Ben Gamari authored Feb 01, 2017 and Ben Gamari committed Feb 01, 2017

This fixes a regression due to D2605 (see #8472) wherein top-level primitive
strings would fail to be noticed by CoreToStg as they were wrapped in a
tick. This resulted in a panic in CoreToStg due to inconsistent CAF information
(or a Core Lint failure, if enabled). Here we document the invariant that
unlifted expressions can only sit at top-level if of the form `Lit (MachStr
...)` with no ticks or other embellishments. Moreover, we fix instance of
this in `Simplify.prepareRhs` and `FloatOut.wrapTick` where this
invariant was being broken.

Test Plan: Validate with `-g`. Run testsuite with `WAY=ghci`.

Reviewers: austin, simonpj

Reviewed By: simonpj

Subscribers: simonpj, akio, scpmw, thomie

Differential Revision: https://phabricator.haskell.org/D3051