Error from batch job with gh installed

[vermiculus]

$ emacs -batch -Q --eval "(progn (toggle-debug-on-error) (require 'package) (message \"initialize\") (package-initialize) (message \"done\"))"
Debug on Error enabled globally
initialize
Eager macro-expansion failure: (void-function eieio-make-class-predicate)
Debugger entered--Lisp error: (void-function eieio-make-class-predicate)
  eieio-make-class-predicate(marshal-driver)
  byte-code("\300\301\302\303!\"\210\300\304\305\303!\"\210\300\306\304\"\210\307\306\310\311#\210\312\303\313\304#\210\314\303\315\316\315$\207" [defalias marshal-driver-p eieio-make-class-predicate marshal-driver marshal-driver--eieio-childp eieio-make-child-predicate marshal-driver-child-p make-obsolete "use (cl-typep ... \\='marshal-driver) instead" "25.1" put cl-deftype-satisfies eieio-defclass-internal nil ((input :initarg :input) (output :initarg :output))] 5)
  require(marshal)
  byte-code("\300\301!\210\300\302!\210\300\303!\210\300\304!\210\300\305!\210\306\307\310\311\312\313%\207" [require eieio dash marshal s gh-profile custom-declare-group gh nil "Github API client libraries." :group applications] 6)
  (gh-defclass gh-object nil nil)
  eval-buffer(#<buffer  *load*> nil "/Users/sean/.emacs.d/elpa/gh-20160728.1525/gh-autoloads.el" nil t)  ; Reading at buffer position 2204
  load-with-code-conversion("/Users/sean/.emacs.d/elpa/gh-20160728.1525/gh-autoloads.el" "/Users/sean/.emacs.d/elpa/gh-20160728.1525/gh-autoloads.el" nil t)
  load("/Users/sean/.emacs.d/elpa/gh-20160728.1525/gh-autoloads" nil t)
  #[0 "\302\303\304\305\300\"\301\"\306\307#\207" [gh "/Users/sean/.emacs.d/elpa/gh-20160728.1525" load expand-file-name format "%s-autoloads" nil t] 5 "\n\n(fn)"]()
  funcall(#[0 "\302\303\304\305\300\"\301\"\306\307#\207" [gh "/Users/sean/.emacs.d/elpa/gh-20160728.1525" load expand-file-name format "%s-autoloads" nil t] 5 "\n\n(fn)"])
  package-activate-1([cl-struct-package-desc gh (20160728 1525) "A GitHub library for Emacs" ((emacs (24 3)) (s (1 9 0)) (dash (2 9 0)) (pcache (0 4 1)) (logito (0 1)) (marshal (0 6 3))) nil nil "/Users/sean/.emacs.d/elpa/gh-20160728.1525" nil nil])
  package-activate(gh (0 9 1))
  #[0 "\300\302H    >\204\303\304\305\306#\210\300\307H\211\2050\211@\310
@\211A@\262
\"\204)\311\312\"\210
A\266\202\202\207" [[cl-struct-package-desc magit-gh-pulls (20160513 310) "GitHub pull requests extension for Magit" ((emacs (24)) (gh (0 9 1)) (magit (2 1 0)) (pcache (0 2 3)) (s (1 6 1))) nil nil "/Users/sean/.emacs.d/elpa/magit-gh-pulls-20160513.310" ((:keywords "git" "tools") (:url . "https://github.com/sigma/magit-gh-pulls")) nil] cl-struct-package-desc-tags 0 error "%s accessing a non-%s" package-desc-reqs package-desc 4 package-activate throw dep-failure] 6 "\n\n(fn)"]()
  funcall(#[0 "\300\302H    >\204\303\304\305\306#\210\300\307H\211\2050\211@\310
@\211A@\262
\"\204)\311\312\"\210
A\266\202\202\207" [[cl-struct-package-desc magit-gh-pulls (20160513 310) "GitHub pull requests extension for Magit" ((emacs (24)) (gh (0 9 1)) (magit (2 1 0)) (pcache (0 2 3)) (s (1 6 1))) nil nil "/Users/sean/.emacs.d/elpa/magit-gh-pulls-20160513.310" ((:keywords "git" "tools") (:url . "https://github.com/sigma/magit-gh-pulls")) nil] cl-struct-package-desc-tags 0 error "%s accessing a non-%s" package-desc-reqs package-desc 4 package-activate throw dep-failure] 6 "\n\n(fn)"])
  package-activate(magit-gh-pulls)
  package-initialize()
  (progn (toggle-debug-on-error) (require (quote package)) (message "initialize") (package-initialize) (message "done"))
  eval((progn (toggle-debug-on-error) (require (quote package)) (message "initialize") (package-initialize) (message "done")))
  command-line-1(("--eval" "(progn (toggle-debug-on-error) (require 'package) (message \"initialize\") (package-initialize) (message \"done\"))"))
  command-line()
  normal-top-level()

I'm not really sure where to post this, but it doesn't happen when marshal is installed but does happen when gh is installed.

Despite the package-activate(magit-gh-pulls) (????), magit-gh-pulls isn't necessary.

[sigma]

unable to reproduce, and that doesn't look gh.el-related. It seems that the problem is really in the eieio layer. That might be related to some outdated bytecode lying around (usually is)