mocl crashes when printing a float

ID	Created	Type	mocl Version	Dev OS	Target	Status
#52	by simon on 2014-08-04 (1113 days ago)	defect	14.05	Mac OS X	iOS	closed

Description

Summary:

mocl crashes when trying to print a float (or maybe double). It seems to crash every time when testing at the REPL. It crashes sometimes when using FORMAT from source code, as in calling the function TEST defined below. 

Steps to Reproduce:

In a lisp source file:

(defun magnitude-sqr (sequence)
  "Returns the squared magnitude of the sequence."
  (reduce #'+ sequence :key (lambda (num) (* num num))))

(defun magnitude (sequence)
  "Returns the magnitude of the sequence."
  (sqrt (magnitude-sqr sequence)))

(defun dot-product (a b)
  (reduce #'+ (map 'list #'* a b)))

(defun vector-min-angle (a b)
  (acos (/ (dot-product a b)
		   (* (magnitude a) (magnitude b)))))

(declaim (call-in test))
(defun test ()
  (let* ((a '(0.2349514788541595 0.9720071000688439 0.0))
		 (b '(0.2349514788541595 0.9720071000688439 0.0))
		 (angle (vector-min-angle a b)))
	(format t "~A~%" angle)))

Compile with mocl. Run on simulator or device.

In REPL: 

🔸  (setq a '(0.2349514788541595 0.9720071000688439 0.0))
(0.23495148 0.9720071 0.0)
🔸  (setq b '(0.2349514788541595 0.9720071000688439 0.0))
(0.23495148 0.9720071 0.0)
🔸  (vector-min-angle a b)   
Error: end of file on #<SB-SYS:FD-STREAM for "socket 192.168.1.102:54355, peer: 192.168.1.102:2222" {1007C6CFE3}>

Expected Results:

0.0

Actual Results:

mocl crashes.

Notes:

In the Xcode output window the following message appears:

Error: 1.0E  in write-internal
  in print-object--method( t t )
  in print-object--subfun( sub382 )
  in print-object--method( simple-condition t )
  in print-object--subfun( sub364 )
  in write-internal
  in print-object--method( t t )
  in print-object--subfun( sub382 )

Using mocl from git, commit 258c6709f118470a8e8f153047bb0b01b0008f72.

Comments

The real problem is not printing the float, is that VECTOR-MIN-ANGLE returns a value that seems to be incorrect. Upon trying to debug by FORMATing mocl crashes.

- simon, 1113 days ago

It appears that, due to ordinary float round-off issues, the argument to acos inside vector-min-angle is equal to (+ 1.0 double-float-epsilon) rather than exactly 1.0. For arguments greater than 1, acos should technically return a complex number; but, mocl does not support the ANSI complex number type, so it returns float NaN, which crashes the printer.

I've updated the mocl printer to print NaN, crash-free, in commit fdebed8.
NaN, +Infinity, and -Infinity will print as #<float NaN>, #<float +Infinity>, and #<float -Infinity>, respectively. I might add some additional stuff like error signaling for acos arguments > 1 in later commits but that's it for now.

You will probably want to round 'slightly above 1' to 'actually 1' in your calculation somehow, e.g. by (acos (min 1.0 FOO)).

Hope that all sounds reasonable... let me know if you have any questions/comments.

Wes

- Wukix, 1113 days ago

This works for me.

Thank you very much.

- simon, 1113 days ago