Re: Having hard time adding to CFLAGS

[Rich Felker <dalias@libc.org>]

On Fri, Oct 23, 2015 at 06:53:29AM +0200, Denys Vlasenko wrote:
> On Fri, Oct 23, 2015 at 6:09 AM, Rich Felker <dalias@libc.org> wrote:
> > On Thu, Oct 22, 2015 at 11:02:36PM -0400, Rich Felker wrote:
> >> > the attached patch makes this work, but i
> >> > consider -Os to be not part of 'needed'
> >>
> >> > diff --git a/Makefile b/Makefile
> >> > index 844a017..f713286 100644
> >> > --- a/Makefile
> >> > +++ b/Makefile
> >> > @@ -94,22 +94,22 @@ crt/crt1.o crt/Scrt1.o crt/rcrt1.o src/ldso/dlstart.lo: $(wildcard arch/$(ARCH)/
> >> >
> >> >  crt/rcrt1.o: src/ldso/dlstart.c
> >> >
> >> > -crt/Scrt1.o crt/rcrt1.o: CFLAGS += -fPIC
> >> > +crt/Scrt1.o crt/rcrt1.o: CFLAGS_ALL += -fPIC
> >>
> >> This is the correct fix. I was not aware that make variables set from
> >> the make command line would take precedence over the target-specific
> >> += concatenations. The intent has always been that editing CFLAGS
> >> should not break the build (unless you put really inapproriate stuff
> >> there, of course).
> >
> > I've committed this with one change (omitting the -O3 thing that's
> > really optional) and another related fix.
> >
> > Note however that overriding CFLAGS at make time is still a bad idea.
> > It will suppress all the warning options configure detected and other
> > useful but non-essential things like -fno-unwind-tables and
> > -fno-asynchronous-unwind-tables. We should probably discuss whether
> > this behavior is desirable. We could factor out all of the stuff
> > configure detects into a CFLAGS_AUTO and leave CFLAGS just containing
> > the user-provided options. Opinions?
> 
> Make configure --help warn/explain what would happen if
> make CFLAGS=foo is run.
> 
> Currently, it is confusing. See for yourself:
> 
> 
> Usage: $0 [OPTION]... [VAR=VALUE]... [TARGET]
> 
> To assign environment variables (e.g., CC, CFLAGS...), specify them as
> VAR=VALUE.  See below for descriptions of some of the useful variables.
> ....
> ("ok, I got it. CFLAGS should go to configure's command line!")
> ....
> Some influential environment variables:
>   CC                      C compiler command [detected]
>   CFLAGS                  C compiler flags [-Os -pipe ...]
>   CROSS_COMPILE           prefix for cross compiler and tools [none]
>   LIBCC                   compiler runtime library [detected]
> 
> Use these variables to override the choices made by configure.
> 
> ("What? You just said that CFLAGS should be on command line!
> Now you are saying it should be in the environment!
> What it is?")

The behavior is the same as autoconf: they're accepted either as
environment variables or on the configure command line. Some users do:

	CFLAGS=... ./configure ...

and others do:

	./configure CFLAGS=... ...

Following a principle of least surprise, I've tried to match the
behavior of autoconf-generated configure scripts, which actually has
something resembling a specification outside of the autoconf
implementation:

https://www.gnu.org/prep/standards/html_node/Configuration.html

I can add some language to the help text to make it explicit that
either is accepted.

Rich