| Message ID | patch-1.3-e03fde1b642-20210921T224944Z-avarab@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Makefile: make "sparse" and "hdr-check" non-.PHONY | expand |
On Wed, Sep 22, 2021 at 12:55:13AM +0200, Ævar Arnfjörð Bjarmason wrote: > On my box with -j8 it was fast before, or around 5 seconds, now it > only takes that long the first time, and the common case is <100ms, or > however long it takes GNU make to stat the *.sp file and see that all > the corresponding *.c files are older. Sort of side note, but I think the main culprit here is all of the $(shell) invocations, etc, we run. Try: make SHELL_PATH='sh -x' to get a taste (do it twice, because the first one will actually rebuild everything due to the changed shell; the second one _should_ be a noop but still runs a bunch of stuff). Though even that doesn't tell the whole story, as it doesn't make clear which shells are invoked by make directly. Try this: cat >/tmp/foo <<-\EOF #!/bin/sh echo >&2 "==> shell $*" exec sh "$@" EOF make SHELL_PATH=/tmp/foo There's lots of low-hanging fruit like: ==> shell -c echo TEST_SHELL_PATH=\''/tmp/foo'\' >>GIT-BUILD-OPTIONS+ ==> shell -c echo PERL_PATH=\''/usr/bin/perl'\' >>GIT-BUILD-OPTIONS+ ==> shell -c echo DIFF=\''diff'\' >>GIT-BUILD-OPTIONS+ ==> shell -c echo PYTHON_PATH=\''/usr/bin/python'\' >>GIT-BUILD-OPTIONS+ [...and over a dozen more...] Those could easily be a single shell invocation, rather than one per echo. Another culprit is GIT-VERSION-GEN, which we run whether it's needed or not, and takes something like 25ms. It's probably not worth spending too much time micro-optimizing here (and certainly it's orthogonal to your series), but there may be some low-hanging fruit (although a hacky attempt at minimizing the shell calls for GIT-BUILD-OPTIONS didn't seem to show any speedup, so maybe it's not so low-hanging after all). -Peff
diff --git a/.gitignore b/.gitignore index 311841f9bed..b02250a50c4 100644 --- a/.gitignore +++ b/.gitignore @@ -224,6 +224,7 @@ *.lib *.res *.sln +*.sp *.suo *.ncb *.vcproj diff --git a/Makefile b/Makefile index a9f9b689f0c..5b09f67aab0 100644 --- a/Makefile +++ b/Makefile @@ -2896,11 +2896,13 @@ check-sha1:: t/helper/test-tool$X SP_OBJ = $(patsubst %.o,%.sp,$(C_OBJ)) -$(SP_OBJ): %.sp: %.c GIT-CFLAGS FORCE +$(SP_OBJ): %.sp: %.c GIT-CFLAGS $(QUIET_SP)cgcc -no-compile $(ALL_CFLAGS) $(EXTRA_CPPFLAGS) \ - $(SPARSE_FLAGS) $(SP_EXTRA_FLAGS) $< + -Wsparse-error \ + $(SPARSE_FLAGS) $(SP_EXTRA_FLAGS) $< && \ + >$@ -.PHONY: sparse $(SP_OBJ) +.PHONY: sparse sparse: $(SP_OBJ) EXCEPT_HDRS := command-list.h config-list.h unicode-width.h compat/% xdiff/%
Change the "sparse" target and its *.sp dependencies to be non-.PHONY. It's now viable to run it as part of a normal compilation target, as we'll only re-generate these checks if the source *.c file has changed. On my box with -j8 it was fast before, or around 5 seconds, now it only takes that long the first time, and the common case is <100ms, or however long it takes GNU make to stat the *.sp file and see that all the corresponding *.c files are older. 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. It is critical that we use -Wsparse-error here, otherwise the error would only show up once, but we'd successfully create the empty *.sp file, and running a second time wouldn't show the error. I'm therefore not putting it into SPARSE_FLAGS or SP_EXTRA_FLAGS, it's not optional, the Makefile logic won't behave properly without it. 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. Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> --- .gitignore | 1 + Makefile | 8 +++++--- 2 files changed, 6 insertions(+), 3 deletions(-)