Re: deprecating SIGDIE

Front page | perl.perl5.porters | Postings from October 1999

Thread Previous | Thread Next

From:

Joshua N Pritikin

Date:

October 1, 1999 11:52

Subject:

Re: deprecating SIGDIE

Message ID:

19991001145239.I16864@eq1062.wks.na.deuba.com

On Fri, Oct 01, 1999 at 01:53:05PM -0400, ilya@math.ohio-state.edu wrote:
> Current situation: as Larry said, you do not need to protect against
> __DIE__.  __DIE__ handler *should not* step on other people's $@.

Agreed.

> However, this "should not" requires some additional work from the
> __DIE__ handler writers.  The above 'local' is in recognition of
> possibility of potentially buggy __DIE__.
> 
> Providing __EXCEPTION__ allows one 
> 
>   a) either document extremely strict requirements on the code;

It should be enforced.  Exceptions are too critical to leave to
(potentially unread!) documentation.

>   b) or make $@ local inside __EXCEPTION__, in pseudocode:
> 	  local $@ = $@;
> 	  $SIG{__EXCEPTION__}->();

Ugh.  That means die within __EXCEPTION__ is not reported except with:

  $SIG{__EXCEPTION__} = sub {
    # local $@=$@;  # Ilya's magic
    eval {
       # ...
       #dangerous stuff
       # ...
    };
    warn if $@;
  };

I guess that's workable but there is still the possibility of leaking
exceptions if the eval {} is forgotten or the "warn if $@" is forgotten.
I hate to allow the possibility of these leaks.  I much prefer

  exit("Exception throw inside $SIG{__EXCEPTION__}");

instead of an implicit "local $@=$@" (which should be outlawed!).  In
fact, I believe I submitted a patch over a year ago for a compile time
warning if $@ is localized.  In my opinion, exiting is far preferable to
any localization of $@.

This exit(...) can always be circumvented with carefully designed
eval{}, anyway.

-- 
"Jokes are not funny; doubt is funny."
    via, but not speaking for Deutsche Bank

Thread Previous | Thread Next