monomorphization+inlining sometimes leads to undefined functions

[stylewarning]

Disable specialization, and compile a file containing:

(in-package #:coalton-user)

(named-readtables:in-readtable coalton:coalton)

(coalton-toplevel
  (inline) (monomorphize)
  (declare f (UFix -> (:t -> :t -> Boolean) -> (UFix -> :t) -> :t))
  (define (f n better? func)
    (assert (< 0 n))
    (rec % ((i 1) (best (func 0)))
      (cond
        ((== i n)
         best)
        (True
         (let ((candidate (func i)))
           (if (better? candidate best)
               (% (1+ i) candidate)
               (% (1+ i) best))))))))

(coalton-toplevel
  (define (g)
    (f 10 < (fn (x) x))))

In the processing of g, you will get a warning akin to:

; caught COMMON-LISP:STYLE-WARNING:
;   undefined function: COALTON-USER::|<#3626|

This does not happen if:

1. They're in the same coalton-toplevel.
2. They're compiled separately with C-c-c.
3. (monomorphize) is removed.
4. (inline) is removed.

The issue seems to be that g contains a function named |<#X|, but f had defined |<#Y| for some Y.

[stylewarning]

If I remove the readtable setting, the issue also goes away. Quizzically, putting a break in the monomorphizer tells me f is being monomorphized twice, once in ordiinary processing, and a second time when a code deletion note is being issued. Very weird but also not surprising given we hijack #\(.