Less Memory-Hungry, More Principled Solution for Decidable Side-Conditions

[Jason Gross]

@andres-erbsen suggests that it is probably decidable side-condition proving at fault for mit-plv/fiat-crypto#1427

This issue tracks having a principled solution to this problem.

COQC src/Bedrock/End2End/X25519/GarageDoor.v
Finished transaction in 46.551 secs (7.101u,0.757s) (successful)
Finished transaction in 17.874 secs (6.964u,0.165s) (successful)
Finished transaction in 10.409 secs (7.577u,0.452s) (successful)
Command terminated by signal 9
src/Bedrock/End2End/X25519/GarageDoor.vo (real: 1194.29, user: 633.57, sys: 22.09, mem: 3800484 ko)

GarageDoor uses too much RAM for Coq's CI. Either we should provide a target that excludes only GarageDoor and it's reverse dependencies, or we should perfomance-optimize GarageDoor.

https://github.com/coq/coq/runs/8838127340 coq/coq#16638 (comment)

cc @samuelgruetter

[Andres Erbsen]

The side-condition I have in mind is that the compiler-emitted instructions fit in 32 bits each.

[Samuel Gruetter]

The side-condition I have in mind is that the compiler-emitted instructions fit in 32 bits each.

This one should be fixed by 335e57a