On 27.07.2018 14:52, Yann Ylavic wrote: > On Thu, Jul 26, 2018 at 3:35 PM, Branko Čibej <br...@apache.org> wrote: >> We've had constant and painful problems with this in Subversion, having >> to track down random crashes at process exit. We've implemented really >> horrible workarounds, and I'm still not sure we've covered every edge case. > Supposedly still, this may be because of a DSO cleanup being called > after the DSO is unloaded (i.e. cleanup code already unmapped). > I'd blame the global pool for this, for not being the one you'd want > to destroy/cleanup at the right time.
Well of course, that's what I said. :) The problem is that we have no control over that ordering ... Subversion is also essentially a bunch of libraries and whilst things are pretty simple for our short-lived command-line tools, a long-lived GUI client linking our libraries has a rather different perspective on time. -- Brane