Odd backtrace in condition-case

Odd backtrace in condition-case

[Lars Ingebrigtsen]

I usually run with `debug-on-error' set, but lately I've started getting
these backtraces:

Debugger entered--Lisp error: (wrong-type-argument char-or-string-p nil)
  capitalize(nil)
  timezone-make-arpa-date(2000 0 0 "01:00:00" (3600 "CET"))
  timezone-make-date-arpa-standard("Session")
  date-to-time("Session")
  url-cookie-expired-p([url-cookie "NRK-Experiment-Bucket" "0" "Session" "/" "www.nrk.no" nil])

[...]

  url-http-create-request()
  url-http-async-sentinel(#<process www.nrk.no> "open\n")

The code is:

(defun url-cookie-expired-p (cookie)
  "Return non-nil if COOKIE is expired."
  (let ((exp (url-cookie-expires cookie)))
    (and (> (length exp) 0)
	 (condition-case ()
	     (time-less-p (date-to-time exp) nil)
	   (error nil)))))

My question is: Why does this give me a backtrace?

If I say something else, like:

(condition-case ()
    (capitalize nil)
  (error t))

=> t

I get no backtrace, but:

(condition-case ()
    (date-to-time "Session")
  (error t))

=>
Debugger entered--Lisp error: (wrong-type-argument char-or-string-p nil)
  capitalize(nil)
  timezone-make-arpa-date(2000 0 0 "01:00:00" (3600 "CET"))
  timezone-make-date-arpa-standard("Session")
  date-to-time("Session")
  (condition-case nil (date-to-time "Session") (error nil))

*scratches head*

Shouldn't the `error' clause in condition-case catch all errors?

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Re: Odd backtrace in condition-case

[Lars Ingebrigtsen]

Lars Ingebrigtsen <larsi@gnus.org> writes:

> (condition-case ()
>     (date-to-time "Session")
>   (error t))

Oh.  The function in question has:

	 (condition-case-unless-debug err
	     (encode-time (parse-time-string
			   (timezone-make-date-arpa-standard date)))

Checked in by Stefan in April.  Stefan, was that checked in by mistake?

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Re: Odd backtrace in condition-case

[Andy Moreton]

On Mon 20 May 2019, Lars Ingebrigtsen wrote:

> I usually run with `debug-on-error' set, but lately I've started getting
> these backtraces:

> Debugger entered--Lisp error: (wrong-type-argument char-or-string-p nil)
>   capitalize(nil)
>   timezone-make-arpa-date(2000 0 0 "01:00:00" (3600 "CET"))
>   timezone-make-date-arpa-standard("Session")
>   date-to-time("Session")
>   (condition-case nil (date-to-time "Session") (error nil))

> Shouldn't the `error' clause in condition-case catch all errors?

`date-to-time' has:

	 (condition-case-unless-debug err
	     (encode-time (parse-time-string
			   (timezone-make-date-arpa-standard date)))

As you have `debug-on-error' set, this looks like the expected
behaviour.

The `condition-case-unless-debug' was added by Stefan in commit
cb411c609d43 on April 19th. Perhaps this is left over debugging code ?

    AndyM

Re: Odd backtrace in condition-case

[Paul Eggert]

On 5/20/19 11:22 AM, Andy Moreton wrote:
> The `condition-case-unless-debug' was added by Stefan in commit
> cb411c609d43 on April 19th. Perhaps this is left over debugging code ?

Yes, that's evidently what happened. I just now patched master to change 
it back to plain 'condition-case'. The problem was partly my fault, 
since I introduced the date-to-time bug that Stefan fixed in April.

Re: Odd backtrace in condition-case

[Stefan Monnier]

> I usually run with `debug-on-error' set, but lately I've started getting
> these backtraces:
>
> Debugger entered--Lisp error: (wrong-type-argument char-or-string-p nil)
>   capitalize(nil)
>   timezone-make-arpa-date(2000 0 0 "01:00:00" (3600 "CET"))
>   timezone-make-date-arpa-standard("Session")
>   date-to-time("Session")
>   url-cookie-expired-p([url-cookie "NRK-Experiment-Bucket" "0" "Session" "/" "www.nrk.no" nil])

FWIW, I think it was useful for Emacs to show this backtrace (hence
validating the use of condition-case-unless-debug there ;-) because
it makes no sense for

    (date-to-time "Session")

to signal the error `(wrong-type-argument char-or-string-p nil)`

IOW, the above backtrace indicates a bug somewhere within date-to-time,
which should signal *another* error.


        Stefan