https://sourceware.org/bugzilla/show_bug.cgi?id=22843
Fangrui Song <maskray at google dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |maskray at google dot com
--- Comment #6 from Fangrui Song <maskray at google dot com> ---
Besides -M, -t (--trace) can also be used.
There is no LOAD command in gold's and lld's -M formats.
The linkers' -t formats have more similarities, so it is easier for a script to
adapt the three linkers.
Assume there is a linker option, say --write-dependencies=final_exe.link.d,
that writes dependency information:
final_exe: source.o /path/to/my/place/liba.a /path/to/somewhere/else/libb.so
The Makefile contains `-include final_exe.link.d`.
If /path/to/my/place/liba.a is deleted, there will be a GNU make error:
make: *** No rule to make target '/path/to/my/place/liba.a', needed by
'final_exe'. Stop.
To prevent this, the linker needs another option that is similar to gcc -MP,
which creates force targets:
/path/to/my/place/liba.a:
/path/to/somewhere/else/libb.so:
An example Makefile:
CFLAGS = -MD -MP -MF [email protected] # note -MP is not required in build.ninja
LDFLAGS = -fuse-ld=bfd -Wl,[email protected] -Wl,-MP
final_exe_deps = source.o -la -lb
final_exe: $(final_exe_deps)
$(LINK.c) $(final_exe_deps) $(LDLIBS) -o $@
-include a.o.d
-include final_exe.link.d
The build system (may be the generator of the Makefile) still has to analyze
force targets to extract information. Then, why can't it obtain dependency
information from -M or -t in the first place?
There is a similar feature request to lld:
https://bugs.llvm.org/show_bug.cgi?id=42806 and my reaction is the same that
this is probably not necessary https://reviews.llvm.org/D65430#1616208
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
bug-binutils mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-binutils
