NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/lang/c++-part1/5520

Internet Message Format | 1996-08-05 | 5.8 KB

Path: dawn.mmm.com!news
From: [e-mail] (Kevin J Hopps)
Newsgroups: comp.lang.c++
Subject: Re: How to handle error in constructor
Date: 5 Feb 1996 13:49:56 GMT
Organization: 3M - St. Paul, MN  55144-1000 US
Message-ID: <4f51u4$[e-mail]>
References: <[e-mail]> <4enbts$[e-mail]>
Reply-To: [e-mail]
X-Newsreader: TIN [version 1.2 PL2]
X-Newsreader: TIN [version 1.2 PL2]
X-Newsreader: TIN [version 1.2 PL2]
X-Newsreader: TIN [version 1.2 PL2]

Walt Howard ([e-mail]) wrote:
> On Tue, 30 Jan 1996 00:54:21 GMT, [e-mail] (Andrew
> Phillips) wrote:

> >I'm converting and enhancing a simple program in C to C++.  Without going
> >into too much detail I have a class that represents a file on disk.  The
> >contructor opens the file, but what should it do if the file cannot be
> >opened?  The C program just calls fopen() tests the return value and
> >if there's an error it prints a message and exits.
> >
> >In C++ I thought to set a flag in the constructor and have a member function
> >that tests the flag to see if the file was opened correctly.  This seems
> >rather inelegant -- I guess exceptions would be the elegant way but seem
> >like overkill for such a simple program.  I've looked in several C++ books but
> >error handling is given scant coverage except for exception handling.
> >
> >Any suggestions would be greatly appreciated.
> >

> There are several approaches. The one you came up with is similar to
> what the C++ stream library does. However, that was written before
> exceptions were available.

> Exceptions is the only way to go on this. And you shouldn't consider

The only way to view statements about "the only way" is with skepticism :-)

> them overkill, they really aren't. All you have to say is, for
> example:

> try
> {
>     if ( action() == fail )
>         throw( "dang file didn't open" );
>             //Here I "throw" a char* pointer
> }
> catch( char* errorMessage ) // and here I catch it and deal with it
> {
>     printf( "%s", errorMessage );
>     exit( -1 ); // or whatever
> }

It is unlikely that the catch statement would be in the constructor.
More than likely the catch would be in a piece of code that would
do something about the problem.

> Try it a few times and you'll see how simple it is. The books make it
> seem complicated but it isn't. The rewards are well worth it.

Throwing exceptions is easy.  Writing code that behaves correctly
when exceptions are thrown is not always so easy.  A recent issue
of C++ Report discussed many of these potential problems.

> Exceptions will reliably trap and clean up a bad construction and I
> recommend using them as such, however there are some other issues
> involved.

Exceptions do not clean up anything.  The "stack unwinding" that
is done in propagating an exception from its throw point to its
handler will call the destructors of any automatic (stack based)
objects that have been fully constructed since entry into the
catch's try block.  Any clean up that is done must be done in
those destructors.

> From my experience, I would say that any class whose constructor has
> ANY chance to fail, should not do the failure-possible action during
> the constructor. Have a function called init(void) (always void
> argument) that does the actual activation of the object.

> Have the constructor do EVERYTHING necessary to prime/lock and load
> the object, but the init( ) function actually pulls the trigger. The
> init() function should just have (void) as an argument because all its
> going to do is pull the trigger right?

> For a file object, the construction might set the name and file access
> mode but the init() function would actually OPEN the file.

Dividing the initialization into two parts like this is a holdover
from the days before exceptions.  This approach allowed you to
get a return status from object initialization (via the init() func)
while the constructor provided no direct error reporting means.

Doing things in this way means that when implementing the class,
you need to keep track of whether init was called.  If one follows
the convention that the constructor throws an exception if it cannot
complete successfully, then the other member functions do not need
to verify that initialization was performed.  (Otherwise there would
be no object on which to call the member function.)

Using the construct/init separation also makes it more difficult
to use the class as a member of another class.  If I do this, now
either I have to use the construct/init separation, or I have to
call the init functions of all my member objects that use the
construct/init mechanism.

Things get even worse with multiple inheritance.  And with virtual
inheritance you have to be careful that the init() functions of
common base classes are not called more than once.

> This has many benefits. For example, if you're object's constructor
> can fail, GLOBAL objects (those constructed at startup time) can fail
> and you can't trap that with an try/catch exception handler. Way
> bummer.

This is indeed a problem.  Global objects and exception-throwing
constructors are not a good mix.  However, global objects sometimes are
a problem because of order-of-initialization dependencies, and might be
avoided for a couple of reasons.  One alternative to global objects is
to provide global or public static functions that construct if necessary
and return a reference.  This provides a way to catch exceptions and
avoid order-of-initialization problems.
--
Kevin J. Hopps                  e-mail: [e-mail]
3M Company                      phone:  [phone]
3M Center, Bldg. 235-2D-57      fax:    [phone]
St. Paul, MN 55144-1000         Opinions are my own.  I don't speak for 3M.
    But 3M speaks for me -- I did not write the following line:

Opinions expressed herein are my own and may not represent those of 3M.