided server side analysis for both of these.~ Edit: Red herring. Same issue after requested analysis again.

[niklasf]

• [x] Implement shutdown for stockfish.wasm (via v0.8.0)
• [x] Rename export of stockfish-mv.wasm (via v0.4.0)
• [x] Fix loading / initialization (via 6ee2da30e1d967c33a4ff8840e633ac7f938f905)
• [x] Asset deploy
• [x] Flush eval cache: Bug was introduced in 1e5543eac7898e67819e883f5eb28aaf97e27d57. So everything after that could be affected, variants more likely than standard chess

[niklasf]

@mmixu @ijhchess @bftjoe @kraktus @xehpuk Thanks to everyone who reported positions. The fix has been deployed to production and the cloud evals removed. Existing bad server side analysis was not deleted. Please report again if there's a second issue, now or in the future.