std::map::find() throws exception when map is empty?

[Matthias Hildebrand]

Hello,

std::map< int, MyClass*> mymap;

it = mymap.find( somekey ) // here happen bad things
if( it != mymap.end() )
{
// do something useful with it->second
}
else
{
// do nothing
}

I traced a strange program crash (unhandled exception on windows,
something similar on Solaris) down to happen at the second line of
the code given above which contains the find(). After some
investigation I found out that it happens when the map is empty.

This behaviour could be reproduced on Windows 2k and XP using MSVC 6.0
and 7.0 and finally also on Solaris unsing the Sun CC 5.4.

I read the documentation as well as books and internet resources but I
could not find a single statement that on an empty map a find() must
not be performed.

Could anybody please give me some information whether this is correct
or at least well known behaviour? Did I miss something? I do not like
the idea of preceeding all "critical" find() with some statement like
if( ! mymap.empty() )...

Any comments would be appreciated! Thanks!

Matthias

[Victor Bazarov]

"Matthias Hildebrand" <hi********@uni-kassel.de> wrote...

std::map< int, MyClass*> mymap;

it = mymap.find( somekey ) // here happen bad things
if( it != mymap.end() )
{
// do something useful with it->second
}
else
{
// do nothing
}

I traced a strange program crash (unhandled exception on windows,
something similar on Solaris) down to happen at the second line of
the code given above which contains the find(). After some
investigation I found out that it happens when the map is empty.

This behaviour could be reproduced on Windows 2k and XP using MSVC 6.0
and 7.0 and finally also on Solaris unsing the Sun CC 5.4.

I read the documentation as well as books and internet resources but I
could not find a single statement that on an empty map a find() must
not be performed.

Could anybody please give me some information whether this is correct
or at least well known behaviour? Did I miss something? I do not like
the idea of preceeding all "critical" find() with some statement like
if( ! mymap.empty() )...

No, this is not correct. As to whether it's a known behaviour, you
have to check with those who make the library for your compiler.

Although it has no exception specification, 'find' usually does not
throw by itself. The reason there is no exception specification is
that 'find' uses the comparison function for keys, that may throw.

I understand your frustration about having to use '.empty()' before
'find', but it may be the only work-around for your particular version
of the library. Contact the library vendor and let them know about
your trouble. They should fix it.

Victor

[lilburne]

Matthias Hildebrand wrote:

Could anybody please give me some information whether this is correct
or at least well known behaviour? Did I miss something? I do not like
the idea of preceeding all "critical" find() with some statement like
if( ! mymap.empty() )...

Any comments would be appreciated! Thanks!

What was the exception, who threw it, and why?

[Bob Bell]

hi********@uni-kassel.de (Matthias Hildebrand) wrote in message news:<65**************************@posting.google. com>...

Hello,

std::map< int, MyClass*> mymap;

it = mymap.find( somekey ) // here happen bad things
if( it != mymap.end() )
{
// do something useful with it->second
}
else
{
// do nothing
}

I traced a strange program crash (unhandled exception on windows,
something similar on Solaris) down to happen at the second line of
the code given above which contains the find(). After some
investigation I found out that it happens when the map is empty.

This behaviour could be reproduced on Windows 2k and XP using MSVC 6.0
and 7.0 and finally also on Solaris unsing the Sun CC 5.4.

I read the documentation as well as books and internet resources but I
could not find a single statement that on an empty map a find() must
not be performed.

Could anybody please give me some information whether this is correct
or at least well known behaviour? Did I miss something? I do not like
the idea of preceeding all "critical" find() with some statement like
if( ! mymap.empty() )...

This is not correct. It is perfectly permissible to use find with an
empty map. The bug must be elsewhere.

Bob

[Matthias Hildebrand]

Hello!

std::map< int, MyClass*> mymap;

it = mymap.find( somekey ) // here happen bad things
if( it != mymap.end() )
{
// do something useful with it->second
}
else
{
// do nothing
}

I traced a strange program crash (unhandled exception on windows,
something similar on Solaris) down to happen at the second line of
the code given above which contains the find(). After some
investigation I found out that it happens when the map is empty.
I stepped through some STL code and found out that the cause is a
simple dereferenced NULL pointer within the STL... In short and
simplified words: find is looking for a node close neighbored to the
given key, which results in a NULL pointer because the map is empty,
and then the STL wants to access the parent of this node by
dereferencing the pointer.

So the crash can only be avoided by not doing "find" on empty maps...

This behaviour could be reproduced on Windows 2k and XP using MSVC 6.0
and 7.0 and finally also on Solaris unsing the Sun CC 5.4.

The STL I stepped through was that contained with MSVC 7.0, but as
already mentioned, the same behaviour I experienced also with MSVC 6.0
and Sun CC 5.4.

Since several library implementations seem to share this particular
characteristic I would have expeted that it is well known or at least
documented somewhere but concluding from he postings here I think it
is not... :-(
This is not correct. It is perfectly permissible to use find with an
empty map. The bug must be elsewhere.

Bob

I dare say I would prefer the bug being elsewhere, because this
behaviour is annyoing for me. Would you have a suggestion what bug
outside the STL code could create the behaviour described above?
Thanks!

(Also thanks to all others who replied!)

Matthias