Re: Cross compile crapout

[Jonathan Andrews]

On Tue, 2012-01-24 at 08:09 -0500, Chet Ramey wrote: 
> On 1/23/12 10:17 PM, Jonathan Andrews wrote:
> 
> > Im not a gcc expect, I thought the idea of the linker was also to drop
> > unused code but this seems not to be true with gcc as default.
> 
> Sure, but it's the difference between all of the libc functions you use
> and none of them, not what you use from libc and what you don't.
gcc never fails to surprise me, at the moment not in a positive way :-(

> > 
> > http://www.google.com/url?sa=t&rct=j&q=gcc%20linker%20unused%20code&source=web&cd=5&ved=0CEYQFjAE&url=http%3A%2F%2Felinux.org%2Fimages%2F2%2F2d%2FELC2010-gc-sections_Denys_Vlasenko.pdf&ei=Tx8eT5mUGczRswaZju3eDA&usg=AFQjCNF3miTnzqYNsiiyei_SQDlDxG10TQ&cad=rja
> > 
> > http://gcc.gnu.org/ml/gcc-help/2003-08/msg00128.html
> > 
> > I've been trying to compile bash static with the extra parameters, no
> > joy so far - finger trouble on my part I expect.
> > 
> > 
> >> I built a minimal bash (--enable-minimal-config) on my machine (Mac OS X,
> >> where static linking isn't possible), and it ended up being about half
> >> as big as a bash-4.2.20 build.  I had to fix a few things to do it.
> > Maybe this is one of the things ?
> 
> It is.  Just comment out the goto as a quick fix.
OK

> > 
> > I've applied the patches, tried this :
> > 
> > (script)
> > export CC=arm-softfloat-linux-gnu-gcc
> > ./configure --build=i386-linux --host=arm-linux --enable-static-link 
> > --without-bash-malloc --enable-minimal-config
> > make
> > 
> > 
> > Get this :
> > 
> > arm-softfloat-linux-gnu-gcc  -DPROGRAM='"bash"' -DCONF_HOSTTYPE='"arm"' 
> > -DCONF_OSTYPE='"linux-gnu"' -DCONF_MACHTYPE='"arm-unknown-linux-gnu"' 
> > -DCONF_VENDOR='"unknown"' -DLOCALEDIR='"/usr/local/share/locale"' 
> > -DPACKAGE='"bash"' -DSHELL -DHAVE_CONFIG_H   -I.  -I. -I./include -I./lib   
> > -g -O2 -c y.tab.c
> > ./parse.y: In function 'shell_getc':
> > ./parse.y:2399: error: label 'pop_alias' used but not defined
> > make: *** [y.tab.o] Error 1
> 
> There are a few cases like this that prevent a minimally-configured
> bash from building.  In this case, it's that there is a `goto pop_alias'
> but the pop_alias label is protected by an #ifdef.

I see bash has many tests to qualify the interpreter, could it not have
a similar thing for builds, a set of the most sensible configure/make
permutations that need to be passed after each patch ?
I'm not familiar enough to the code to offer anything sensible or I
would write some.


Another one, introduced in patch5 possibly? ... 

arm-softfloat-linux-gnu-gcc  -DPROGRAM='"bash"' -DCONF_HOSTTYPE='"arm"' 
-DCONF_OSTYPE='"linux-gnu"' -DCONF_MACHTYPE='"arm-unknown-linux-gnu"' 
-DCONF_VENDOR='"unknown"' -DLOCALEDIR='"/usr/local/share/locale"' 
-DPACKAGE='"bash"' -DSHELL -DHAVE_CONFIG_H   -I.  -I. -I./include -I./lib   -g 
-O2 -c variables.c
variables.c:4234: error: 'sv_tz' undeclared here (not in a function)
make: *** [variables.o] Error 1


Also from previous builds:

bash-4.2/bashline.c:2105: warning: Using 'endservent' in statically linked 
applications requires at runtime the shared libraries from the glibc version 
used for linking

WTF ! Words fail me, I bet some c++ lover has made this mess for
me ..... I feel so old..... take me back to 1986, a happy time ....

Is it even possible to build a statically linked bash?
I've done it before but that was in the land time forget, around kernel
2.4.24, when gcc started with a "3." and worked :-( ggrrrrr

Thanks,
Jon