Re: [libmicrohttpd] Memory leak?

[Erik Slagter]

Hi Christian,

> Based on your code, I can clearly see that the callback is called, but
> obviously with the wrong address. However, the MHD code internally clearly
> uses the right address, so the best explanation for the difference would be
> some kind of memory corruption going on in between.  Have you tried valgrind
> on your code?  Again, if you can produce a stand-alone *compiling* example
> where the issue still arises, I'll be happy to take another look, but so far
> this seems to be most likely a bug in the rest of your C++ code.  The way you
> use the API (from what I can see) seems to be fine.

It's interesting that you mention valgrind, because that is the tool I 
found the memory leak with in the first place. So yes, the program has 
been running under valgrind control, with almost all options enabled. 
After running for a few hours, valgrinds concludes that a few little 
blocks of memory are lost inside ncurses (so out of my control) and that 
a huge pile of memory is lost that is allocated inside the 
access_handler_callback, allocated exactly at this line:

ConnectionData * ncd = new(ConnectionData)

And that is exactly the memory that is not free'd because of "wrong 
pointer" problem.

Valgrind does not mention other memory loss or memory corruption.

It strikes me that the pointer I'm receiving in the completed callback 
seems to be a valid pointer; it's address is very similar to the address 
that's actually allocated. Also the pointer always points to a null 
pointer, which suggests the pointer is allocated as should be, no stray 
pointer.

To me it looks like libmicrohttpd allocates a new pointer at some point 
while discarding the current pointer. I haven't had a look at the code, 
so this may be uterly nonsense I'm talking.

My planning is to make very stripped down version of the program, test 
that, and if the problem persists, I'll post the source.

smime.p7s
Description: S/MIME Cryptographic Signature