On 9 December 2015 at 21:48, Giuseppe Scrivano <gscri...@redhat.com> wrote: > Nick Coghlan <ncogh...@gmail.com> writes: >> There's one trick we discussed that I think is actually fairly safe: >> using hard links to share the same set of compiled files for normal >> execution and -O, rather than having two copies. Assuming the "-O" >> files are used, then any asserts and "if __debug__:" blocks in the >> system provided libraries would get skipped even in the non-optimised >> case. > > thinking more of it: as the trick is quite generic and not as aggressive > as my first proposed patch which stripped all the source code, shouldn't > be -OO used in any case where the source code is available as well and > not only for Atomic?
No, as -OO strips docstrings, and there's plenty of Python code that assumes docstrings are always available. It can be a useful memory optimisation trick as an application integrator (since you know whether or not you're using any libraries that rely on docstrings being present), but it's a recipe for obscure compatibility bugs as a platform provider. "-O" is safer, as it's much closer semantically to normal execution (the main changes relative to the default execution mode are that assert statements and "if __debug__:" blocks get compiled out) Cheers, Nick. -- Nick Coghlan | ncogh...@gmail.com | Brisbane, Australia
