Using the same options for compilation and linking

[Pau Koning]

Hi,

there are currently some things going on regarding the analysis of the
build logs using blhc. This is extreme nice and made me think about a
sentence most people missed in the GCC documentation (it is repeated often):

"For predictable results, you must also specify the same set of options
used for compilation (-fpic, -fPIC, or model suboptions) when you specify
this linker option."

Most of the time it works quite well, but what happens when somebody wants
to enable -flto (just using the environment CFLAGS variable)? For example
libtool will then stop to generate debug info because it drops the -g from
Debians CFLAGS in the linking stage.

Or as another example take the builtin rules from GNU make:

LINK.o = $(CC) $(LDFLAGS) $(TARGET_ARCH)

%: %.o
#  commands to execute (built-in):
$(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -o $@

It is missing the CFLAGS completely. So, the requested optimizations
(either added by the Makefile to CFLAGS or by the user environment) is now
completely dropped and only linker flags are used during the -flto compile
step during the linking (when CFLAGS and LDFLAGS have -flto). Or the
link-time optimization is not done as expected (-flto only specified in
CFLAGS)

This is rather problematic for my example but also subtle defects can
happen.

Any thoughts on making this a long term QA goal? Or are there other ideas
about this problem? I've Cc'ed the mentioned projects to allow them to make
statements about their position.