What happens today, without these enhancements?

Do function calls within if condition expressions simply not get expanded, and then it is up to the back end to deal with it, or not?

What happens today, without these enhancements?

We're having a compiler internal error. So, this is not a performance enhancement, rather then ability to compile certain code (see testcase :) )

To be precise, this BUG_CHECK fires: https://github.com/p4lang/p4c/pull/5073/files#diff-e4d3e15d12ab83675782d2e894a3a319c7b9b4bc72c85af23571a194efc0f8e4L40

One would expect that side-effect ordering would pull the function call out of the if condition into a separate assignment to a temp, but perhaps if running with a backend that disables side-effect ordering. The change just looks like it is duplicating the effect of side-effect ordering of calls in if conditions as part of the inline, which is perhaps more efficient -- only create one temp bool instead of two, though later local-copyprop will get rid of the extra one.

@ChrisDodd

but perhaps if running with a backend that disables side-effect ordering.

It is not a backend that disables side effect ordering. The testcase reproduces with plain p4c. Side effect ordering is "broken" after actions inlining: see, there are no "side effects" for y1 in action a, however, they will appear after actions inlining.

Overall, relying on side effect ordering seems to be a huge downside as it is pretty fragile (in terms of optimizations), it does create lots of temporaries that make memory consumption of e.g. use-def to skyrocket. It looks like side effect ordering is an alternative to e.g. SSA form, however, it does not hold the same invariants as proper SSA form, so one cannot rely on them.