Re: Compilation errors when trying to build GCC

在 2026/8/10 06:02, Hrishi via Gcc-help 写道:
> Hello!
> I've been attempting  a Linux From Scratch install as a summer project for
> university applications and what not. I've run into a little bit of a
> problem when compiling gcc however, whenever i try to compile gcc (version
> 16.2.0) on the system after chrooting in; (a cross compilation is first
> done so I do have access to a gcc compiler, not a complete one though)
>
> I have glibc version 2.43, binutils version 2.45 and a cross compiled gcc
> and libstdc++ of version 16.2.0 .
>
> here are the configure flags:
>
> ../configure --prefix=/usr            \
>               LD=ld                    \
>               --enable-languages=c,c++ \
>               --enable-default-pie     \
>               --enable-default-ssp     \
>               --enable-host-pie        \
>               --enable-targets=all     \
>               --disable-multilib       \
>               --disable-bootstrap      \
>               --disable-fixincludes    \
>               --with-system-zlib
>
>
> and below are the errors i encountered during my attempt at compilation:
> ---
> options.cc:27216:2: error: #error 0Wabsolute-value does not have a Var()
> flag
> 27216 | #error 0Wabsolute-value does not have a Var() flag

Because gawk is buggy.  If gawk-5.4.1 is built without mpfr you'll hit 
the bug.
That's why we (the LFS editors) moved ch.08 gawk earlier than GCC.

> I've attempted to ask for help in the #lfs-support irc channel, but felt it
> would probably be well to also request help from the GCC devs themselves.

No, it contradicts with what we say in the book
(https://www.linuxfromscratch.org/lfs/view/systemd/chapter01/askforhelp.html):

    In case you've hit an issue building one package with the LFS
    instruction, we strongly
    discourage posting the issue directly onto the upstream support
    channel before
    discussing via a LFS support channel listed in Section 1.4,
    “Resources.” Doing so is
    often quite inefficient because the upstream maintainers are rarely
    familiar with LFS
    building procedure. Even if you've really hit an upstream issue, the
    LFS community
    can still help to isolate the information wanted by the upstream
    maintainers and
    make a proper report.

And this incident itself proves the points: this is actually a gawk 
issue instead of
gcc issue; the gawk issue only manifests if it's not built with mpfr 
(and such a build
does not occur frequently out of LFS so the GCC maintainers are 
generally not familiar
with such a build); and the gawk issue was firstly identified by the LFS 
manage editor
and he has already worked it around for the book.

Apparently you are reading the instructions from an old LFS release but 
changing
the gawk version on your own (because we don't have a revision where 
gawk is 5.4.1
but it's not moved before GCC in chapter 8).  This is not good for a 
first-time builder
and we already state that.  In the book
(https://www.linuxfromscratch.org/lfs/view/systemd/partintro/generalinstructions.html):

    During a development cycle of LFS, the instructions in the book are
    often modified
    to adapt for a package update or take the advantage of new features
    from updated
    packages. Mixing up the instructions of different versions of the
    LFS book can cause
    subtle breakages.

And in the FAQ (https://www.linuxfromscratch.org/faq/#why-not-version):

    If this is your first time building LFS, using a version not in the
    book or varying from the
    book in any way is not a good idea. The IRC channel regulars have a
    saying, "FBBG",
    meaning "Follow Book, Book Good." They and the people on the lists
    have helped many
    an unhappy newbie who deviated from the book during that first build.

Now you can see we state things in the book for a reason... Please do 
not skip them.