Re: [PATCH] convert: mark a file-local symbol static

[Ramsay Jones <ramsay@ramsayjones.plus.com>]

On 17/10/16 03:18, Jeff King wrote:
> On Mon, Oct 17, 2016 at 02:37:58AM +0100, Ramsay Jones wrote:
> 
>> Hmm, well, you have to remember that 'make clean' sometimes
>> doesn't make clean. Ever since the Makefile was changed to only
>> remove $(OBJECTS), rather than *.o xdiff/*.o etc., you have to
>> remember to 'make clean' _before_ you switch branches. Otherwise,
>> you risk leaving some objects laying around. Since the script
>> runs 'nm' on all objects it finds, any stale ones can cause problems.
>> (Of course, I almost always forget, so I frequently have to manually
>> check for and remove stale objects!)
> 
> Gross. I would not be opposed to a Makefile rule that outputs the
> correct set of OBJECTS so this (or other) scripts could build on it.
> 
> IIRC, BSD make has an option to do this "make -V OBJECTS" or something,
> but I don't thnk there's an easy way to do so.

Hmm, I would go in the opposite direction and take a leaf out of
Ævar's book (see commit bc548efe) and this one-liner:

diff --git a/Makefile b/Makefile
index ee89c06..c08c25e 100644
--- a/Makefile
+++ b/Makefile
@@ -2506,7 +2506,7 @@ profile-clean:
 
 clean: profile-clean coverage-clean
        $(RM) *.res
-       $(RM) $(OBJECTS)
+       $(RM) $(addsuffix *.o,$(object_dirs))
        $(RM) $(LIB_FILE) $(XDIFF_LIB) $(VCSSVN_LIB)
        $(RM) $(ALL_PROGRAMS) $(SCRIPT_LIB) $(BUILT_INS) git$X
        $(RM) $(TEST_PROGRAMS) $(NO_INSTALL)

This would actually solve my problem, but it actually isn't a
_complete_ solution. (Hint: think about what isn't in $(OBJECTS),
depending on the configuration). ;-)

> Or, since it seems to find useful results quite frequently, maybe it
> would be worth including the script inside git (and triggering it with
> an optional Makefile rule). It sounds like we'd need a way to annotate
> known false positives, but if it were in common use, it would be easier
> to get people to keep that list up to date.

Hmm, I suspect that wouldn't happen, which would reduce it usefulness
and ultimately lead to it not being used. (Updating the 'stop list' would
fast become a burden.)

I find it useful to flag these issues automatically, but I still need
to look at each symbol and decide what to do (you may not agree with
some of my choices either - take a look at the output on the master
branch!).

The way I use it, I effectively ignore the 'stop list' maintenance issues.

ATB,
Ramsay Jones