Re: [PATCH v3 5/5] Makefile: normalize clobbering & xargs for tags targets

[Jeff King <peff@peff.net>]

On Thu, Jul 22, 2021 at 01:23:06AM +0200, Ævar Arnfjörð Bjarmason wrote:

> This is because the xargs command may decide to invoke the program
> multiple times. We need to make sure we've got a union of its results
> at the end.
> 
> For "ctags" and "etags" we used the "-a" flag for this, for cscope
> that behavior is the default. Its "-u" flag disables its equivalent of
> an implicit "-a" flag.

Hrm, that's not the experience I get with cscope. E.g.:

  $ cscope -b wt-status.c
  $ grep -m1 wt-status.c cscope.out
	@wt-status.c
  $ cscope -b git.c
  $ grep -m1 wt-status.c cscope.out
  [no output]

I wondered if I was being too hacky with my grep there. But if I
simulate more extreme cmdline-splitting like so:

diff --git a/Makefile b/Makefile
index c7c46c017d..8a1ec00938 100644
--- a/Makefile
+++ b/Makefile
@@ -2751,7 +2751,7 @@ tags: FORCE
 
 cscope:
 	$(RM) cscope*
-	$(FIND_SOURCE_FILES) | xargs cscope -b
+	$(FIND_SOURCE_FILES) | xargs -n1 cscope -b
 
 ### Detect prefix changes
 TRACK_PREFIX = $(bindir_SQ):$(gitexecdir_SQ):$(template_dir_SQ):$(prefix_SQ):\

then I get a file that is much smaller (1MB versus 9MB), and fails to
find lots of things:

  $ cscope -d -L1cmd_pack_objects
  [no output]

(by the way, I confused myself several times while testing this because
without "-d", it will actually rebuild the index using files in the
current directory. So something like " cscope -L1main" gives the same
result in both cases, or even if you don't have a cscope.out file at
all!)

But it really seems like cscope is not appending as we'd want. From
skimming the manpage, I think replacing xargs with "cscope -b -i -"
should work (and seems to for me).

-Peff