[PATCH v4] Makefile: add a non-.PHONY "sparse-incr" target

Add a "sparse-incr" target to compliment the existing "sparse" target. The new "sparse-incr" target doesn't depend on "FORCE", and will create empty *.sp files as markers for the corresponding *.c file having been checked.

Those *.sp files in turn depend on the *.o files, so we can be certain that the dependencies are correct by either depending on all header files, or under "COMPUTE_HEADER_DEPENDENCIES=yes" the headers the relevant file needs.

The "sparse-incr" target is slower on a fresh git.git checkout, as it depends on the creation of the *.o files. But once the *.o and *.sp files are built it's able to incrementally update them. It's thus viable to run:

    make all sparse-incr

As part of a regular edit/compile/test cycle, or as a fast "git rebase --exec" command.

On my box (with -j8) the initial run of the "sparse" target takes ~5s, but ~16s for "sparse-incr". When using CC="ccache cc" the difference between the two is negligible.

Running e.g.:

    make grep.sp

Will behave the same way and will always re-run "cgcc", we'll only use the new dependency chain of "sparse-incr" is part of the MAKECMDGOALS.

I think it would make sense to just remove the "sparse" target entirely, and to say that anyone who cares about the speed of an initial "sparse" run should use "CC='ccache cc'". But per [1] and [2] there are existing users of "make sparse" and "make <file>.sp" that prefer the current semantics.

I.e. per [2] want "make <file>.sp" to *always* run "sparse", even though a corresponding "make <file>.o" would only re-run the "real" compilation if needed. I don't think that makes any sense, especially in combination with DEVELOPER=1 which'll ensure that -Werror would have made any errors in a "make <file>.sp" sticky.

But since we have existing users relying on it, and I don't really care at al about "make <file>.sp", I just want an incremental target I can use, let's create this new "make sparse-incr" instead of "fixing" the existing "make sparse".

See 0bcd9ae85d7 (sparse: Fix errors due to missing target-specific variables, 2011-04-21) for the modern implementation of the "sparse" target being changed here.

Appending to $@ without a move is OK here because we're using the .DELETE_ON_ERROR Makefile feature. See 7b76d6bf221 (Makefile: add and use the ".DELETE_ON_ERROR" flag, 2021-06-29). GNU make ensures that on error this file will be removed.

1. https://lore.kernel.org/git/xmqqtuib199x.fsf@gitster.g/
2. https://lore.kernel.org/git/457ec039-1e26-9da9-55f6-9ea79b962bfe@ramsayjones.plus.com/

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
---

Not so long since v3, it was hopelessly broken, sorry. I was doing some last-minute tweaking of the ifeq/ifneq conditions, so any target in the Makefile that wasn't "sparse" or "sparse-incr" broke.

Now we do the right thing by just splitting up the check, if you supply "sparse" and "sparse-incr" (in any order, and even among other targets) we'll error, they're incompatible.

Then we separately check if the "sparse-incr" target has been specified, if it has we'll use the new dependency mechanism, if not we'll use the old behavior.

Range-diff against v3:
1:  b6ba99ca4cc ! 1:  f31fa3e8282 Makefile: add a non-.PHONY "sparse-incr" target
    @@ Makefile: check-sha1:: t/helper/test-tool$X
      		$(SPARSE_FLAGS) $(SP_EXTRA_FLAGS) $<
     +endef
     +
    -+ifneq ($(filter sparse-incr,$(MAKECMDGOALS)),sparse-incr)
    -+ifneq ($(filter sparse,$(MAKECMDGOALS)),)
    ++ifeq ($(sort $(filter sparse sparse-incr,$(MAKECMDGOALS))),sparse sparse-incr)
     +$(error The sparse and sparse-incr targets cannot be combined!)
     +endif
    ++
    ++ifneq ($(filter sparse-incr,$(MAKECMDGOALS)),sparse-incr)
     +$(SP_OBJ): %.sp: %.c GIT-CFLAGS FORCE
     +	$(cmd_run_sparse)
     +else

 .gitignore |  1 +
 Makefile   | 19 +++++++++++++++++--
 2 files changed, 18 insertions(+), 2 deletions(-)