Hello Thomas, Sorry for the delay, I had some problems with my mail lately,
> The trouble with reverting my commit is that readding the double free > completely > crashes ed if more than one join is performed. I think this patch (which also > reverts > back to having no double free) should handle your concern via blocking signal > handling > until the join is finished. Ok, I understand. Mixing longjmp signals and dynamic memory was a really bad idea. Your idea is not bad, but I think we should block sigprocmask and related functions, instead of playing with variables. Regards,
