Hello,

After doing some more tests, the use case I described yesterday is actually not producing the libc assert. I thought that the cleanup set the default memory manager back, but it's not (which in fact seems ok to me).

I guess the fix I suggested is still legit, but for information, here is the use case producing the assert :
1) curl_global_init_mem() // initialization with custom memory manager
2) curl_global_init_mem()
3) curl_global_cleanup() // cleanup
4) curl_multi_init() // initialization with default memory manager
5) curl_global_init_mem()
6) curl_global_cleanup() // cleanup
7) curl_global_cleanup()

I didn't know where the default memory managers were set back, and I just discovered the step 4 in my code and that it was calling curl_global_init().

Sorry for the wrong use case from yesterday

---
** [bugs:#1362] Increment the initialized counter in curl_global_init_mem()**
**Status:** open
**Labels:** curl_global_init_mem 
**Created:** Thu Apr 17, 2014 09:28 PM UTC by Tristan
**Last Updated:** Thu Apr 17, 2014 09:28 PM UTC
**Owner:** nobody
Hello,
I think there is an issue in easy.c, in the curl_global_init_mem() function
The initialized counter is not incremented if the libcurl is already initialized.
I'm currently doing two inits and two shutdowns, and here is the problem I met :
1) curl_global_init_mem(); // Actually initialize libcurl (initialized = 1)
2) curl_global_init_mem(); // Should do nothing (initialized stays to 1 instead of being incremented)
3) curl_global_cleanup(); // Actually cleanup libcurl because initialized = 1 
4) curl_global_cleanup();
After step 3 the variables that were created using the custom malloc are released using the default free, which raises an assert by libc on non-matching heap :
Invalid address specified to RtlValidateHeap
I could fix that by modifying easy.c the following way:
    @@ -300,8 +300,10 @@
         return CURLE_FAILED_INIT;
    
       /* Already initialized, don't do it again */
    -  if(initialized)
    +  if(initialized) {
    +    initialized++;
         return CURLE_OK;
    +  }
    
       /* Call the actual init function first */
       code = curl_global_init(flags);
However, maybe there is a better fix than duplicating the increment from curl_global_init()?
I'm using Libcurl 7.34.0, and checked quickly for a fix in 7.36.0, but I didn't find it.
---
Sent from sourceforge.net because curl-tracker@cool.haxx.se is subscribed to https://sourceforge.net/p/curl/bugs/
To unsubscribe from further messages, a project admin can change settings at https://sourceforge.net/p/curl/admin/bugs/options.  Or, if this is a mailing list, you can unsubscribe from the mailing list.