ftbfs: gnat cross targeting powerpc

[Matthias Klose]

both seen with gcc-8-cross and gcc-9-cross (search the build logs for "unfinished":

/<<PKGBUILDDIR>>/gcc/build/./gcc/xgcc -B/<<PKGBUILDDIR>>/gcc/build/./gcc/ -B/usr/powerpc-linux-gnu/bin/ -B/usr/powerpc-li
nux-gnu/lib/ -isystem /usr/powerpc-linux-gnu/include -isystem /usr/powerpc-linux-gnu/sys-include -isystem /<<PKGBUILDDIR>
>/gcc/build/sys-include -c -g -O2 -W -Wall -gnatpg -nostdinc -gnatn a-direct.adb -o a-direct.o
a-direct.adb:734:27: "NAME_MAX" is undefined
a-direct.adb:743:28: "SIZEOF_struct_dirent_alloc" is undefined
a-direct.adb:1210:24: "ENOENT" is undefined
s-filatt.ads:62:18: "SIZEOF_struct_file_attributes" not declared in "OS_Constants"
s-oscons.ads:50:26: "Target_OS" is undefined
s-oscons.ads:54:01: (style) multiple blank lines
s-oscons.ads:59:01: (style) multiple blank lines
s-oscons.ads:64:01: (style) multiple blank lines
s-oscons.ads:69:01: (style) multiple blank lines
s-oscons.ads:74:01: (style) multiple blank lines
s-oscons.ads:79:01: (style) multiple blank lines
s-oscons.ads:86:01: (style) multiple blank lines
s-oscons.ads:91:01: (style) multiple blank lines
s-oscons.ads:96:01: (style) multiple blank lines
s-oscons.ads:101:01: (style) multiple blank lines
s-oscons.ads:106:01: (style) multiple blank lines
s-oscons.ads:111:01: (style) multiple blank lines
s-oscons.ads:116:01: (style) multiple blank lines
s-oscons.ads:121:01: (style) multiple blank lines
s-oscons.ads:126:01: (style) multiple blank lines
s-oscons.ads:131:01: (style) multiple blank lines
s-oscons.ads:136:01: (style) multiple blank lines
s-oscons.ads:147:01: (style) multiple blank lines
s-oscons.ads:166:01: (style) multiple blank lines
s-oscons.ads:173:01: (style) multiple blank lines
s-oscons.ads:176:01: (style) multiple blank lines
s-oscons.ads:181:01: (style) multiple blank lines
make[9]: *** [../gcc-interface/Makefile:299: a-direct.o] Error 1
make[9]: Leaving directory '/<<PKGBUILDDIR>>/gcc/build/gcc/ada/rts'
make[8]: *** [gcc-interface/Makefile:623: gnatlib] Error 2

[Matthias Klose (doko)]

last successful builds were in mid June in eoan

[Matthias Klose (doko)]

not specific on the host, ftbfs on amd64 as well

[Matthias Klose (doko)]

(cd rts ; \
    /<<PKGBUILDDIR>>/gcc/build/./gcc/xgcc -B/<<PKGBUILDDIR>>/gcc/build/./gcc/ -B/usr/powerpc-linux-gnu/bin/ -B/usr/powerp
c-linux-gnu/lib/ -isystem /usr/powerpc-linux-gnu/include -isystem /usr/powerpc-linux-gnu/sys-include -isystem /<<PKGBUILD
DIR>>/gcc/build/sys-include -W -Wall -g -O2 -g -O2 -fexceptions -DIN_RTS -DHAVE_GETIPINFO -E -C -DTARGET=\"powerpc-u
nknown-linux-gnu\" -iquote /<<PKGBUILDDIR>>/gcc/src/gcc/ada /<<PKGBUILDDIR>>/gcc/src/gcc/ada/s-oscons-tmplt.c > s-oscons-
tmplt.i ; \
    /<<PKGBUILDDIR>>/gcc/build/./gcc/xgcc -B/<<PKGBUILDDIR>>/gcc/build/./gcc/ -B/usr/powerpc-linux-gnu/bin/ -B/usr/powerp
c-linux-gnu/lib/ -isystem /usr/powerpc-linux-gnu/include -isystem /usr/powerpc-linux-gnu/sys-include -isystem /<<PKGBUILD
DIR>>/gcc/build/sys-include -W -Wall -g -O2 -g -O2 -fexceptions -DIN_RTS -DHAVE_GETIPINFO -S s-oscons-tmplt.i ; \
    ../bldtools/oscons/xoscons s-oscons)
In file included from /usr/powerpc-linux-gnu/include/bits/libc-header-start.h:33,
                 from /usr/powerpc-linux-gnu/include/limits.h:26,
                 from /<<PKGBUILDDIR>>/gcc/build/gcc/include-fixed/limits.h:194,
                 from /<<PKGBUILDDIR>>/gcc/build/gcc/include-fixed/syslimits.h:7,
                 from /<<PKGBUILDDIR>>/gcc/build/gcc/include-fixed/limits.h:34,
                 from /<<PKGBUILDDIR>>/gcc/src/gcc/ada/gsocket.h:177,
                 from /<<PKGBUILDDIR>>/gcc/src/gcc/ada/s-oscons-tmplt.c:103:
/usr/powerpc-linux-gnu/include/features.h:185:3: warning: #warning "_BSD_SOURCE and _SVID_SOURCE are deprecated, use _DEF
AULT_SOURCE" [-Wcpp]
 # warning "_BSD_SOURCE and _SVID_SOURCE are deprecated, use _DEFAULT_SOURCE"
   ^~~~~~~
/<<PKGBUILDDIR>>/gcc/src/gcc/ada/s-oscons-tmplt.c: In function 'main':
/<<PKGBUILDDIR>>/gcc/src/gcc/ada/s-oscons-tmplt.c:1256:46: error: 'SO_SNDTIMEO_OLD' undeclared (first use in this function)
 CND(SO_SNDTIMEO, "Emission timeout")
                                              ^
/<<PKGBUILDDIR>>/gcc/src/gcc/ada/s-oscons-tmplt.c:1256:46: note: each undeclared identifier is reported only once for each function it appears in
/<<PKGBUILDDIR>>/gcc/src/gcc/ada/s-oscons-tmplt.c:1261:46: error: 'SO_RCVTIMEO_OLD' undeclared (first use in this function)
 CND(SO_RCVTIMEO, "Reception timeout")
                                              ^
test -f rts/s-oscons.ads || exit 1

[...]

/<<PKGBUILDDIR>>/gcc/build/./gcc/xgcc -B/<<PKGBUILDDIR>>/gcc/build/./gcc/ -B/usr/powerpc-linux-gnu/bin/ -B/usr/powerpc-li
nux-gnu/lib/ -isystem /usr/powerpc-linux-gnu/include -isystem /usr/powerpc-linux-gnu/sys-include -isystem /<<PKGBUILDDIR>
>/gcc/build/sys-include -c -g -O2 -W -Wall -gnatpg -nostdinc -gnatn a-direct.adb -o a-direct.o
checking for sinl... a-direct.adb:722:27: "NAME_MAX" is undefined
a-direct.adb:731:28: "SIZEOF_struct_dirent_alloc" is undefined
a-direct.adb:1099:24: "ENOENT" is undefined
s-filatt.ads:62:18: "SIZEOF_struct_file_attributes" not declared in "OS_Constants"
s-oscons.ads:50:26: "Target_OS" is undefined
s-oscons.ads:54:01: (style) multiple blank lines
s-oscons.ads:59:01: (style) multiple blank lines
s-oscons.ads:64:01: (style) multiple blank lines
s-oscons.ad...

[Seth Forshee (sforshee)]

SO_SNDTIMEO_OLD and SO_RCVTIMEO_OLD are new in 5.1, due to some shuffling to support y2038-safe socket timeout options. They should be defined in headers installed by linux-libc-dev, as follows:

usr/include/asm-generic/socket.h:
  #define SO_RCVTIMEO (sizeof(time_t) == sizeof(__kernel_long_t) ? SO_RCVTIMEO_OLD : SO_RCVTIMEO_NEW)
  #define SO_SNDTIMEO (sizeof(time_t) == sizeof(__kernel_long_t) ? SO_SNDTIMEO_OLD : SO_SNDTIMEO_NEW)

usr/include/powerpc64le-linux-gnu/asm/socket.h:
  #define SO_RCVTIMEO_OLD 18
  #define SO_SNDTIMEO_OLD 19

So it seems likely that the would be coming from the SO_RCVTIMEO/SO_SNDTIMEO definitions, which should get pulled in from the ppc asm header pulling in the asm-generic header. But I can't see how it would end up with those definitions but not the definitions for the corresponding *_OLD symbols.

Looking at kernel changes since 5.0, as far as I can tell nothing has changed wrt NAME_MAX and ENOENT. The other symbols don't appear to be from the kernel.

[Matthias Klose (doko)]

comparing the source trees of 5.0 and 5.2, it looks like all architectures except powerpc were updated (fgrep -r SO_RCVTIMEO). So this is a blocker for the powerpc cross compilers, which we need to build some firmware.

[Seth Forshee (sforshee)]

That's not true; most architectures don't have a match for SO_RCVTIMEO, as they're using the definition from asm-generic, just as powerpc should be. The others were updated differently because they do not include asm-generic/socket.h. And powerpc clearly does have a definition for it, since the build complains that SO_RCVTIMEO_OLD is not defined and not SO_RCVTIMEO. The question is why isn't SO_RCVTIMEO_OLD defined.

I tried cross-compiling the following for ppc64el and did not receive any errors about SO_RCVTIMEO or SO_RCVTIMEO_OLD being undefined, nor did it complain when I used SO_RCVTIMEO_OLD, so it seems like it's working as intended. That leaves me wondering whether something isn't including asm-generic/socket.h when it should be including asm/socket.h, or something like that. I don't see any such mistakes in linux-libc-dev though.

#include <stdio.h>
#include <sys/socket.h>

int main(void)
{
        printf("%d\n", SO_RCVTIMEO);
        return 0;
}

[Matthias Klose (doko)]

gcc/ada/gsocket.h:#include <sys/socket.h>

so this is using the glibc header, no kernel header directly.

It looks like that with linux-libc-dev built using linux 5.2, I'm able to build the cross compiler. I'll update once the build finishes.

[Matthias Klose (doko)]

> /usr/include/powerpc64le-linux-gnu/asm/socket.h:

this is the wrong header, I am talking about the powerpc-linux-gnu target.

[Matthias Klose (doko)]

now built linux-libc-dev-powerpc-cross from the linux 5.2 sources, and the cross builds are working again. I'm currently not spending time to investigate that further.

[Steve Langasek (vorlon)]

If the cross-builds are working again, it sounds like this is resolved.