On 16.07.2024 02:43, Stefano Stabellini wrote: > On Mon, 15 Jul 2024, Jan Beulich wrote: >> On 13.07.2024 00:38, Stefano Stabellini wrote: >>> On Wed, 3 Jul 2024, Jan Beulich wrote: >>>> I further have to note that, as indicated during the earlier discussion, >>>> I still cannot see how occasional ambiguity is going to be dealt with. >>>> IOW from the rules above two different headers could still end up with >>>> the same guard identifier. >>> >>> Maybe something like this? >>> >>> "In the event of naming collisions, exceptions to the coding style may >>> be made at the discretion of the contributor and maintainers." >> >> Hmm, maybe I wasn't clear enough then. My take is that the scheme should >> simply not allow for possible collisions. Neither the contributor nor the >> reviewer may spot such a collision, and it may therefore take until the >> first full scan that one is actually noticed. Which I consider too late >> in the process, even if we already were at the point where commits were >> checked pre-push. > > Looking at the proposal, copy/pasted here for convenience: > > - private headers -> <dir>_<filename>_H > - asm-generic headers -> ASM_GENERIC_<filename>_H > - #ifndef ASM_GENERIC_X86_PERCPU_H > #define ASM_GENERIC_X86_PERCPU_H > //... > #endif /* ASM_GENERIC_X86_PERCPU_H */ > - arch/<architecture>/include/asm/<subdir>/<filename>.h -> > ASM_<architecture>_<subdir>_<filename>_H > - #ifndef ASM_X86_DOMAIN_H > #define ASM_X86_DOMAIN_H > //... > #endif /* ASM_X86_DOMAIN_H */ > - xen/include/xen/<filename>.h -> XEN_<filename>_H > - xen/include/xen/<subdir>/<filename>.h -> XEN_<subdir>_<filename>_H > > > The only possibility for collision that I can see is from the first > point: > > - private headers -> <dir>_<filename>_H
I don't think this is the only possibility of collisions. The <subdir>_<filename> parts can similarly cause problems if either of the two involved names contains e.g. a dash (which would need converting to an underscore) or an underscore. To avoid this, the name separators (slashes in the actual file names) there may need representing by double underscores. > two directories like this could collide: > > - arch/arm/arm64/lib/something.h -> LIB_SOMETHING_H > - arch/arm/arm32/lib/something.h -> LIB_SOMETHING_H > - arch/x86/lib/something.h -> LIB_SOMETHING_H > > (Leaving aside that in this example it would not be an issue because the > three headers are not meant to be all included in the same file.) > > Can we specify that <dir> should go all the way back to the arch/ or or > common or drivers directory? > > - arch/arm/arm64/lib/something.h -> ARM_ARM64_LIB_SOMETHING_H > - arch/arm/arm32/lib/something.h -> ARM_ARM32_LIB_SOMETHING_H > - arch/x86/lib/something.h -> X86_LIB_SOMETHING_H We can of course, so long as we're okay(ish) with the length and redundancy. As already indicated before, there are downsides to this. Yet a firm scheme with few rules has the benefit that it might even be possible to use a script to do all the guard adjustments. Jan > We can rely on the filesystem paths to make sure to avoid collisions.
