We have MAX_BRANCHES to deal with trace explosion, but for a switch compiling a number of traces equal to the number of cases (which can easily be more than 32) is the expected and desired behavior, right? I think we should consider not counting switch "branches" in MAX_BRANCHES, at least once bug 516264 is fixed for switches.

We already do that. From jstracer.cpp: if (anchor && anchor->exitType != CASE_EXIT) ++treeInfo->branchCount;

Ah, ok. Sounds like this bug is invalid then?

Mostly. But even if the main split point is exempt, the overall tree still hits the limit quickly if the cases are not trivial and branch themselves.

Obsolete with the removal of tracejit.