A couple of questions preceded by the usual curmudgeonly comment :-)
I'm not a big fan of checkpatch.pl. It's 5965 lines of dense perl spaghetti code, continues to grow, and is attempting to achieve that which I understand may be impossible: parsing C with REs and random blobs of PERL. Given that the problem may be impossible, checkpatch finds it hard, and makes mistakes, as experienced on the projects that use it.
It's not surprising it continues to grow, and not surprising it's got lots of errors. You can find discussions on the various lists which boil down to "just ignore checkpatch.pl in this case." This is not confidence-inspiring.
I'd most prefer to move to clang-format for this sort of thing, which is what we're doing on at least one other kernel project I work on.
I continue to not understand the injunction on typedefs; operating systems written by the inventors of C continued to use them until Bell Labs disbanded the OS group in 2015 ...
Finally, why the use of packed? In general you really only want to use packed when you're using it to directly address data. Why not deserialize the data from memory into an unpacked struct? That's how ACPICA seems to do it.
thanks
ron
On Tue, Jul 26, 2016 at 5:05 PM Julius Werner jwerner@chromium.org wrote:
Does git commit --no-verify (or -n for short) allow you to commit?
I think we should try to do a little of option 1 within reason, not by forking the script but rather by disabling more check steps that don't match the coreboot code style (by having our wrapper pass --ignore XXX flags to checkpatch) and possibly upstreaming checkpatch.pl patches with new features we need to the Linux kernel (to make the detection more accurate or add a more specific --ignore flag we can turn on). In the Chromium fork of coreboot we're already using a bunch of those flags that we should probably use in upstream coreboot as well:
--ignore C99_COMMENTS --ignore GLOBAL_INITIALISERS --ignore INITIALISED_STATIC --ignore LINE_SPACING --ignore NEW_TYPEDEFS --ignore PREFER_ALIGNED --ignore PREFER_PACKED --ignore PREFER_PRINTF --ignore SPLIT_STRING
However, in my experience there will always be edge cases checkpatch doesn't get right, no matter how well you tweak it. It should always be up to the discretion of the committer/reviewer which warnings they act upon, so we need to retain a mechanism (like --no-verify) to work around it if necessary. Maybe we should update the pre-commit hook to print an explanation to this effect. (Ideally, I think it would be awesome if Gerrit could separately run checkpatch and make it explicit in the interface if someone circumvented a warning this way so it can be discussed, but that's probably not and easy thing to add.)
-- coreboot mailing list: coreboot@coreboot.org https://www.coreboot.org/mailman/listinfo/coreboot
-
ron minnich