On Thu, May 21, 2020 at 12:31:32PM +0200, Arnd Bergmann wrote: > On Thu, May 21, 2020 at 12:14 PM Russell King - ARM Linux admin > <[email protected]> wrote: > > > > On Thu, May 21, 2020 at 11:06:23AM +0200, Arnd Bergmann wrote: > > > Note that the warning should come up for either W=1 or C=1, and I also > > > think that > > > new code should generally be written sparse-clean and have no warnings > > > with > > > 'make C=1' as a rule. > > > > No, absolutely not, that's a stupid idea, there are corner cases > > where hiding a sparse warning is the wrong thing to do. Look at > > many of the cases in fs/ for example. > > > > See https://lkml.org/lkml/2004/9/12/249 which should make anyone > > who sees a use of __force in some random code stop and question > > why it is there, and whether it is actually correct, or just there > > to hide a sparse warning. > > > > Remember, sparse is there to warn that something isn't quite right, > > and the view taken is, if it isn't right, then we don't "cast the > > warning away" with __force, even if we intend not to fix the code > > immediately. > > > > So, going for "sparse-clean" is actually not correct. Going for > > "no unnecessary warnings" is. > > > > And don't think what I've said above doesn't happen; I've rejected > > patches from people who've gone around trying to fix every sparse > > warning that they see by throwing __force incorrectly at it. > > > > The thing is, if you hide all the warnings, even for incorrect code, > > then sparse becomes completely useless to identify where things in > > the code are not quite correct. > > Adding __force is almost always the wrong solution, and I explictly > was not talking about existing code here where changing it would > risk introducing bugs or require bad hacks.
I'm using existing code to illustrate the problem with your idea of "sparse-clean" new code, trying to show you that it is not about being sparse clean, but about being correct. > However, when writing a new driver, sparse warnings usually > indicate that you are doing something wrong that is better addressed > by doing something different that does not involve adding __force. Right, but if you lay down a rule that says "new submissions must be sparse clean" you will get people using __force to shut sparse up. -- RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTC for 0.8m (est. 1762m) line in suburbia: sync at 13.1Mbps down 424kbps up
