> On Aug 19, 2024, at 7:44 AM, Jeff Layton <jlayton@xxxxxxxxxx> wrote: > > On Mon, 2024-08-19 at 10:04 +1000, NeilBrown wrote: >> On Fri, 16 Aug 2024, Jeff Layton wrote: >> >>> +// Generated by lkxdrgen, with hand-edits. >> >> I *really* don't like having code in the kernel that is partly >> tool-generated and partly human-generated, and where the boundary isn't >> obvious (like separate files). >> >> If we cannot use tool-generated code as-is, then let's fix the tool. >> If we cannot fix the tool, then include the raw output and a >> human-generated patch which the makefile combines. >> >> Ideally the tool should be in tools/, the .x file should be in fs/nfsd/ >> and the makefile should apply the one to the other. We are going to >> want to do that eventually and I think it should be priority. The tool >> doesn't have to be bug-free before it lands (nothing is). >> >> A particular reason for this is that I cannot review tool-generated >> hand-editted code. It is too noisy and I don't know which parts are >> worth closer inspection etc. > > Fair point. Chuck made some similar comments to me privately, and it > looks like he has updated his xdrgen tool as well. > > I'll plan to just respin that part from scratch and regenerate from the > .x files. I'll also keep my hand-edits in a separate commit for the > next version. > > It'll probably take me a few days, but I'll try to have something to > resend within the next week or so. Meanwhile, I'll post the current xdrgen tool for review. It already lives under tools/ as Neil suggested above. I'm hoping that by providing the .x snippet used to generate the source, and by adding one or two "pragma" annotations to the tool to handle certain exceptions, no hand-edits or extra patching will be needed. -- Chuck Lever
