GCC-4.9.0

[Armin]

​http://ftp.gnu.org/gnu/gcc/gcc-4.9.0/

​http://gcc.gnu.org/gcc-4.9/changes.html

[Pierre Labastie]

Changes needed for building in chapter 5:
Add two "disables" to gcc-pass1:

--disable-libvtv --disable-libcilkrts

And binutils-pass2 needs:

--disable-werror

I have not run the tests in chapter 6 yet.

[bdubbs@…]

Chapter 6 binutils also needs --disable-werror

Chapter 6 gcc no longer has libmudflap so the mv line there needs to be taken out.

Running Chapter 6 gcc tests now.

[bdubbs@…]

My full LFS build is finishing up now, but gcc-4.9.0 completed with no unexpected errors.

Totalseconds: 10507 -- That's 2hr 55min at -j1 (1 SBU = 125 seconds, so gcc with tests takes 84 SBU. :(

Space used 1.8G

Checking glibc with this compiler, the errors I get are:

libio/tst-ftell-partial-wide.out
posix/tst-getaddrinfo4.out
posix/annexc.out
conform/run-conformtest.out

And we've seen all those before.

[bdubbs@…]

Now there is an error in building systemd. I have fixed that with adding to config.cache:

cc_cv_CFLAGSflto=no

I did find a reference in the systemd mailing list that seemed to complain that this is a gcc bug. That may be, but I take that sort of things from the systemd devs with a grain of salt. (Nothing is ever their fault.)

Otherwise, I've finished the full LFS build without other issues, but have not tested yet.

Full build with all Chapter 6 tests is 196 SBU
gcc-pass1-4.9.0 7.2 SBU
gcc-pass2-4.9.0 9.5 SBU
Ch 6 gcc-4.9.0 84.1 SBU

So just gcc is over half the total LFS build time when running all tests (-j1).

[Armin]

You can fix systemd build by using NM=gcc-nm RANLIB=gcc-ranlib too. That will make the package use -flto when building which means far longer compile times and bigger build tree (and maybe executables) just for small code speed impact. I don't think it's worth it, your workaround seems fine to me.

[bdubbs@…]

Fixed at revision 10550.