Talk about the goto() overhead in the documentation

[perl/modules/Scope-Upper.git] / lib / Scope / Upper.pm

diff --git a/lib/Scope/Upper.pm b/lib/Scope/Upper.pm
index ad3f339cf3ff37d933c0d590c9dc318cefd96361..eefe25362df85b9fdabf14a419e3351f9917434a 100644 (file)
--- a/lib/Scope/Upper.pm
+++ b/lib/Scope/Upper.pm
@@ -310,7 +310,7 @@ Both are identical, with the following caveats :
 =item *
 
 The L<Sub::Uplevel> implementation of C<uplevel> may execute a code reference in the context of B<any> upper stack frame.
-The L<Scope::Upper> version only allows to uplevel to a B<subroutine> stack frame, and will croak if you try to target an C<eval> or a format.
+The L<Scope::Upper> version can only uplevel to a B<subroutine> stack frame, and will croak if you try to target an C<eval> or a format.
 
 =item *
 
@@ -520,6 +520,12 @@ However, it's possible to hook the end of the current scope compilation with L<B
 Some rare oddities may still happen when running inside the debugger.
 It may help to use a perl higher than 5.8.9 or 5.10.0, as they contain some context-related fixes.
 
+Calling C<goto> to replace an L</uplevel>'d code frame does not work when a custom runloop is used or when debugging flags are set with C<perl -D>.
+In those two cases, L</uplevel> will look for a C<goto &sub> statement in its callback and, if there is one, throw an exception before executing the code.
+
+Moreover, in order to handle C<goto> statements properly, L</uplevel> currently has to suffer a run-time overhead proportional to the size of the the callback in every case (with a small ratio), and proportional to the size of B<all> the code executed as the result of the L</uplevel> call (including subroutine calls inside the callback) when a C<goto> statement is found in the L</uplevel> callback.
+Despite this shortcoming, this XS version of L</uplevel> should still run way faster than the pure-Perl version from L<Sub::Uplevel>.
+
 =head1 DEPENDENCIES
 
 L<XSLoader> (standard since perl 5.006).