Jonny H kirjoitti 9.11.2017 klo 12:36:
> On Thursday, 9 November 2017, 8:14, Kai Ruottu
<kai.ruottu@xxxxxxxxxxx> wrote:
>> The three step builds with a "bootstrap GCC", "newlib" and "full GCC"
>> should be some remnant from the old ages. Building both GCC and
newlib (plus gmp, mpc, mpfr
>> etc) in only one step should work.
>
> Thanks, I should add a couple of things, I couldn't build the newlib
2.5 sources, I hit the same error
> as described here:
> https://sourceware.org/ml/newlib/2016/msg01117.html
My old slow 32-bit Linux PC is still producing libstdc++-v3... Forgot to
disable multilib which made it
still slower :( But producing newlib-2.5 with gcc-7.2.0 for ppc-eabi
succeeded without problems
(it didn't stop anywhere). Ok, now it did stop here :
Checking multilib configuration for libbacktrace...
Configuring in ppc-eabi/libbacktrace
configure: loading cache ./config.cache
checking build system type... i686-pc-linux-gnu
checking host system type... powerpc-unknown-eabi
checking target system type... powerpc-unknown-eabi
checking for ppc-eabi-gcc... /home/src-old/gcc-7.2.0/build/./gcc/xgcc
-B/home/src-old/gcc-7.2.0/build/./gcc/ -nostdinc
-B/home/src-old/gcc-7.2.0/build/ppc-eabi/newlib/ -isystem
/home/src-old/gcc-7.2.0/build/ppc-eabi/newlib/targ-include -isystem
/home/src-old/gcc-7.2.0/newlib/libc/include
-B/home/src-old/gcc-7.2.0/build/ppc-eabi/libgloss/rs6000
-L/home/src-old/gcc-7.2.0/build/ppc-eabi/libgloss/libnosys
-L/home/src-old/gcc-7.2.0/libgloss/rs6000 -B/opt/cross/ppc-eabi/bin/
-B/opt/cross/ppc-eabi/lib/ -isystem /opt/cross/ppc-eabi/include -isystem
/opt/cross/ppc-eabi/sys-include
checking for C compiler default output file name...
configure: error: in `/home/src-old/gcc-7.2.0/build/ppc-eabi/libbacktrace':
configure: error: C compiler cannot create executables
Of course the "target system" for ppc-eabi is something unknown so no
default startups, libgloss, linker script
etc... So no executables for it as default. Why the '--with-newlib'
didn't cause disabling all linking tests is weird.
Seemingly I must add those '--disable-multilib' and
'--disable-libbacktrace' in the GCC configure and start again.
Producing 'libssp' succeeded but you had disabled it and seemingly
had'nt problems with 'libbacktrace'. Weird...
The other thing is that my test with gcc 6.2 was actually done using crosstool-ng
http://crosstool-ng.github.io/
Which automates the cross-build. I don't know if this uses the three step build method, it may not.
Maybe this modifies the build defaults, disabling those libs which
wouldn't succeed...