On Oct 29, 2015, at 10:39 PM, Andi Kleen <a...@firstfloor.org> wrote: > David Malcolm <dmalc...@redhat.com> writes: >> >> * adds -Wmisleading-indentation to -Wall > > I have doubts this is a good idea. I'm sure this will break > a bazillion packages which (misguidedly) ship with -Wall -Werror. > > Would be better to leave the user the choice, given that such a thing > was historically never checked before and not include it in -Wall.
So, my experience is that folks that run large scale builds have ways of adding a flag like -Wno-misleading-indentation to all builds or otherwise turning off -Werror for exactly this reason. An OS build from source does make for a nice smoke test for the feature. These people also usually fix(ate) the compiler and don’t update it, for much the same reason. The plan would be for gcc to update, packages to update, the OS or package integrator then pull in the updated packages and the updated compiler and then release. As this time, things would then be ok. We should turn it on now (for trunk only), and send in fixes and bug reports to all packages that don’t build. Those packages then have time to update. When we go for release, maybe the RM decides that we will give people 1 more year to fix their code and leave it out of -Wall for 1 more year, if there are too many packages dragging their feet. I think that’s a fine use of an RM. We can record in a gcc PR to disable for Wall the packages that break, and then consider disabling for Wall if the package list remains too long. If the list is short by the time we go to release, then the RM closes as not to be fixed. This is why I favor putting the change in Wall now (but subject to reversion on all release branches at RM discretion).