Revert "Fix TCompLoop printing being stateful"

[michael-schwarz]

This reverts commit eded39e. (c.f. #134)

It causes non-termination of the printers (see goblint/analyzer#1290)

[sim642]

Instead of just reverting, I think we should figure out why #134 should somehow cause non-termination because that's not clear at all. It's only removing things which weren't in the hashtable before and were just added.

By never removing them, the visited structs just accumulate and the extraneously accumulated structs somehow prune evaluation. But that pruned evaluation isn't then happening recursively from within the modified block, rather somewhere later in some sequential process. But that seems like something else odd is also going on.

[michael-schwarz]

But that pruned evaluation isn't then happening recursively from within the modified block, rather somewhere later in some sequential process. But that seems like something else odd is also going on.

The sampling seems to point at the recursion happening entirely within pOnlyType.

[michael-schwarz]

One idea: Maybe something is delayed by these doc things that are everywhere and then only evaluated later?

[sim642]

One idea: Maybe something is delayed by these doc things that are everywhere and then only evaluated later?

That can't be because the Pretty.doc is just an ADT of strings and nothing delayable.

[michael-schwarz]

Other possibility: Pretty uses Obj.magic, maybe breaking some assumptions of flambda. Though I think it is too early to suspect the compiler.

[michael-schwarz]

Seems like it is working as intended (as @sim642 explains here goblint/analyzer#1290)