Jonathan Blow

The unfortunate thing about SSA is it is a pre-obfuscated concept, so all explanations tend to be kind of terrible. The standard conception needs to be switched to a thing without phi functions, which are silly.

I read phi as “this value may come from any of these source computations”. I coopted flow graph dominators for reasoning about multiple unifications in a functional logic setting. Here the phis denote places where failure or contradiction can occur. Definitely sane there.

They are just hard to understand and give someone who is trying to learn no idea what an implementation would look like. If instead you say “these are parameters to this basic block, they must be supplied by anyone entering the block,” suddenly there is no magic and everything