NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/std/c/136

Internet Message Format | 1996-08-06 | 2.8 KB

Path: lyra.csx.cam.ac.uk!nmm1
From: [e-mail] (Nick Maclaren)
Newsgroups: comp.std.c
Subject: Re: atexit() and return from main()
Date: 18 Jan 1996 08:15:48 GMT
Organization: University of Cambridge, England
Message-ID: <4dkvjk$[e-mail]>
References: <4dhfja$[e-mail]> <4diicg$[e-mail]> <[e-mail]>
NNTP-Posting-Host: bootes.cus.cam.ac.uk

In article <[e-mail]>,
Lawrence Kirby  <[e-mail]> wrote:
>In article <4diicg$[e-mail]>
>           [e-mail] "Nick Maclaren" writes:
>
>>In article <4dhfja$[e-mail]>, [e-mail]
>> (Thomas Koenig) writes:
>>|> I don't see a problem when I replace the 'return 0' above with
>>|> an 'exit(0)'.
>>
>>I do.  exit() may unwind the stack before calling the atexit()
>>functions, and that is actually the way that most C run-time
>>systems implement close down.
>
>Implementations that do that are non-conforming. According to 7.10.4.3:
>
>"First, all functions registered by the atexit function are called, in
> the reverse order of registration."
>
>There is no license given to destroy any existing objects before that happens.

Hang on.  The question was whether there was likely to be a problem.
There is.  The fact that such implementations are non-conformant is
not a denial of that fact.

I can remember discussing this very point at great length with other
people, and we came to the conclusion that the standard was confusing
at best.  The clarification mentioned by Clive Feather makes it much
clearer that exit() is NOT supposed to unwind the stack before calling
the atexit() functions.

But you are making a serious mistake in assuming that run-time systems
can be implemented to allow everything that the standard permits by
omission.  They can't.  In THIS case, they can usually preserve the
stack before calling the atexit() functions.


However, for the advanced paper, think about the following problems:

What is the effect of an atexit() function calling exit()?  Does it
restart the atexit() queue, carry on as if the function had returned
or does it abandon the atexit() queue?

I chose to do the second, as by far the most user-friendly, but I
would NEVER write serious code that relied upon that action!

main() calls setjmp() and then calls exit().  An atexit() function
calls longjmp() and returns to main().  This repeats (or does it?)

This is ALSO permitted by omission, but a good run-time system will
trap it, issue an error message and stop without calling more atexit()
functions.


Nick Maclaren,
University of Cambridge Computer Laboratory,
New Museums Site, Pembroke Street, Cambridge CB2 3QG, England.
Email:  [e-mail]
Tel.:  [phone]    Fax:  [phone]