Declaimed Optional Types Lead to Bad Type Inference

Minimal test case:

(declaim (ftype (function (real &optional real) real) foo))
(defun foo (a &optional b)
  (cond (b (+ a b))
        (T (+ a 1))))

This only works when compiled and then loaded, so save to file, (load (compile-file "test.lisp")). Calling the function like (foo 1) then results in an error:

NIL is not of type REAL.
   [Condition of type TYPE-ERROR]

Restarts:
 0: [RETRY] Retry SLIME REPL evaluation request.
 1: [*PROCESS-INPUT] Continue reading input.
 2: [ABORT] Return to SLIME's top level.
 3: [CLOSE-CONNECTION] Close SLIME connection.
 4: [RESTART-TOPLEVEL] Go back to Top-Level REPL.

Backtrace:
  0: DEBUG-IN-EMACS
  1: INVOKE-SLIME-DEBUGGER
  2: G992
  3: CALL-WITH-DEBUGGER-HOOK
  4: SWANK-DEBUGGER-HOOK
  5: FOO
  6: #<bytecompiled-function BYTECODES>
  7: EVAL-REGION
  8: G102
  9: TRACK-PACKAGE
 10: G101
 11: CALL-WITH-RETRY-RESTART
 12: G100
 13: CALL-WITH-BUFFER-SYNTAX
 14: REPL-EVAL
 15: LISTENER-EVAL
 16: #<bytecompiled-function BYTECODES>
 17: EVAL-FOR-EMACS
 18: PROCESS-REQUESTS
 19: G372
 20: HANDLE-REQUESTS
 21: G651
 22: CALL-WITH-BINDINGS
 23: REPL-INPUT-STREAM-READ
 24: G643
 25: G226
 26: CALL-WITH-LOCK-HELD
 27: G225
 28: SIMPLE-REPL
 29: G594
 30: G585
 31: CALL-WITH-USER-BREAK-HANDLER
 32: G582
 33: CALL-WITH-DEBUGGER-HOOK
 34: G626
 35: CALL-WITH-BINDINGS
 36: SIMPLE-SERVE-REQUESTS
 37: SERVE-REQUESTS
 38: ACCEPT-CONNECTIONS
 39: SERVE
 40: SERVE-LOOP
 41: SETUP-SERVER
 42: START-SERVER
 43: #<bytecompiled-function BYTECODES>
 --more--

So most likely ECL just doesn't augment the proclaimed optional type with (or null ..).

Changed title: Declaimed Optional Types Leads to Bad Type Inference → Declaimed Optional Types Lead to Bad Type Inference

confirmed

Added ansi-standard bug labels

Hmm, thinking about this further, I'm not actually quite sure how the type information is supposed to be handled for optional and keyword arguments when they default to a value. Does the default value have to be of the proclaimed type or can it diverge? Since the default value isn't really part of the function interface, I don't know how that should be handled in the presence of function type declarations.

Wouldn't know where to look in the CLHS for this either, so it would probably be wise to consult someone like Robert Strandh on this.

changed milestone to %ECL 16.2.0