Marcel Holtmann wrote:

Hi Pedro,

In the application I am developing, I want to log some information about 
Bluetooth devices, pretty much like "sdptool browse" and "hcitool info" 
do. In the code of these applications, there are calls to some library 
functions like hci_dflagstostr(), hci_lmtostr()... that create a string 
with information about the device (flags, features, etc.).

The problem with these functions is that they call malloc(), expecting 
the caller to call free(). The problem with that is that this approach 
is not save, as it is not guaranteed that calling malloc() in a library 
function and free() in the main program works. In fact, I have seen a 
segmentation fault doing that.


explain why this is not safe.

I am not an expert in these issues, but I have read many times that there is no guarantee that the malloc/free implementation is the same in the main program and in the libraries. This is due to how the different modules are compiled: compiler version, compile flags, etc. But, as I said, I am not 100% sure. I am quite sure that I have read this for Windows, but maybe (probably) Linux is different.

AFAIK, there are two alternatives: either the library functions expect a 
buffer (and its size) to be filled or malloc() and free() are both done 
within the library, which means creating a new library function call for 
freeing buffers allocated with malloc(). Are any of this alternatives 
implemented or on the roadmap? What workaround do you suggest?


We can do that, but it is not on my roadmap. Send my patches for it and
I will review them.

The easiest way to do it (and to continue with the actual API without modifications) would be to have a function call like:

void hci_str_free(char *str)
{
free(str);
}

So after a call to hci_dflagstostr() (for example), the resulting pointer should be freed with hci_str_free(). In any case, there is one exception: hci_dtypetostr(); it returns a pointer to static data.

Cheers,

Pedro