On 12/5/2023 8:11 AM, Russell King (Oracle) wrote: > On Tue, Dec 05, 2023 at 07:29:12AM -0800, Jakub Kicinski wrote: >> On Tue, 5 Dec 2023 15:10:50 +0000 Russell King (Oracle) wrote: >>> I've raised this before in other subsystems, and it's suggested that >>> it's better to have it in the .c file. I guess the reason is that it's >>> more obvious that the function is documented when modifying it, so >>> there's a higher probability that the kdoc will get updated when the >>> function is altered. >> >> Plus I think people using IDEs (i.e. not me) may use the "jump to >> definition" functionality, to find the doc? >> >> TBH I thought putting kdoc in the C source was documented in the coding >> style, but I can't find any mention of it now. > > Well, in Documentation/doc-guide/kernel-doc.rst: > > The function and type kernel-doc comments should be placed just before > the function or type being described in order to maximise the chance > that somebody changing the code will also change the documentation. > > That implies (but not explicitly) that it should be at the function > definition site, since "changing the code" is used as an argument as > I did in my previous email. > > Secondly, this document goes on to give an example of running > scripts/kernel-doc on a .c file. > > Thirdly, there are seven references in this document of kernel-doc > in .c files, and only one for kernel-doc in a .h file. So this suggests > that "it will be in a .c file" isn't a rule (it can't be because of > documenting structures!) > > So let's not get hung up on whether it should be in .c or .h because I > think that isn't relevant. Instead, I think it's about "it should be at > the definition site" - that being a structure definition or a function > definition, and not at a function prototype. > > The only exception I can think of is the style I've used in > linux/phylink.h for the _method_ definitions which look like function > prototypes - that's just a work-around because one can't kernel-doc > the structure-of-function-pointers and document the function parameters > without jumping through that hoop, and it would be silly to document > the methods in some random driver! > The Linux Kernel philosophy of documenting functions instead of prototypes has always bothered me since I'm "old school" and am ingrained with the software engineering philosophy that you document interfaces, not implementations. This was reinforced early in my career by working on multiple projects in different programming languages using processes outlined in DOD-STD-2167A, and for some projects, especially ones written in Ada, the header files were the design and the documentation. This philosophy was further enforced when working with closed source projects (Windows, IOS, VxWorks) where all the documentation was contained in shared header files. So in my experience a function prototype IS the function definition, and the actual function is just the implementation of that definition. But that thinking obviously isn't shared by others. /jeff
