gcc 4.9.1 and stdint.h

gcc 4.9.1 and stdint.h

[Holger Hans Peter Freyther]

Hi,

I have a small piece of initial bootloader and it fails to compile
using master (f19b4e995ea47f9243f152b39337330307453c9f). The code
does include "stdint.h" which ends up with a failure like this:

tmp/sysroots/i686-linux/usr/lib/arm-poky-linux-gnueabi/gcc/arm-poky-linux-gnueabi/4.9.1/include/stdint.h:9:26: fatal error: stdint.h: No such file or directory
|  # include_next <stdint.h>
|                           ^
| compilation terminated.


So somehow gcc doesn't know how to find the glibc version of
stdint.h. With edison and dora the #include_next will move to
the sysroot.

Is this a regression of gcc-4.9? is this a wanted change?

kind regards
	holger

Re: gcc 4.9.1 and stdint.h

[Richard Purdie]

On Sun, 2014-11-02 at 21:05 +0100, Holger Hans Peter Freyther wrote:
> Hi,
> 
> I have a small piece of initial bootloader and it fails to compile
> using master (f19b4e995ea47f9243f152b39337330307453c9f). The code
> does include "stdint.h" which ends up with a failure like this:
> 
> tmp/sysroots/i686-linux/usr/lib/arm-poky-linux-gnueabi/gcc/arm-poky-linux-gnueabi/4.9.1/include/stdint.h:9:26: fatal error: stdint.h: No such file or directory
> |  # include_next <stdint.h>
> |                           ^
> | compilation terminated.
> 
> 
> So somehow gcc doesn't know how to find the glibc version of
> stdint.h. With edison and dora the #include_next will move to
> the sysroot.
> 
> Is this a regression of gcc-4.9? is this a wanted change?

What was the compiler commandline you used and did you specify the
sysroot? Was this was master or dizzy?

Cheers,

Richard

Re: gcc 4.9.1 and stdint.h

[Holger Hans Peter Freyther]

On Sun, Nov 02, 2014 at 10:03:35PM +0000, Richard Purdie wrote:

Good Morning,

> What was the compiler commandline you used and did you specify the
> sysroot? Was this was master or dizzy?

that was with master. The command line for the compilation is:

arm-poky-linux-gnueabi-gcc -c -Os -Wall -mabi=aapcs-linux -DDM644x -DFLASH_TYPE_NAND -Dboard_sysmobts_v2 -DENABLE_BOOT_INTERRUPT   -c -o davinci.o davinci.c

Okay the Makefile didn't honor the CC variable from the environment.
How did that work before? Did gcc have a default sysroot?

holger

Re: gcc 4.9.1 and stdint.h

[Richard Purdie]

On Mon, 2014-11-03 at 08:22 +0100, Holger Hans Peter Freyther wrote:
> On Sun, Nov 02, 2014 at 10:03:35PM +0000, Richard Purdie wrote:
> 
> Good Morning,
> 
> > What was the compiler commandline you used and did you specify the
> > sysroot? Was this was master or dizzy?
> 
> that was with master. The command line for the compilation is:
> 
> arm-poky-linux-gnueabi-gcc -c -Os -Wall -mabi=aapcs-linux -DDM644x -DFLASH_TYPE_NAND -Dboard_sysmobts_v2 -DENABLE_BOOT_INTERRUPT   -c -o davinci.o davinci.c
> 
> Okay the Makefile didn't honor the CC variable from the environment.
> How did that work before? Did gcc have a default sysroot?

Yes.

http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=678e8798ebe0f4fd1bd347db136f1499b8fe00c9

Cheers,

Richard

Re: gcc 4.9.1 and stdint.h

[Holger Hans Peter Freyther]

On Mon, Nov 03, 2014 at 08:04:59AM +0000, Richard Purdie wrote:

Dear Richard,

> http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=678e8798ebe0f4fd1bd347db136f1499b8fe00c9

thanks a lot for explaining. I see how using the default sysroot
with external toolchains is going to cause issues. What is the
reasoning to have it for the canadian cross (hence meta-toolchain)?

E.g. when compiling U-Boot linking will fail when just passing
make CROSS_COMPILE=/opt/poky/1.7/.../arm-..- (not going through
the sourcing of the environment file) the linking wil fail due a
missing -lgcc but maybe I am doing something wrong as well.

  /opt/poky/1.7/sysroots/i686-pokysdk-linux/usr/bin/arm-poky-linux-gnueabi/arm-poky-linux-gnueabi-ld.bfd   -g -Ttext 0xc100000 -o examples/standalone/hello_world -e hello_world examples/standalone/hello_world.o examples/standalone/libstubs.o -L . -lgcc
/opt/poky/1.7/sysroots/i686-pokysdk-linux/usr/bin/arm-poky-linux-gnueabi/arm-poky-linux-gnueabi-ld.bfd: cannot find -lgcc
gpl-uboot/examples/standalone/Makefile:67: recipe for target 'examples/standalone/hello_world' failed
make[2]: *** [examples/standalone/hello_world] Error 1


holger

Re: gcc 4.9.1 and stdint.h

[Richard Purdie]

On Mon, 2014-11-03 at 16:12 +0100, Holger Hans Peter Freyther wrote:
> On Mon, Nov 03, 2014 at 08:04:59AM +0000, Richard Purdie wrote:
> 
> Dear Richard,
> 
> > http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=678e8798ebe0f4fd1bd347db136f1499b8fe00c9
> 
> thanks a lot for explaining. I see how using the default sysroot
> with external toolchains is going to cause issues. What is the
> reasoning to have it for the canadian cross (hence meta-toolchain)?

The reasoning is that we have one canadian-cross compiler targeting
multiple sysroots (one arm compiler targets all arm machines). The
compiled in sysroot may not be the final installation path anyway,
depending on where you install meta-toolchain.

We prefer deterministic builds rather than some toolchain installations
that work and some that don't.

Cheers,

Richard

Re: gcc 4.9.1 and stdint.h

[Holger Hans Peter Freyther]

On Mon, Nov 03, 2014 at 03:37:09PM +0000, Richard Purdie wrote:

> We prefer deterministic builds rather than some toolchain installations
> that work and some that don't.

Do you intend to create a script like the current environment that
could work with "bare" apps (like u-boot, linux kernel, initial
bootloaders, etc.)?

holger

Re: gcc 4.9.1 and stdint.h

[Richard Purdie]

On Fri, 2014-11-07 at 17:58 +0100, Holger Hans Peter Freyther wrote:
> On Mon, Nov 03, 2014 at 03:37:09PM +0000, Richard Purdie wrote:
> 
> > We prefer deterministic builds rather than some toolchain installations
> > that work and some that don't.
> 
> Do you intend to create a script like the current environment that
> could work with "bare" apps (like u-boot, linux kernel, initial
> bootloaders, etc.)?

I'd like to understand what differences such an environment script would
have but would be happy to consider it...

Cheers,

Richard

Re: gcc 4.9.1 and stdint.h

[Holger Hans Peter Freyther]

On Wed, Nov 12, 2014 at 01:21:11PM +0000, Richard Purdie wrote:

> I'd like to understand what differences such an environment script would
> have but would be happy to consider it...

I am not sure. My question is how to compile a kernel with the
result of meta-toolchain? Until this commit setting CROSS_COMPILE=/opt/..-
was enough. But now my users probably need to pass CC and LD as
variables as well? The current environment file passes -Wl, to
the linker which doesn't work with the kernel compilation. So I
wondered if one could create a script with enough variables to
easily compile things like UBL, U-Boot, Linux?