icu @72.1 does not build on PPC Tiger, Mac OS X 10.4.11, because of problems with assembly code

[ballapete (Peter "Pete" Dyballa)]

gnumake[2]: Leaving directory `/opt/local/var/macports/build/_opt_local_var_macports_sources_nue.de.rsync.macports.org_macports_release_tarballs_ports_devel_icu/icu/work/icu/source/data'
/bin/sh ../mkinstalldirs ./out
DYLD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$DYLD_LIBRARY_PATH  ../bin/icupkg -tb ./in/icudt72l.dat ./out/icudt72b.dat
/bin/sh ../mkinstalldirs ./out/tmp ./out/build/icudt72b
mkdir ./out/tmp
mkdir ./out/build
mkdir ./out/build/icudt72b
Unpacking ./out/icudt72b.dat and generating out/tmp/icudata.lst (list of data files)
DYLD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$DYLD_LIBRARY_PATH  ../bin/icupkg -d ./out/build/icudt72b --list -x \* ./out/icudt72b.dat -o out/tmp/icudata.lst
echo timestamp > build-local
DYLD_LIBRARY_PATH=../stubdata:../tools/ctestfw:../lib:$DYLD_LIBRARY_PATH  ../bin/pkgdata -O ../data/icupkg.inc -q -c -s /opt/local/var/macports/build/_opt_local_var_macports_sources_nue.de.rsync.macports.org_macports_release_tarballs_ports_devel_icu/icu/work/icu/source/data/out/build/icudt72b -d ../lib -e icudt72  -T ./out/tmp -p icudt72b -L icudata -m static -r 72.1 ./out/tmp/icudata.lst
./out/tmp/icudt72b_dat.S:7:Unknown pseudo-op: .balign
./out/tmp/icudt72b_dat.S:7:Rest of line ignored. 1st junk character valued 49 (1).
-- return status = 256
Error creating with assembly code. Failed command: /opt/local/bin/gcc-mp-7 -DU_ATTRIBUTE_DEPRECATED=   -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1  -pipe -Os -arch ppc -std=c11 -Wall -pedantic -Wshadow -Wpointer-arith -Wmissing-prototypes -Wwrite-strings   -fno-common -c -I../common -I../common  -dynamic -o ./out/tmp/icudt72b_dat.o ./out/tmp/icudt72b_dat.S
Error generating assembly code for data.
pkgdata: /opt/local/bin/gcc-mp-7 -DU_ATTRIBUTE_DEPRECATED=   -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1  -pipe -Os -arch ppc -std=c11 -Wall -pedantic -Wshadow -Wpointer-arith -Wmissing-prototypes -Wwrite-strings   -fno-common -c -I../common -I../common  -dynamic -o ./out/tmp/icudt72b_dat.o ./out/tmp/icudt72b_dat.S
gnumake[1]: *** [packagedata] Error 1

[ballapete (Peter "Pete" Dyballa)]

Main.log from Tiger PPC, Mac OS X 10.4.11

[catap (Kirill A. Korinsky)]

Well, seems that as is too old to build it. Switching to .align may introduce some bugs because its behaviour is depend on hardware for GCC. See: ​https://sourceware.org/binutils/docs/as/Pseudo-Ops.html

I have a crazy idea to use gcc10-bootstrap as compiler for 10.4. May I ask you to try to build gcc10-bootstrap port?

[ballapete (Peter "Pete" Dyballa)]

In a few days, after finishing as much updates as possible (by telling MacPorts that there is no icu upgrade), I could try. The PowerBook can compile in the background and heat my apartment fine…

BTW, I have also nasm @2.15.05 and yasm @1.3.0 installed…

[catap (Kirill A. Korinsky)]

Peter, when you may try a patch like that:

diff --git a/devel/icu/Portfile b/devel/icu/Portfile
index 638d1d87fb7..aa2a13bd8f2 100644
--- a/devel/icu/Portfile
+++ b/devel/icu/Portfile
@@ -84,9 +84,14 @@ if {${subport} eq ${name} || ${subport} eq "${name}-lx"} {
     configure.cppflags-delete -I${prefix}/include
     configure.ldflags-delete  -L${prefix}/lib
 
+    configure.env-append \
+                        AS=${prefix}/bin/yasm
+    building.env-append AS=${prefix}/bin/yasm
+
     configure.args      --disable-layoutex \
                         --disable-samples \
-                        --enable-static
+                        --enable-static \
+                        AS=${prefix}/bin/yasm
 
     configure.universal_args-delete --disable-dependency-tracking
 

It might work...

[kencu (Ken)]

​http://yasm.tortall.net/

"Yasm currently supports the x86 and AMD64 instruction sets"

[kencu (Ken)]

There are only two options, I believe.

1. change ".balign" to ".align"
2. use Iains new "llas" PowerPC assembler. This is a big project, as it requires pulling in his llvm-7.1-for-PPC project, which he uses for the back-end for the new "llas" assembler he wrote. And that project, the llvm-7.1-for-PPC project, is not necessarily something we want in MacPorts right now as it is unfinished and raw still. Unfinished raw things just cause 100,000 tickets without resolutions, you may have noticed.

[catap (Kirill A. Korinsky)]

Ken, but he said that it is installed (and I've assumed that it works! :) )

Peter, a patch to change .balign to .align is trivial and you need to change it at the only one place: source/tools/toolutil/pkg_genc.cpp but I really not sure about side effects that it may introduce. Probably none, but I really not sure.

[kencu (Ken)]

I guess yasm would probably work on PPC, if it installed, but still, it can only build x86 and AMD64 assembly.

[ballapete (Peter "Pete" Dyballa)]

Replying to catap:

Peter, when you may try a patch like that:

+ building.env-append AS=${prefix}/bin/yasm

}}}

This is illegal, port says… (Could build.args-append, build.env-append, or configure.env be more approriate?)

Configure does not check for Assembler. No Makefile in build directory tree contains the word yasm.

BTW, shouldn't the word build on line #23 of Portfile be built?

[ballapete (Peter "Pete" Dyballa)]

I think the file in question was compiled. TWICE:

/opt/local/bin/g++-mp-7 -DU_ATTRIBUTE_DEPRECATED= -DU_TOOLUTIL_IMPLEMENTATION -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1 -I. -I../../common -I../../i18n "-DU_BUILD=\"powerpc-apple-darwin8.11.0\"" "-DU_HOST=\"powerpc-apple-darwin8.11.0\"" "-DU_CC=\"/opt/local/bin/gcc-mp-7\"" "-DU_CXX=\"/opt/local/bin/g++-mp-7\"" -DUNISTR_FROM_CHAR_EXPLICIT=explicit -DUNISTR_FROM_STRING_EXPLICIT=explicit -pipe -Os -D_GLIBCXX_USE_CXX11_ABI=0 -arch ppc -W -Wall -pedantic -Wpointer-arith -Wwrite-strings -Wno-long-long -std=c++11 -fvisibility=hidden -fno-common -c -MMD -MT "pkg_genc.d pkg_genc.o pkg_genc.ao" -o pkg_genc.ao pkg_genc.cpp
pkg_genc.cpp: In function 'uint32_t write32(FileStream*, uint32_t, uint32_t)':
pkg_genc.cpp:584:23: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
         for (i = 0; i < sizeof(uint32_t); i++)
                     ~~^~~~~~~~~
/opt/local/bin/g++-mp-7 -DU_ATTRIBUTE_DEPRECATED= -DU_TOOLUTIL_IMPLEMENTATION -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1 -I. -I../../common -I../../i18n "-DU_BUILD=\"powerpc-apple-darwin8.11.0\"" "-DU_HOST=\"powerpc-apple-darwin8.11.0\"" "-DU_CC=\"/opt/local/bin/gcc-mp-7\"" "-DU_CXX=\"/opt/local/bin/g++-mp-7\"" -DUNISTR_FROM_CHAR_EXPLICIT=explicit -DUNISTR_FROM_STRING_EXPLICIT=explicit -pipe -Os -D_GLIBCXX_USE_CXX11_ABI=0 -arch ppc -W -Wall -pedantic -Wpointer-arith -Wwrite-strings -Wno-long-long -std=c++11 -fvisibility=hidden -fno-common -c -MMD -MT "pkg_gencmn.d pkg_gencmn.o pkg_gencmn.ao" -o pkg_gencmn.ao pkg_gencmn.cpp

and then

/opt/local/bin/g++-mp-7 -DU_ATTRIBUTE_DEPRECATED= -DU_TOOLUTIL_IMPLEMENTATION -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1 -I. -I../../common -I../../i18n "-DU_BUILD=\"powerpc-apple-darwin8.11.0\"" "-DU_HOST=\"powerpc-apple-darwin8.11.0\"" "-DU_CC=\"/opt/local/bin/gcc-mp-7\"" "-DU_CXX=\"/opt/local/bin/g++-mp-7\"" -DUNISTR_FROM_CHAR_EXPLICIT=explicit -DUNISTR_FROM_STRING_EXPLICIT=explicit -pipe -Os -D_GLIBCXX_USE_CXX11_ABI=0 -arch ppc -W -Wall -pedantic -Wpointer-arith -Wwrite-strings -Wno-long-long -std=c++11 -fvisibility=hidden -fno-common -c -dynamic -MMD -MT "pkg_genc.d pkg_genc.o pkg_genc.ao" -o pkg_genc.o pkg_genc.cpp
pkg_genc.cpp: In function 'uint32_t write32(FileStream*, uint32_t, uint32_t)':
pkg_genc.cpp:584:23: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
         for (i = 0; i < sizeof(uint32_t); i++)
                     ~~^~~~~~~~~
/opt/local/bin/g++-mp-7 -DU_ATTRIBUTE_DEPRECATED= -DU_TOOLUTIL_IMPLEMENTATION -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1 -I. -I../../common -I../../i18n "-DU_BUILD=\"powerpc-apple-darwin8.11.0\"" "-DU_HOST=\"powerpc-apple-darwin8.11.0\"" "-DU_CC=\"/opt/local/bin/gcc-mp-7\"" "-DU_CXX=\"/opt/local/bin/g++-mp-7\"" -DUNISTR_FROM_CHAR_EXPLICIT=explicit -DUNISTR_FROM_STRING_EXPLICIT=explicit -pipe -Os -D_GLIBCXX_USE_CXX11_ABI=0 -arch ppc -W -Wall -pedantic -Wpointer-arith -Wwrite-strings -Wno-long-long -std=c++11 -fvisibility=hidden -fno-common -c -dynamic -MMD -MT "pkg_gencmn.d pkg_gencmn.o pkg_gencmn.ao" -o pkg_gencmn.o pkg_gencmn.cpp

It's -c -MMD -MT "pkg_genc.d pkg_genc.o pkg_genc.ao" -o pkg_genc.ao pkg_genc.cpp vs. -c -dynamic -MMD -MT "pkg_genc.d pkg_genc.o pkg_genc.ao" -o pkg_genc.o pkg_genc.cpp – a fault? Not capable to find a decision for either this or that? First output file is pkg_genc.ao and second output file is pkg_genc.o

But in the end I see again:

gnumake[2]: Leaving directory `/opt/local/var/macports/build/_opt_local_var_macports_sources_nue.de.rsync.macports.org_macports_release_tarballs_ports_devel_icu/icu/work/icu/source/data'
/bin/sh ../mkinstalldirs ./out
DYLD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$DYLD_LIBRARY_PATH  ../bin/icupkg -tb ./in/icudt72l.dat ./out/icudt72b.dat
/bin/sh ../mkinstalldirs ./out/tmp ./out/build/icudt72b
mkdir ./out/tmp
mkdir ./out/build
mkdir ./out/build/icudt72b
Unpacking ./out/icudt72b.dat and generating out/tmp/icudata.lst (list of data files)
DYLD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$DYLD_LIBRARY_PATH  ../bin/icupkg -d ./out/build/icudt72b --list -x \* ./out/icudt72b.dat -o out/tmp/icudata.lst
echo timestamp > build-local
DYLD_LIBRARY_PATH=../stubdata:../tools/ctestfw:../lib:$DYLD_LIBRARY_PATH  ../bin/pkgdata -O ../data/icupkg.inc -q -c -s /opt/local/var/macports/build/_opt_local_var_macports_sources_nue.de.rsync.macports.org_macports_release_tarballs_ports_devel_icu/icu/work/icu/source/data/out/build/icudt72b -d ../lib -e icudt72  -T ./out/tmp -p icudt72b -L icudata -m static -r 72.1 ./out/tmp/icudata.lst
./out/tmp/icudt72b_dat.S:7:Unknown pseudo-op: .balign
./out/tmp/icudt72b_dat.S:7:Rest of line ignored. 1st junk character valued 49 (1).
-- return status = 256
Error creating with assembly code. Failed command: /opt/local/bin/gcc-mp-7 -DU_ATTRIBUTE_DEPRECATED=   -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1  -pipe -Os -arch ppc -std=c11 -Wall -pedantic -Wshadow -Wpointer-arith -Wmissing-prototypes -Wwrite-strings   -fno-common -c -I../common -I../common  -dynamic -o ./out/tmp/icudt72b_dat.o ./out/tmp/icudt72b_dat.S
Error generating assembly code for data.
pkgdata: /opt/local/bin/gcc-mp-7 -DU_ATTRIBUTE_DEPRECATED=   -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1  -pipe -Os -arch ppc -std=c11 -Wall -pedantic -Wshadow -Wpointer-arith -Wmissing-prototypes -Wwrite-strings   -fno-common -c -I../common -I../common  -dynamic -o ./out/tmp/icudt72b_dat.o ./out/tmp/icudt72b_dat.S
gnumake[1]: *** [packagedata] Error 1

[ballapete (Peter "Pete" Dyballa)]

Replying to catap:

I have a crazy idea to use gcc10-bootstrap as compiler for 10.4. May I ask you to try to build gcc10-bootstrap port?

Probably this evening, 18:00 UTC, all ports that can be upgraded or that can be fixed by me, will be built. What should I be aware of when trying to build gcc10-bootstrap and some more?

[catap (Kirill A. Korinsky)]

Peter as far as I know nobody tries to build gcc10-bootstrap on 10.4 PPC. It may be build, and if yes, you may try to use it to build icu-72; I doubt that it help, but anyway interesting to ask :)

I plan to make a patch for ICU that should replaces a .balign to .align and prepare a branch to icu-devel that adds this patch, after that I'll ask you to test it. It should allow to build it, and probably it will work fine ;)

[ballapete (Peter "Pete" Dyballa)]

The build has started and reached stage1.

Would it make sense to open a new case where I can report observations? And possible failures?

[catap (Kirill A. Korinsky)]

Peter, if gcc10-bootstrap is failed, yes, I'd like to ask you to open an issue for me.

[ballapete (Peter "Pete" Dyballa)]

After almost 50 hours the build succeeded three hours ago. (Log is saved.)

[catap (Kirill A. Korinsky)]

Peter, thanks for testing!

May I ask you to try build icu-devel from this branch? ​https://github.com/catap/macports-ports/tree/icu-devel

I've switched from .balign to .align for the case of macOS on PPC before 10.5 It may work.

Thanks.

P.S. it is a kind of blind shot, I haven't tested it

[ballapete (Peter "Pete" Dyballa)]

I tried to run gmake check from this port's "home directory" – only FAILures were reported. (Because conftest.dSYM directories are left?)

Right now icu-devel/Portfile has a date from two weeks ago. Is this usable?

[catap (Kirill A. Korinsky)]

I see. I haven't created a PR to MacPorts ports tree because I haven't tested it yet :)

So, it exists only as a branch in my fork of MacPorts ports tree.

To make things easy may you try this patch ​https://raw.githubusercontent.com/catap/macports-ports/091c149c8ba07cca682a1008fe4d866ebdeb3b59/devel/icu-devel/files/patch-balign-to-align.diff to any used icu port?

[kencu (Ken)]

it would be unexpected to find the "gas" assembler on Leopard PPC can accept ".balign".

[ballapete (Peter "Pete" Dyballa)]

The build of icu-devel has not yet reached the patched pkg_genc.cpp, but it throws complaints as the build of gcc10-bootstrap did:

/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'm' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: Flag option -m has already been seen!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'm' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'a' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'c' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: Flag option -o has already been seen!

So there must be some "typo" while/from building gcc10-bootstrap. Before all Assembler code was compiled invisibly.

[kencu (Ken)]

That happens because of a flag the ancient cctools on Tiger doesn’t understand. Usually causes no failures, but clutters output w the warning.

[ballapete (Peter "Pete" Dyballa)]

Finally a build failure for icu-devel:

gnumake[2]: Leaving directory `/opt/local/var/macports/build/_opt_local_var_macports_sources_nue.de.rsync.macports.org_macports_release_tarballs_ports_devel_icu-devel/icu-devel/work/icu/source/data'
/bin/sh ../mkinstalldirs ./out
DYLD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$DYLD_LIBRARY_PATH  ../bin/icupkg -tb ./in/icudt72l.dat ./out/icudt72b.dat
/bin/sh ../mkinstalldirs ./out/tmp ./out/build/icudt72b
mkdir ./out/tmp
mkdir ./out/build
mkdir ./out/build/icudt72b
Unpacking ./out/icudt72b.dat and generating out/tmp/icudata.lst (list of data files)
DYLD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$DYLD_LIBRARY_PATH  ../bin/icupkg -d ./out/build/icudt72b --list -x \* ./out/icudt72b.dat -o out/tmp/icudata.lst
echo timestamp > build-local
DYLD_LIBRARY_PATH=../stubdata:../tools/ctestfw:../lib:$DYLD_LIBRARY_PATH  ../bin/pkgdata -O ../data/icupkg.inc -q -c -s /opt/local/var/macports/build/_opt_local_var_macports_sources_nue.de.rsync.macports.org_macports_release_tarballs_ports_devel_icu-devel/icu-devel/work/icu/source/data/out/build/icudt72b -d ../lib -e icudt72  -T ./out/tmp -p icudt72b -L icudata -m static -r 72.1 ./out/tmp/icudata.lst
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'm' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: Flag option -m has already been seen!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'm' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'a' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'c' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: Flag option -o has already been seen!
./out/tmp/icudt72b_dat.S:7:77: error: token "." is not valid in preprocessor expressions
    7 | #if defined(__ppc__) &&__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ < 1050 .align 16
      |                                                                             ^
./out/tmp/icudt72b_dat.S:8:7: warning: extra tokens at end of #else directive [-Wendif-labels]
    8 | #else .balign 16
      |       ^
./out/tmp/icudt72b_dat.S:7: error: unterminated #else
    7 | #if defined(__ppc__) &&__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ < 1050 .align 16
      | 
-- return status = 256
Error creating with assembly code. Failed command: /opt/local/libexec/gcc10-bootstrap/bin/gcc -DU_ATTRIBUTE_DEPRECATED=   -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1  -pipe -Os -arch ppc -std=c11 -Wall -pedantic -Wshadow -Wpointer-arith -Wmissing-prototypes -Wwrite-strings   -fno-common -c -I../common -I../common  -dynamic -o ./out/tmp/icudt72b_dat.o ./out/tmp/icudt72b_dat.S
Error generating assembly code for data.
pkgdata: /opt/local/libexec/gcc10-bootstrap/bin/gcc -DU_ATTRIBUTE_DEPRECATED=   -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1  -pipe -Os -arch ppc -std=c11 -Wall -pedantic -Wshadow -Wpointer-arith -Wmissing-prototypes -Wwrite-strings   -fno-common -c -I../common -I../common  -dynamic -o ./out/tmp/icudt72b_dat.o ./out/tmp/icudt72b_dat.S
gnumake[1]: *** [packagedata] Error 1
gnumake[1]: Leaving directory `/opt/local/var/macports/build/_opt_local_var_macports_sources_nue.de.rsync.macports.org_macports_release_tarballs_ports_devel_icu-devel/icu-devel/work/icu/source/data'

[ballapete (Peter "Pete" Dyballa)]

Main.log for icu-devel with gcc10-bootstrap from PPC Tiger, Mac OS X 10.4.11

[catap (Kirill A. Korinsky)]

Ken, indeed, included it into the condition.

Peter, thanks for testing!

Sorry, I've forgot about \n in my patch :(

An updated patch is force-pushed to my branch or available via ​https://raw.githubusercontent.com/catap/macports-ports/29e43d7dc514e114fb074912747ba47ac5f0bf32/devel/icu-devel/files/patch-balign-to-align.diff

May I ask you to try it one more time?

Thanks.

[ballapete (Peter "Pete" Dyballa)]

Replying to kencu:

That happens because of a flag the ancient cctools on Tiger doesn’t understand. Usually causes no failures, but clutters output w the warning.

Is it possible to patch cctools to work more accurately?

[catap (Kirill A. Korinsky)]

BTW, I not sure that gcc10-bootstrap is required to build icu with this patch. I think that usual GCC should be enough :)

[ballapete (Peter "Pete" Dyballa)]

I can try that, too. Will become evening then, here…

[kencu (Ken)]

Replying to ballapete:

Replying to kencu:

That happens because of a flag the ancient cctools on Tiger doesn’t understand. Usually causes no failures, but clutters output w the warning.

Is it possible to patch cctools to work more accurately?

Done, in 2005 or so.

But this build is using the original assembler in Tiger, it seems.

[ballapete (Peter "Pete" Dyballa)]

Replying to kencu:

But this build is using the original assembler in Tiger, it seems.

Maybe I installed the wrong variant: cctools @949.0.1_2+xcode. Its /opt/local/bin/as is a shell script:

#!/bin/bash
if [[ -x /usr/bin/xcrun ]] ; then
  exec /usr/bin/xcrun as "${@}"
elif [[ -x /usr/bin/as ]] ; then
  exec /usr/bin/as "${@}"
else
  exec as "${@}"
fi

which leads to excuting `Apple Computer, Inc. version cctools-622.9~2, GNU assembler version 1.38´, the only Assembler compiler on board…

[kencu (Ken)]

Yes, you don’t want the +xcode variant on Tiger.

[kencu (Ken)]

We should add a bunch more tests for stuff like this to

port diagnose

[ballapete (Peter "Pete" Dyballa)]

The build succeeded with gcc10-bootstrap:

/opt/local/libexec/gcc10-bootstrap/bin/g++ -DU_ATTRIBUTE_DEPRECATED= -DU_TOOLUTIL_IMPLEMENTATION -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1 -I. -I../../common -I../../i18n "-DU_BUILD=\"powerpc-apple-darwin8.11.0\"" "-DU_HOST=\"powerpc-apple-darwin8.11.0\"" "-DU_CC=\"/opt/local/libexec/gcc10-bootstrap/bin/gcc\"" "-DU_CXX=\"/opt/local/libexec/gcc10-bootstrap/bin/g++\"" -DUNISTR_FROM_CHAR_EXPLICIT=explicit -DUNISTR_FROM_STRING_EXPLICIT=explicit -pipe -Os -D_GLIBCXX_USE_CXX11_ABI=0 -arch ppc -W -Wall -pedantic -Wpointer-arith -Wwrite-strings -Wno-long-long -std=c++11 -fvisibility=hidden -fno-common -c -dynamic -MMD -MT "pkg_genc.d pkg_genc.o pkg_genc.ao" -o pkg_genc.o pkg_genc.cpp
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'm' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: Flag option -m has already been seen!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'm' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'a' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: I don't understand 'c' flag!
/usr/bin/../libexec/gcc/darwin/ppc/as: Flag option -o has already been seen!
pkg_genc.cpp: In function 'uint32_t write32(FileStream*, uint32_t, uint32_t)':
pkg_genc.cpp:588:23: warning: comparison of integer expressions of different signedness: 'int32_t' {aka 'int'} and 'long unsigned int' [-Wsign-compare]
  588 |         for (i = 0; i < sizeof(uint32_t); i++)
      |                     ~~^~~~~~~~~~~~~~~~~~

[ballapete (Peter "Pete" Dyballa)]

port test … resulted in [All tests passed successfully...]. So I am going to try a build with the default compiler.

[ballapete (Peter "Pete" Dyballa)]

Replying to catap:

BTW, I not sure that gcc10-bootstrap is required to build icu with this patch. I think that usual GCC should be enough :)

Yes, it built also with GCC 7. The patch rules!

[ballapete (Peter "Pete" Dyballa)]

I think the regular port will upgrade with that patch…

[catap (Kirill A. Korinsky)]

Peter, thanks for testing.

I've created a PR for icu-devel ​https://github.com/macports/macports-ports/pull/16740 and would like to wait a few weeks before update the main ICU port, that allows folks with PPC to test it.

[ballapete (Peter "Pete" Dyballa)]

Regular icu port was upgraded thanks to the patch supplied.

[catap (Kirill A. Korinsky)]

In 2f7e934ed19f5de797a4460eade648a7d96a7d82/macports-ports (master):

icu-devel: added a patch to support PPC

It also syncronized patches with main port

See: #66258

[skip ci]

[catap (Kirill A. Korinsky)]

Peter, you may install icu-devel port instead of icu ;) it works just fine.

[catap (Kirill A. Korinsky)]

I've backported a patch to upstream: ​https://github.com/unicode-org/icu/pull/2257

And in one week I'll open a PR to main ICU port that should make things better for PPC

[ballapete (Peter "Pete" Dyballa)]

Replying to catap:

Peter, you may install icu-devel port instead of icu ;) it works just fine.

I do not see the proof:

tiger pete 224 /\ foreach PORT ( `port echo dependentof:icu` )
foreach -> port deps $PORT
foreach -> end
Full Name: boost176 @1.76.0_6+no_single+no_static+python310
Extract Dependencies: lbzip2
Build Dependencies:   gcc7
Library Dependencies: bzip2, expat, icu, libiconv, lzma, zlib, zstd, python310, libgcc
Full Name: harfbuzz-icu @5.3.1_0
Extract Dependencies: xz
Build Dependencies:   pkgconfig, gcc7
Library Dependencies: harfbuzz, icu, libgcc
Full Name: libxml2 @2.10.3_1
Extract Dependencies: xz
Build Dependencies:   pkgconfig, apple-gcc42
Library Dependencies: libiconv, icu, xz, zlib
Full Name: libxslt @1.1.37_1
Extract Dependencies: xz
Build Dependencies:   pkgconfig, apple-gcc42
Library Dependencies: icu, libiconv, libxml2, zlib
Full Name: openldap @2.6.3_4
Build Dependencies:   apple-gcc42
Library Dependencies: perl5, tcp_wrappers, cyrus-sasl2, libfetch, openssl, db48, icu, legacy-support
Test Dependencies:    coreutils
Full Name: texlive-bin @2022.62882_2+x11
Extract Dependencies: xz
Build Dependencies:   perl5, pkgconfig, gcc7
Library Dependencies: fontconfig, freetype, libpng, ncurses, zlib, libzzip, cairo, libpixman, graphite2, icu, harfbuzz, harfbuzz-icu, libpaper, gmp, mpfr, potrace, poppler, texlive-common, xorg-libXp, xpm, xorg-libXaw, xorg-libXi, libgcc
Runtime Dependencies: ghostscript

The ports that use icu do not accept icu-devel. And icu-devel has Conflicts with: icu, icu-lx, icu-docs.

[catap (Kirill A. Korinsky)]

Yes, it installs icu by default. But all ports depends on icu as path:lib/pkgconfig/icu-uc.pc:icu and not port:icu that allows to install icu-devel port instead icu ;)

[ballapete (Peter "Pete" Dyballa)]

I am building GCC 10.4 manually (./gcc-10.4.0/configure --enable-languages=c,c++,objc,obj-c++ --enable-bootstrap --with-system-zlib --without-zstd --disable-nls --disable-tls --enable-multilib --enable-multiarch --with-dwarf2 --enable-targets=i386-apple-darwin8,powerpc-apple-darwin8 --with-gmp-include=/opt/local/include --with-gmp-lib=/opt/local/lib --with-mpfr-include=/opt/local/include --with-mpfr-lib=/opt/local/lib --with-mpc-include=/opt/local/include --with-mpc-lib=/opt/local/lib), using GCC 7 and other modern tools. One of them is cctools @949.0.1. Its as understands .balign and .p2align. So it might be better to build icu with modern tools…?

[kencu (Ken)]

newer cctools does NOT understand .balign or .p2align . The assembler is dated about 2006, and not updated much since.

cctools passes assembly to clang on Intel systems, and clang does understand those, but not on ppc of course.

so you’re wasting your time with that.

If you want gcc10 Right Now, use gcc10-bootstrap, as it is complicated to build and integrate a new gcc into macports.

A serious effort to update older systems to gcc12 is just in the beginning stages now. Wait for that to finish

[ballapete (Peter "Pete" Dyballa)]

Replying to kencu:

newer cctools does NOT understand .balign or .p2align . The assembler is dated about 2006, and not updated much since.

cctools passes assembly to clang on Intel systems, and clang does understand those, but not on ppc of course.

so you’re wasting your time with that.

That's alright. I am unemployed since a few weeks, and in February retirement is awaiting me…

Ken, my Tiger runs on a PowerPC G4 processor. And when I use /opt/local/bin/as, configure (from manually building GCC 10.4 unpatched) tells me in stages 1 and 2 (hasn't come further yet) when performing its work in ./gcc:

checking assembler for .balign and .p2align... yes

instead of

checking assembler for .balign and .p2align... no

because /opt/local/bin/as is almost up-to-date:

-rw-r--r-- fredzlotnick/staff  26792 2020-02-28 19:02 cctools-949.0.1/as/as.c

(According to my calendar.)

OK, when this only works on PPC, then a patch is of course necessary in order to not discriminate intel CPUs.

Besides, my effort is to find faults in the configure scripts. I reported them from the GCC 10.3 bootstrap builds and was asked to better report those from the recent version.

To me it seems an improvement to start building GCC 10 with modern tools. In the end the installed software will be built by GCC 10 and not the system's tools. Maybe it's worth to create a variant for gcc10-bootstrap that just uses the system's installations of gmp, isl, mpfr, and mpc in order to save some disk space (there aren't so many 2,5" disks with PATA interface and more than 100 GB capacity):

   183915   2436 -rw-r--r--   2 macports admin     2493916 Nov 19  2020 distfiles/gcc/gmp-6.2.1.tar.bz2
   188589   2212 -rw-r--r--   2 macports admin     2261594 Jan  1  2022 distfiles/gcc/isl-0.24.tar.bz2
   183577    820 -rw-r--r--   2 macports admin      838731 Okt 24  2020 distfiles/gcc/mpc-1.2.1.tar.gz
   191530   1708 -rw-r--r--   1 macports admin     1747243 Nov 17 14:13 distfiles/gcc/mpfr-4.1.0.tar.bz2
   183915   2436 -rw-r--r--   2 macports admin     2493916 Nov 19  2020 distfiles/gmp/gmp-6.2.1.tar.bz2
   188589   2212 -rw-r--r--   2 macports admin     2261594 Jan  1  2022 distfiles/isl/isl-0.24.tar.bz2
   183577    820 -rw-r--r--   2 macports admin      838731 Okt 24  2020 distfiles/libmpc/mpc-1.2.1.tar.gz
   148931   1412 -rw-r--r--   1 macports admin     1441996 Feb 10  2019 distfiles/mpfr/4.0.2/mpfr-4.0.2.tar.xz
   148932   1492 -rw-r--r--   1 macports admin     1525476 Okt  3  2020 distfiles/mpfr/mpfr-4.1.0.tar.xz

Hard links created by me. (Could be cleaner, and closer to perfection, as I see now – by using more up-to-date compression.)

[kencu (Ken)]

Just for the record:

/opt/local/bin/gcc-mp-7

already uses

/opt/local/bin/as

as the assembler.

So you don't have to do anything to make it use that as the assembler.

Can we ask for you help? We need someone to try building the new gcc12/libgcc12 transition on Tiger as part of the upcoming big move. I gave up on it two years ago, as it is too much work for one person. But now there is a group, perhaps.

​https://github.com/macports/macports-ports/pull/16843

[ballapete (Peter "Pete" Dyballa)]

Replying to kencu:

Just for the record:

/opt/local/bin/gcc-mp-7

already uses

/opt/local/bin/as

as the assembler.

This did not happen when I built gcc10-bootstrap. I tried it twice…

Can we ask for you help? We need someone to try building the new gcc12/libgcc12 transition on Tiger as part of the upcoming big move. I gave up on it two years ago, as it is too much work for one person. But now there is a group, perhaps.

​https://github.com/macports/macports-ports/pull/16843

Alright! The build of GCC 10.4 will presumingly end today. Then I'll check the configure output and shall report flaws. Are a Portfile or a useful description available?

Yesterday I forgot to mention these:

   191528   1528 -rw-r--r--   1 macports admin     1564560 Nov 17 14:19 distfiles/gcc/make-3.81.tar.gz
   183068   1240 -rw-r--r--   1 macports admin     1266180 Mai 26  2020 distfiles/gmake/make-4.3.tar.lz
   191475   1256 -rw-r--r--   1 macports admin     1283355 Nov 11 00:56 distfiles/gmake/make-4.4.tar.lz

[catap (Kirill A. Korinsky)]

This did not happen when I built gcc10-bootstrap. I tried it twice…

gcc10-bootstrap is quite special things. It contains more than 80 patches and it is built in environment that prevent it to use anything from MacPorts.

[kencu (Ken)]

we are going to need to force gcc10-bootstrap to use /opt/local/bin/as at times, so hopefully the way to do that won’t prove difficult.

similar w ld

[kencu (Ken)]

icu builds and installs for me on Tiger PPC without touching a single thing in the Portfile:

$ port -v installed icu
The following ports are currently installed:
  icu @67.1_4 requested_variants='' platform='darwin 8' archs='ppc' date='2022-08-17T19:55:44-0700'
  icu @72.1_0 (active) requested_variants='' platform='darwin 8' archs='ppc' date='2022-11-28T13:32:37-0800'

It builds with the stock gcc7 currently in use for MacPorts.

So ... ? how about you clean everything up, update your ports, and try again? Make sure your system is all stock, no modified toolchains, no alternative cctools or linker versions, no forcing of any compilers.

Just stock.

[catap (Kirill A. Korinsky)]

Ken, an easy way to enforce used as or ld is create a folder which contains symlink to as/ld which you would like to use, and put this folder as the first one at PATH.

It has a few edge cases but it works :)

[kencu (Ken)]

I showed you that trick :)

It might work generally, but if you think for a moment, gcc10-bootstrap should ALREADY find the newer cctools and ld if installed then, and maybe it does, but this user indicated it did not.

TBH, I have to test these things myself to be sure.

But need to have something that is the recommended way, that people can just copy/paste and not mess up…. and that might be it…

Also, similar for clang-11-bootstrap, but there are many more tools to consider there…

[ballapete (Peter "Pete" Dyballa)]

configure --help tells that setting an environment variable AS with the correct value for configure should work. I simplified my manual build of GCC 10.4 by setting PATH accordingly, /opt/local/bin as first element.

[ballapete (Peter "Pete" Dyballa)]

GCC 10.4 has built manually (within 40 hours), the compilation log is evaluated and I can send two eMails soon. In the meantime I selfupdated port on PPC Tiger and upgraded gawk. All installed ports are according to "stock" since I upgraded Python3 to Python 3.10. But icu still wants Python 2.7, and I do not want it any more. It's not secure, it's outdated, it's past perfect. Therefore I cannot try to build icu again.

On Big Sur icu has no dependencies. Why do such exist on Tiger? (Besides GCC 7?) On Big Sur only Python 3 exists – why is this not working or sufficient on Tiger?

[kencu (Ken)]

OK, I'm going to close this ticket now, as icu builds just fine on Tiger without doing anything to any Portfiles. I will attach the log from the rebuild for you to look at.

It is something you are doing on your system that must have broken it, and to be honest, I can't follow everything you did.

Please just undo it all, and try again. ICU does not need fixing.

[ballapete (Peter "Pete" Dyballa)]

Replying to kencu:

Please just undo it all, and try again. ICU does not need fixing.

My problem seems to be that /usr/bin/as is used instead of /opt/local/bin/as, correct? The latter does understand the assembler code.

[kencu (Ken)]

/usr/bin/as will never be used by MacPorts gcc7 unless you did something to make it do that.

[ballapete (Peter "Pete" Dyballa)]

Ken, there is some twist…

I start to build with:

/usr/bin/install -c ./config/icu-config-top config/icu-config
/usr/bin/gnumake[0]: Making `all' in `stubdata'
/usr/bin/gnumake[0]: Making `all' in `common'
/usr/bin/ranlib: file: ../lib/libicuuc.a(cwchar.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicuuc.a(icuplug.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicuuc.a(restrace.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicuuc.a(wintz.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicuuc.a(cwchar.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicuuc.a(icuplug.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicuuc.a(restrace.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicuuc.a(wintz.ao) has no symbols
/usr/bin/gnumake[0]: Making `all' in `i18n'
/usr/bin/ranlib: file: ../lib/libicui18n.a(windtfmt.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicui18n.a(winnmfmt.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicui18n.a(wintzimpl.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicui18n.a(windtfmt.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicui18n.a(winnmfmt.ao) has no symbols
/usr/bin/ranlib: file: ../lib/libicui18n.a(wintzimpl.ao) has no symbols
/usr/bin/gnumake[0]: Making `all' in `io'

and you start to build on two CPU cores with:

/usr/bin/install -c ./config/icu-config-top config/icu-config
/usr/bin/gnumake[0]: Making `all' in `stubdata'
/usr/bin/gnumake[0]: Making `all' in `common'
/opt/local/bin/ranlib: file: ../lib/libicuuc.a(cwchar.ao) has no symbols
/opt/local/bin/ranlib: file: ../lib/libicuuc.a(icuplug.ao) has no symbols
/opt/local/bin/ranlib: file: ../lib/libicuuc.a(restrace.ao) has no symbols
/opt/local/bin/ranlib: file: ../lib/libicuuc.a(wintz.ao) has no symbols
/usr/bin/gnumake[0]: Making `all' in `i18n'
/opt/local/bin/ranlib: file: ../lib/libicui18n.a(windtfmt.ao) has no symbols
/opt/local/bin/ranlib: file: ../lib/libicui18n.a(winnmfmt.ao) has no symbols
/opt/local/bin/ranlib: file: ../lib/libicui18n.a(wintzimpl.ao) has no symbols
/usr/bin/gnumake[0]: Making `all' in `io'

I am not aware of any change I applied to the MacPorts build system. On Tiger my search path is: /Users/pete/bin:/opt/local/bin:/opt/local/libexec/perl5.34:/usr/X11R6/bin:/usr/local/clamXav/bin:/usr/local/bin:/bin:/usr/bin:/usr/sbin:/opt/local/sbin:/usr/local/clamXav/sbin:/usr/local/sbin:/Developer/Tools:/Users/pete/bin:/sbin:/usr/local/texlive/2009/bin:/Users/pete/bin/FDK/Tools/osx. And root, which operates MacPorts, has: /var/root/bin:/usr/libexec:/opt/local/bin:/bin:/sbin:/usr/bin:/usr/local/bin:/usr/sbin:/opt/local/bin:/opt/local/sbin.

[kencu (Ken)]

I have the stock path setup, for both user and root:

PATH=/opt/local/bin:/opt/local/sbin:/bin:/sbin:/usr/bin:/usr/sbin

I am not sure what, if anything, is in your path that may or may not cause trouble, but it is a 10 second fix to check if the stock path fixes you up, and then you'll know.

[ballapete (Peter "Pete" Dyballa)]

I think I fixed the PATH issue. Particularly GNU Emacs, in which I work with MacPorts, has a proper process environment. It's "tested" by logging out and in and by rebooting. Final proof: icu built – without the balign patch in icu-devel.

[kencu (Ken)]

that’s great!

[ballapete (Peter "Pete" Dyballa)]

When will GCC 1x test work start? In case this has some time I would boot into Leopard and clean it from Python 3.9 and old stuff.

[ballapete (Peter "Pete" Dyballa)]

It built in my corrected setup, so the ticket can be closed. (Or is there a need to contact me for GCC 1x testing?)

[catap (Kirill A. Korinsky)]

Peter, I'd like to ask your help and test ​https://github.com/macports/macports-ports/pull/16922 on your hardware.

[ballapete (Peter "Pete" Dyballa)]

Replying to catap:

Building icu-devel and testing it?

[catap (Kirill A. Korinsky)]

Yep. Thanks!

[ballapete (Peter "Pete" Dyballa)]

Up-to-date? -rw-r--r-- 1 root x11 10277 21 Nov 06:39 /opt/local/var/macports/sources/nue.de.rsync.macports.org/macports/release/tarballs/ports/devel/icu-devel/Portfile

[catap (Kirill A. Korinsky)]

Peter, may I ask you to replace patch patch-balign-to-align.diff to ​https://raw.githubusercontent.com/macports/macports-ports/dfa7bc0d278e0a26ad3fbcce5055406d700b3850/devel/icu-devel/files/patch-balign-to-align.diff ?

Thanks.

[ballapete (Peter "Pete" Dyballa)]

So I do not have to wait for a new version of Portfile but just substitute the patch-balign-to-align.diff file? And start building and testing right away?

[catap (Kirill A. Korinsky)]

Yes, can you please do that.

[ballapete (Peter "Pete" Dyballa)]

I removed the dependancy to dead Python 2.7.

[ballapete (Peter "Pete" Dyballa)]

Build started some time ago with /usr/bin/gnumake and /opt/local/bin/g++-mp-7 -std=c++11, new patch succeeded…

[ballapete (Peter "Pete" Dyballa)]

Build succeeded, and test too, OK, one hour ago. Do you want any results?

[catap (Kirill A. Korinsky)]

Nope, thanks!

I've pushed this patch to upstream.