chap-5.10.1 gcc unneeded reference to sysv4.h

[Gilles Espinasse]

gcc/config/rs6000/sysv4.h contain the define for the linker GLIBC_DYNAMIC_LINKER "/lib/ld.so.1" But gcc/config/i386/sysv4.h does not contain that define.

LFS does not support powerpc, so that's useless to adjust the linker in rs6000/sysv4.h file.

Greg only add STANDARD_INCLUDE_DIR and STANDARD_STARTFILE_PREFIX_... in one DL_HEADER file, so gcc/config/i386/sysv4.h need no changes.

I would suggest to refer to i386 directory directly and write

for file in $(find gcc/config/i386 -name linux64.h -o -name linux.h)

Similary for unsetting MULTILIBS on x86_64, there is only one file to change, so a for loop is a big luxury. That could be written directly with

file=gcc/config/i386/t-linux64; \
cp -v $file{,.orig}
sed '/MULTILIB_OSDIRNAMES/d' $file.orig > $file

[Bryan Kadzban]

Replying to gespinasse:

gcc/config/rs6000/sysv4.h contain the define for the linker GLIBC_DYNAMIC_LINKER "/lib/ld.so.1" But gcc/config/i386/sysv4.h does not contain that define.

Um... so? :-)

LFS does not support powerpc, so that's useless to adjust the linker in rs6000/sysv4.h file.

Oh, I see.

This was the result of making the book a bit friendlier to other architectures a while ago, when it didn't overly complicate things. (This is also why the text in the "toolchain technical notes" section is not specific to the x86 32-bit linker, but is generic when that's easy.)

I would rather not change this back to be x86/32-bit specific. It's not very complicated the way it is, and we hit all the possible linker paths this way. No, the book doesn't *support* other architectures, but it might possibly *work* on some of them this way. It's up to the reader to ensure that it does (since we don't), but there's at least a chance.

(Same with the multilib comment.)

I won't set this to WONTFIX yet, though (I'll let a few other people chime in; maybe everybody else wants to revert to x86-and-32-bit-only).

[Matthew Burgess]

Replying to bryan@…:

I won't set this to WONTFIX yet, though (I'll let a few other people chime in; maybe everybody else wants to revert to x86-and-32-bit-only).

I think the current plan is to try and get a bit firmer support for architectures other than x86-and-32-bit-only into the book for 7.0. Therefore, I don't think it's worth changing this stuff now. As Bryan says, it may just work on other archictectures now, we just don't make any claims to that effect.

[bdubbs@…]

Marked as WONTFIX for the above reasons.