Issues using Perl exit

[chrstphrchvz]

Various issues (errors, crashes, segmentation faults) are encountered when Perl exit is called before destroying the main/root window. Examples include:

• using exit in callbacks (see https://rt.cpan.org/Ticket/Display.html?id=128654)
• not destroying root toplevel ("main window") before the implicit exit at the end of a script, as was often done in test scripts which never called MainLoop (see https://rt.cpan.org/Ticket/Display.html?id=128804)

Tcl::Tk appears to be affected by this as well (https://www.perlmonks.org/?node_id=1191134). I do not know whether wrappers can/should support exit usage, so for now I recommend always destroying the root toplevel before Perl exits (either by entering MainLoop and waiting for the user to close it, or programmatically using the destroy method).

[chrstphrchvz]

I looked into this a while ago. Perl’s exit relies on C’s longjmp(), which is why perlinterp says “code should always save values using SAVE_FOO rather than in auto variables”. But Tcl sees no problem with keeping its state on the C stack. So embedding Tcl seems to violate Perl’s assumption that XS or C code called from Perl won’t keep important state on the C stack, because it will be clobbered by longjmp(). Once Perl exit has done longjmp(), it is too late for Tcl.xs to call Tcl_Finalize(). I have yet to find if there is a reasonable way to override Perl’s exit so that Tcl_Finalize() can be done before longjmp().