On Wed, Mar 11, 2020 at 12:49 AM Sergei Trofimovich <sly...@gentoo.org>
wrote:
> On Tue, 10 Mar 2020 20:54:12 -0700
> Manoj Gupta <manojgu...@google.com> wrote:
>
> > On Tue, Mar 3, 2020 at 1:17 AM Sergei Trofimovich <sly...@gentoo.org>
> wrote:
> >
> > > On Mon, 2 Mar 2020 19:03:48 -0800
> > > Manoj Gupta <manojgu...@google.com> wrote:
> > >
> > > > On Thu, Feb 27, 2020 at 11:20 PM Sergei Trofimovich <
> sly...@gmail.com>
> > > > wrote:
> > > >
> > > > > On Thu, 27 Feb 2020 at 22:41, Manoj Gupta <manojgu...@google.com>
>
> > > wrote:
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Thu, Feb 27, 2020 at 11:22 AM Manoj Gupta <
> manojgu...@google.com>
> > >
> > > > > wrote:
> > > > > >>
> > > > > >> gcc-config installs cc/f77 by default. This may be undesired on
> > > > > >> systems that want to set their own versions of cc/f77.
> > > > > >>
> > > > > >> Add option "-n"/"--no-default-vars" to not install the cc/f77
> > > > > >> wrappers.
> > > > > >>
> > > > > >> Signed-off-by: Manoj Gupta <manojgu...@google.com>
> > > > > >> ---
> > > > > >> gcc-config | 6 +++++-
> > > > > >> 1 file changed, 5 insertions(+), 1 deletion(-)
> > > > > >>
> > > > > >> diff --git a/gcc-config b/gcc-config
> > > > > >> index f03a46a..6f306db 100755
> > > > > >> --- a/gcc-config
> > > > > >> +++ b/gcc-config
> > > > > >> @@ -262,7 +262,7 @@ update_wrappers() {
> > > > > >> # For all toolchains, we want to create the fully
> qualified
> > > > > >> # `tuple-foo`. Only native ones do we want the simple
> > > `foo`.
> > > > > >> local all_wrappers=( ${new_wrappers[@]/#/${CTARGET}-} )
> > > > > >> - if ! is_cross_compiler ; then
> > > > > >> + if ! is_cross_compiler && [[ "${DEFAULT_PROGS}" ==
> "yes"
> > > ]];
> > > > > then
> > > > > >> all_wrappers+=( "${new_wrappers[@]}" )
> > > > > >> # There are a few fun extra progs which we have
> to
> > > > > handle #412319
> > > > > >> all_wrappers+=( cc:gcc f77:g77 )
> > > > > >> @@ -951,6 +951,7 @@ FORCE="no"
> > > > > >> CC_COMP=
> > > > > >> ENV_D="${EROOT}etc/env.d"
> > > > > >> GCC_ENV_D="${ENV_D}/gcc"
> > > > > >> +DEFAULT_PROGS="yes"
> > > > > >>
> > > > > >> for x in "$@" ; do
> > > > > >> case "${x}" in
> > > > > >> @@ -972,6 +973,9 @@ for x in "$@" ; do
> > > > > >> -l|--list-profiles)
> > > > > >> set_doit list_profiles
> > > > > >> ;;
> > > > > >> + -n|--no-default-vars)
> > > > > >> + DEFAULT_PROGS="no"
> > > > > >> + ;;
> > > > > >> -S|--split-profile)
> > > > > >> if [[ ( $1 != "-S" && $1 !=
> > > "--split-profile" )
> > > > > || $# -eq 1 ]] ; then
> > > > > >> usage 1
> > > > > >> --
> > > > > >>
> > > > > >
> > > > > > Not sure of the correct mailing list for patches to gcc-config
> so
> > > also
> > > > > adding toolchain@gentoo .
> > > > > >
> > > > >
> > > > > toolch...@gentoo.org should generally be fine.
> > > > >
> > > > > Today cc->gcc and gcc->${CHOST}-gcc symlinks are effectively owned
> by
> > > > > a single sys-devel/gcc-config package.
> > > > > gcc-config is calld to update symlinks every time sys-devel/gcc is
> > > > > installed/updated. That way we never get cc/gcc
> > > > > out of sync.
> > > > >
> > > > > Your change makes /usr/bin/cc an orphan symlink. I think we need to
> > > > > still keep a 'cc'/'f77' ownership somewhere
> > > > > (say, a separate package).
> > > > >
> > > > > I suggest making a decision to handle or not handle 'cc'/'f77' and
> > > > > gcc-config build-time, not gcc-config call-time.
> > > > > That way sys-devel/gcc updates will behave the same as manual
> > > > > 'gcc-config-' calls.
> > > > >
> > > > > Mechanically that could be a Makefile variable that switches the
> > > > > behaviour on/off at
> > > > > https://gitweb.gentoo.org/proj/gcc-config.git/tree/Makefile
> > > > > and exposed as an USE flag on sys-devel/gcc-config ebuild.
> > > > >
> > > > > Later we can create a separate ebuild to manage /usr/bin/cc. For
> gcc
> > > > > it's not hard, as gcc-config always provides /usr/bin/gcc and
> > > > > /usr/bin/${CHOST}-gcc.
> > > > > These can be static symlinks that don't require maintenance
> updates.
> > > > >
> > > > > Thanks for the suggestion. I will look into adding a Makefile
> > > variable
> > > > exposed via an USE flag.
> > >
> > > You might also need to look in the detail at 'c++', 'cpp' and
> ${CHOST}-*
> > > equivalents
> > > as those also get linked by gcc-config:
> > >
> > > $ LANG=C ls -l /usr/bin/ | fgrep 10.0.1 | fgrep -v -- '-10.0.1 ->'
> > > lrwxrwxrwx 1 root root 43 Feb 4 10:45 c++ ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/c++
> > > lrwxrwxrwx 1 root root 43 Feb 4 10:45 cc ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcc
> > > lrwxrwxrwx 1 root root 43 Feb 4 10:45 cpp ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/cpp
> > > lrwxrwxrwx 1 root root 43 Feb 4 10:45 g++ ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/g++
> > > lrwxrwxrwx 1 root root 43 Feb 4 10:45 gcc ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcc
> > > lrwxrwxrwx 1 root root 46 Feb 4 10:45 gcc-ar ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcc-ar
> > > lrwxrwxrwx 1 root root 46 Feb 4 10:45 gcc-nm ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcc-nm
> > > lrwxrwxrwx 1 root root 50 Feb 4 10:45 gcc-ranlib ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcc-ranlib
> > > lrwxrwxrwx 1 root root 45 Feb 4 10:45 gccgo ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gccgo
> > > lrwxrwxrwx 1 root root 44 Feb 4 10:45 gcov ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcov
> > > lrwxrwxrwx 1 root root 49 Feb 4 10:45 gcov-dump ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcov-dump
> > > lrwxrwxrwx 1 root root 49 Feb 4 10:45 gcov-tool ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcov-tool
> > > lrwxrwxrwx 1 root root 48 Feb 4 10:45 gfortran ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gfortran
> > > lrwxrwxrwx 1 root root 45 Feb 4 10:45 go-10 ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/go-10
> > > lrwxrwxrwx 1 root root 48 Feb 4 10:45 gofmt-10 ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gofmt-10
> > > lrwxrwxrwx 1 root root 48 Feb 4 10:45 lto-dump ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/lto-dump
> > > lrwxrwxrwx 1 root root 63 Feb 4 10:45
> x86_64-pc-linux-gnu-c++
> > > -> /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-c++
> > > lrwxrwxrwx 1 root root 63 Feb 4 10:45
> x86_64-pc-linux-gnu-cpp
> > > -> /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-cpp
> > > lrwxrwxrwx 1 root root 63 Feb 4 10:45
> x86_64-pc-linux-gnu-g++
> > > -> /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-g++
> > > lrwxrwxrwx 1 root root 63 Feb 4 10:45
> x86_64-pc-linux-gnu-gcc
> > > -> /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-gcc
> > > lrwxrwxrwx 1 root root 66 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gcc-ar ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-gcc-ar
> > > lrwxrwxrwx 1 root root 66 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gcc-nm ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-gcc-nm
> > > lrwxrwxrwx 1 root root 70 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gcc-ranlib ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-gcc-ranlib
> > > lrwxrwxrwx 1 root root 65 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gccgo ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-gccgo
> > > lrwxrwxrwx 1 root root 64 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gcov ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-gcov
> > > lrwxrwxrwx 1 root root 49 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gcov-dump ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcov-dump
> > > lrwxrwxrwx 1 root root 49 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gcov-tool ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gcov-tool
> > > lrwxrwxrwx 1 root root 68 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gfortran ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/x86_64-pc-linux-gnu-gfortran
> > > lrwxrwxrwx 1 root root 45 Feb 4 10:45
> > > x86_64-pc-linux-gnu-go-10 ->
> /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/go-10
> > > lrwxrwxrwx 1 root root 48 Feb 4 10:45
> > > x86_64-pc-linux-gnu-gofmt-10 ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/gofmt-10
> > > lrwxrwxrwx 1 root root 48 Feb 4 10:45
> > > x86_64-pc-linux-gnu-lto-dump ->
> > > /usr/x86_64-pc-linux-gnu/gcc-bin/10.0.1/lto-dump
> > >
> > > > Regarding the separate ebuild, I hope someone more knowledgeable
> > > > than me regarding ebuilds can handle that.
> > >
> > > How do you plan to manage those symlinks meanwhile?
> > >
> > > I guess there is a confusion in the scope of change I want to do.
> >
> > GCC package does not provide cc and f77 binaries or symlinks. These are
> > provided by the gcc-config
> > which adds /usr/bin/cc and /usr/bin/f77.
> > These are not needed for building packages via portage but were added
> > in https://bugs.gentoo.org/412319 just because it is nice to ensure that
> > invocations like
> > $ make
> > work as is without setting CC explicitly e.g.
> > $ CC=gcc make
>
> I'm afraid "not needed" is desired but not strictly true. I keep seeing
> packages that use 'cc' accidentally (dev-util/radare2) or explicitly.
> Don't know if most of them can safely fall back to 'gcc' at ./configure
> time. I suspect many don't.
>
> > I only want to change gcc-config to NOT create /usr/bin/cc and
> /usr/bin/f77
> > via an option.
> > Everything else stays unchanged.
>
> I see. Initially you said you want to set your own 'cc' and 'f77' wrapper.
> Can you clarify what is the motivation to change only a subset of programs?
> Maybe describe whole intended setup?
>
> This is only for Chrome OS setup where I want to switch "cc" to point to
clang or a different compiler (Switching will be done using a
different ebuild not used in Gentoo).
I do not want to disturb Gentoo's current setup in any way other than
adding an option to "gcc-config"
that will serve our needs.
It feels very dangerous to have 'cc' point to one compiler (manually set)
> and 'c99' to another ('gcc').
> We will start producing more inconsistent environments than we do today.
>
> Not installing 'cc'/'f77' at all might be less bad.
>
Yes, this is exactly what I'd like but some internal teams want '$ make' to
work
as is. This option to "not install" these links will be a good first step
to find out
if there are any packages relying on 'cc' being present in Chrome OS builds.
Thanks,
Manoj
> --
>
> Sergei
>
