RE: [ECOS] malloc vs. new

[Scott Dattalo <scott at dattalo dot com>]

On Sat, 22 Jun 2002, Dan Conti wrote:

> new shouldn't be broken, since it just calls malloc. does malloc work
> when making allocations of the exact same size?

malloc works fine but new fails. I did a real simple case where I malloc'd 
75 bytes and new'd a 75 byte char array.

> 
> two things to try as workarounds:
> 
> 1) write your own operator new/new[]/delete/delete[]. tossing the
> following into some file should work:

I thought about this, but when you overload new and delete the constructor 
is no longer called (actually, that's an assumption on my part). So I'll 
have to add special code to initialize the objects as they're new'd.

> 2) use placement new, which is what you were getting at below. be wary
> of alignment considerations for the buffer you use. also at clean up you
> have to manually call the destructor IIRC. C++ is not my forte.
> 
> #include <new.h>
> 
> char buffer[1024] __attribute__((align(4));
> 
> CObject* pObj = new(buffer) CObject();
> 
> ...
> pObj->~CObject();
> 
> Good luck.

Thanks for the help. Little tid-bits about alignment and such are good to 
know.

I'm going to dig a little deeper with the debugger and determine exactly 
the mode of failure. new'ing is such a fundamental thing that I'd be 
extremely surprised if it didn't work in an eCos world. While I could work 
around new (with malloc for example), I don't wish to artificially 
constrain other developers that will eventually be working on the code 
base that I'm creating. OTOH, this *is* an embedded application and most 
of the desktop assumptions (like infinite memory) are invalid!

Scott


-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss