Am 22.07.2013 14:49, schrieb Massimo Manghi: > On 07/22/2013 01:20 PM, Harald Oehlmann wrote: >> - If so, Massimo could introduce in Rivet a call to Tcl_InitNotifier, >> eventually conditioned on patched tcl (say 8.5.14, 8.6.1) and no >> HAVE_PTHREAD_ATFORK... IMHO the call never harms, but it might be not >> necessary in most cases... > > Jan's approach is probably the wisest and we should simply look at the > problem from the Rivet's perspective: just call Tcl_InitNotifier > unconditionally. Either the notifier is started correctly or not it > doesn't matter. We could let 'configure' print a warning if the version > of Tcl is among those that doesn't handle the fork consistently and in > case say loud that fileevent handlers won't work. Did I get the picture? > > If it's all about adding that call I can do it tonight and do some tests.
Massimo, I would appreciate, if you could do it. If we have no answer on the question, if there are any platforms not having pthread_atfork(), we could just do it. On the strict rivet view, we still have an issue, which is perhaps not as important as I thought: 1) we have no solution for unpatched tcl (<8.5.15) threaded builds. or 2) the solution for those platforms would be to completly restructure rivet_mod.c and loose Rivet_Init_Proc (all names from memory). I would forget (2), as: - if people use a standard linux distribution, the builds are mostly unthreaded - if people compile tcl on their own, they may recompile tcl. - It only concerns Linux, MacOS has already a solution for this (who may test ?) Perhaps, we should modify the requirements on tcl: - threaded build on Unix, not MacOS: tcl8.6.1, tcl8.5.15 - no threaded build: as current Regards, Harald --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
