On Wed, Apr 12 2023, Glen Choo via GitGitGadget wrote: > From: Glen Choo <chooglen@xxxxxxxxxx> > > This isn't set in stone; we expect this to be updated as the project > evolves. > > Signed-off-by: Glen Choo <chooglen@xxxxxxxxxx> > --- > contrib/coccinelle/README | 23 +++++++++++++++++++++++ > 1 file changed, 23 insertions(+) > > diff --git a/contrib/coccinelle/README b/contrib/coccinelle/README > index 9b28ba1c57a..055e3622e5c 100644 > --- a/contrib/coccinelle/README > +++ b/contrib/coccinelle/README > @@ -92,3 +92,26 @@ that might be useful to developers. > > The absolute times will differ for you, but the relative speedup > from caching should be on that order. > + > +== Authoring and reviewing coccinelle changes > + > +* When introducing and applying a new .cocci file, both the Git changes and > + .cocci file should be reviewed. > + > +* Reviewers do not need to be coccinelle experts. To give a Reviewed-By, it is > + enough for the reviewer to get a rough understanding of the proposed rules by > + comparing the .cocci and Git changes, then checking that understanding > + with the author. Maybe it would be useful here to add something about how you can reproduce the application of the coccinelle rule(s). I sometimes do this on an ad-hoc basis, something like (untested): git checkout HEAD^ -- ':!contrib/coccinelle' '*.[ch]' make coccicheck <apply any suggested patches> git add -A Then see if I ended up with a no-op, or if there's suggested changes. With changes that modify both the header & source files this can be tricky with the default of SPATCH_USE_O_DEPENDENCIES=Y, but disabling it will take care of any potential circular dependency issues. I.e. when the header doesn't contain a required construct that we're replacing. > +* Conversely, authors should consider that reviewers may not be coccinelle > + experts. The primary aim should be to make .cocci files easy to understand, > + e.g. by adding comments or by using rules that are easier to understand even > + if they are less elegant. I agree that simple things should be kept simple, but this seems to come quite close (or perhaps past the line of) suggesting that we use only the simpler features of the language when a more elegant solution would be available with something less well-known. I think we should clarify that that's not the intent. Just as with C, shellscript, Perl etc. we should aim for simplicity, but ultimately we should expect that we can target the full available language available to us. > +* .cocci rules should target only the problem it is trying to solve; "collateral > + damage" is not allowed. I think what you mean here is that you should be able to apply the rule and still build the project. I think that's correct, but I also think that rather than define this in prose, how about we just modify the current CI job to apply the result of non-pending rules, and do a build at the end? Wouldn't that assert this going forward. > +* .cocci files used for refactoring should be temporarily kept in-tree to aid > + the refactoring of out-of-tree code (e.g. in-flight topics). They should be > + removed when enough time has been given for others to refactor their code, > + i.e. ~1 release cycle. Maybe s/should/can/? E.g. for my recent "index" and "the_repository" patches I think they can, but we often keep unused code in-repo for longer than that. If e.g. that code stayed in for more than one release until someone cared to remove it we'd also be fine. I also don't know if some long-running forks (e.g. GfW?) would benefit from the rules for longer than that...
