autark makes the excellent suggestion to use $SIG{__DIE__}. The die signal, AKA sigdie, can be caught and handled by a user defined function.
This is done by adding the handler to the %SIG associative array. (read %SIG)
I've spent plenty of time playing with this device, and I thought
I might take a moment to point out a few of the pitfalls. Many of these
are mentioned in the documentation for die, warn, eval, $^S,
and for %SIG. (The last two are parts of perlvar.)
So then, on to the pitfalls.
Pitfall 1: While eval traps the fatality of sigdie, it does not
block the $SIG{__DIE__} handler from being called.
Solution 1: If you don't want your handler to be used by any
eval block, then check $^S at the begining.
Solution 2: If you have a particular eval block where you don't want the
handler to be invoked, then turn it off using local:
sub Something
{
local $@; # Don't step on other code.
# try
eval{
local $SIG{__DIE__}; # No sigdie handler
# do stuff that might die
}
# catch
HandleEvalError( $@ ) if $@;
}
Pitfall 2: The sigdie handler can be set at any time, including during compilation (via a BEGIN block).
If you set the handler during a BEGIN block and complitation fails before compiling your subroutine, you not only
get a function not found error, but you lose the information about the compilation error.
Solution 1: Declare your handler subroutine before setting the handler.
|
<Note>It occurs to me that you might not have realized that you can do this.
Here is an example of to what I am refering. The solution is to move the subroutine
up in your code, before the BEGIN block:
BEGIN{ $SIG{__DIE__} = \&FatalErr }
# Real problem if compile fails before getting here.
sub FatalErr { # do stuff, maybe print @_ or something.
}
Why does moving the sub up work? Well... BEGIN is actually just
another sub routine, except that it gets called as soon as its compiled.
INIT, by the way, is also a special sub routine, except that it gets called
as the last step of the compilation phase. </Note> |
Solution 2: Don't set your handler in a BEGIN block like that. If you want the
handle available after compilation is complete, but before code runs, then use an INIT block.
Pitfall 3: Not so much a pitfall as it is a reminder: Functions called with an ampersand
and no argument list are passed the current values of @_. This means (as Plaid once pointed out to me)
that your sigdie handler gets the same argument list that die got. One thing to note is that the file/line number
appending has already taken place by the time your handler gets @_.
Pitfall 4: When your handler is finished, Perl continues dieing. This means that not only does the
execution end, but Perl prints the error message to STDERR.
Solution: You might not have thought of that as a pitfall, but I certainly do. Sometimes you want to include
a lot of information about the failure. Maybe write it to a file or something, and then make the error message
the user sees be something like, "An error occurred. See log, '$log'\n". The trick here is
to keep in mind that Perl is about to exit anyway, so why not accelerate the process? BTW Don't do this in
your signal handler unless you are also checking $^S. People get mad
when eval blocks don't return.
Pitfall 5: Your handler only gets called if you use die. This means that if you aren't checking your
system calls and using die like C's Assert, then you won't get any added functionality.
Solution: die. Everywhere.
I hope this helps you out.
Please do read the documentation sited above.
Also, don't forget the power of an END block.
A few of my past posts on the subject:
Fun with $SIG{__DIE__}
And yet more $SIG{__DIE__} fun...
END failed--call queue aborted
| [reply] [d/l] [select] |
| [reply] |
Just to clarify things... when anybody executes a die the program is basically done. Even if you wrap an eval around it the script is still going to terminate?
The reason I am asking is because a module that somebody at work wrote executes a die when there is a request for data that doesn't exist. In the program he has used this for that might be a valid path of execution, but in the program I have written I can't just terminate. I was planning on wrapping an eval around the function call to trap it and more or less ignore it because for me it isn't fatal, but maybe eval won't help?
Just curious, because I'd rather not have to ask this guy to change his code and just deal with the problem myself. This man scares me to death, hehe ;)
| [reply] |
The basic rule is that if die is called, your program terminates. As with all rules, this one has exceptions. eval is like a separate program. The eval call will terminate, but will not end your program.
However, it WILL call any handler you have defined in $SIG{__DIE__}. This is something you must keep in mind when writing a sigdie handler. Either check $^S, or don't plan on actually terminating. A large part of my post concerned the pitfalls surounding this point.
| [reply] |
The $@ variable is only set by the eval code if it somehow fails.
In your last example, you don't use eval to "catch" this die, so $@
will not be set.
However, even if you used eval, this will not work for your END-block. You might
want to try to install a $SIG{__DIE__} handler instead. Using that to
set a variable or something similar, so that you can access the information within
the END-block.
$SIG{__DIE__} = sub { our @reason = @_ };
eval { die "Let's Try to die." };
END {
our @reason;
if (@reason) {
print "We caught a call to die: @reason\n";
}
}
Autark. | [reply] [d/l] [select] |
| [reply] [d/l] [select] |
Yes -- but that's only because it's a Perl-generated signal that never touches the OS. So Windows' lack of signal support doesn't even get brought into play.
perl -e 'print "Just use $^X$\"$]!$/"'
| [reply] |
A more complete solution if you wish to trap errors in OO-ish
way is Graham Barr's
Error.
module.
Although I am not a big fan of throw, try and catch (personal illogical old fart
reasons). I have used it where eval... if $@ just
wouldn't do and it was relativley painless.
mitd-Made in the Dark
'My favourite colour appears to be grey.' | [reply] [d/l] |