[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-kernel
Subject:    Re: [PATCH] kbuild: clang: remove crufty HOSTCFLAGS
From:       Nick Desaulniers <nick.desaulniers () gmail ! com>
Date:       2017-09-30 23:14:50
Message-ID: 20170930231450.cggvysrt4zedsbkq () lostoracle ! net
[Download RAW message or body]

On Thu, Sep 28, 2017 at 07:52:35PM +0900, Masahiro Yamada wrote:
> 2017-09-26 11:28 GMT+09:00 Nick Desaulniers <nick.desaulniers@gmail.com>:
> >  HOSTCFLAGS   := -Wall -Wmissing-prototypes -Wstrict-prototypes -O2 \
> > +               $(call hostcc-option,-fno-delete-null-pointer-checks) \
> >                 -fomit-frame-pointer -std=gnu89 $(HOST_LFS_CFLAGS)
> 
> You call hostcc-option
> before Kbuild.include is included around line 341.
> 
> So, $(call hostcc-option, ...) returns always an empty string here
> whether the compiler supports the option or not.

So calling a yet-to-be defined variable results in an empty string
rather than a loud failure?  Chalk that up there with language features
no one ever asked for.  That kind of implicit conversion gets languages
like JavaScript (with its loose type system, not that C is without its
own implicit type conversions/promotions) in a lot of hot water.

If that's the case, why are includes not at the top of Makefiles, if
silent failure is a possibility?  Is there a reason the include is so
far into the Makefile?

Is your sugguestion to raise the include or lower the HOSTCFLAGS
definition?

> > -ifeq ($(shell $(HOSTCC) -v 2>&1 | grep -c "clang version"), 1)
> > -HOSTCFLAGS  += -Wno-unused-value -Wno-unused-parameter \
> > -               -Wno-missing-field-initializers -fno-delete-null-pointer-checks
> > -endif
>
> The logic is very strange in the first place.
>
> Even very old GCC supports -fno-delete-null-pointer-checks,
> but clang does not.
>
> Here, -fno-delete-null-pointer-checks is added only when
> we are using clang for HOSTCC.  This is opposite.
>
> I guess we can remove all of them
> unless somebody can explain the rationale.

+llvm-linux

I suppose maybe different ARCH's have different host binaries made
during the build?  I tested x86_64 and arm64.  The commit message that
added them missed any context or justification.
[prev in list] [next in list] [prev in thread] [next in thread]