From 5d6702bffc20c0eeae85a84851576b7ee63a4450 Mon Sep 17 00:00:00 2001
From: Mike Vitousek <mmvitousek@gmail.com>
Date: Mon, 16 Sep 2024 16:26:18 -0700
Subject: [PATCH] Update base for Update on "[compiler] Initialize abstract
 values of places in InferReferenceEffects"

Test Plan:
In this diff, we now populate the abstractValue field of places during the InferReferenceEffects pass. The value we populate it with is the same value that we use internally in this pass, but it now will remain accessible to downstream phases as part of the Place.

For phis specifically, we need to do a bit of extra work to compute the appropriate value for the phi, since InferReferenceEffects currently doesn't need to infer a value for phis themselves, only values downstream of them. However, the value we compute should correspond to the value available downstream of the phi.

This changes the error message for one todo test case, because we now are querying for the valueKind earlier in the pass, and hitting an invariant violation as a result of that rather than a later invariant violation.

[ghstack-poisoned]