On Sat, 23 Aug 2008, Max L. Eidswick wrote:

> I reran the test with the call in the spawned thread (all init code in
> process init, not in the libcurl thread). It consumes roughly a 4k page
> each 7 thread hits. Move the test libcurl call out of the spawned thread
> and the leak is plugged. Remove the call to easy perform in the test
> thread, the leak is plugged.

Now this is getting curious. What can possibly differ depending on which
thread that does what?

Does the "all init code" include curl_easy_init() ? If so, can there possibly
some windows call we use in libcurl that can't be done in another thread to
free up resources if the init was done in a different thread?

-- 
  / daniel.haxx.se