ghostscript-10.02.0-1: error: lcms2 not found, or too old

[cooljeanius]

ghostscript fails to upgrade during its configure step:

checking for local lcms2mt library source... no
checking for local lcms2 library source... no
checking for system lcms2 library... checking for cmsCloneTransformChangingFormats in -llcms2mt... no
configure: error: lcms2 not found, or too old
### execution of /tmp/fink.JPzEc failed, exit code 1
### execution of /tmp/fink.yowoI failed, exit code 1
Removing runtime build-lock...
Removing build-lock package...
/opt/sw/bin/dpkg-lockwait -r fink-buildlock-ghostscript-10.02.0-1
(Reading database ... 
dpkg: serious warning: files list file for package `fink-buildlock-ccp4-6.5-1' missing, assuming package has no files currently installed.
1036159 files and directories currently installed.)
Removing fink-buildlock-ghostscript-10.02.0-1 ...

config.log says:

configure:8868: checking for cmsCloneTransformChangingFormats in -llcms2mt
configure:8893: flag-sort gcc -o conftest  -DHAVE_RESTRICT=1 -DUSE_LIBICONV_GNU -I/opt/sw/include -DUSE_LIBPAPER -MD -Wno-shift-negative-value -I/opt/sw/include -L/opt/sw/lib/0 -L/opt/sw/lib -Wl,-headerpad_max_install_names conftest.c -llcms2mt  -lm -ldl  -liconv -L/opt/sw/lib -lidn -lpaper >&5
ld: warning: directory not found for option '-L/opt/sw/lib/0'
Undefined symbols for architecture x86_64:
  "_cmsCloneTransformChangingFormats", referenced from:
      _main in conftest-34a14d.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

I do in fact have lcms2 installed:

$ fink apropos lcms2
Information about 15258 packages read in 5 seconds.
 i   lcms2                                              2.15-1                                 Free color management engine in 100K
 i   lcms2-bin                                          2.15-1                                 Free color management engine in 100K
 i   lcms2-shlibs                                       2.15-1                                 Free color management engine in 100K
 i   lcms2mt2                                           2.12-1                                 Color management engine (multithreaded)
 i   lcms2mt2-shlibs                                    2.12-1                                 Color management engine (multithreaded)

(I'm on Big Sur with Xcode 13)

[dmacks]

This is specifically about lcms2mt, not lcms2 itself. Seems like it is getting mis-compiled? On my 10.13, nm /sw/lib/liblcms2mt.dylib | grep _cmsCloneTransformChangingFormats says:

000000000002c580 (__TEXT,__text) external _cmsCloneTransformChangingFormats

[cooljeanius]

This is specifically about lcms2mt, not lcms2 itself. Seems like it is getting mis-compiled? On my 10.13, nm /sw/lib/liblcms2mt.dylib | grep _cmsCloneTransformChangingFormats says:

000000000002c580 (__TEXT,__text) external _cmsCloneTransformChangingFormats

I get:

$ nm /opt/sw/lib/liblcms2mt.dylib | grep _cmsCloneTransformChangingFormats
000000000002aee0 t _cmsCloneTransformChangingFormats
000000000004b2b0 t _cmsCloneTransformChangingFormats.cold.1

[dmacks]

I mis-pasted the 'nm' flags, but regardless I can already see the problem: liblcms2mt is indeed mis-compiled. The symbol is marked private rather than public. During 'fink rebuild lcms2mt2', I see:

checking for __attribute__((visibility))... yes
checking whether C compiler accepts -fvisibility=hidden... yes

I'm going to guess you get no/yes? My config.log has:

configure:16014: checking for __attribute__((visibility))
configure:16036: gcc -o conftest -g -O2 -I/sw/include -L/sw/lib -Wl,-headerpad_max_install_names conftest.c  >&5
configure:16036: $? = 0
configure:16049: result: yes
configure:16072: checking whether C compiler accepts -fvisibility=hidden
configure:16091: gcc -c -g -O2  -fvisibility=hidden -I/sw/include conftest.c >&5
configure:16091: $? = 0
configure:16100: result: yes

[cooljeanius]

I mis-pasted the 'nm' flags,

Forgot the -m flag? Let me try with that one:

$ nm -m /opt/sw/lib/liblcms2mt.dylib | grep _cmsCloneTransformChangingFormats
000000000002aee0 (__TEXT,__text) non-external (was a private external) _cmsCloneTransformChangingFormats
000000000004b2b0 (__TEXT,__text) non-external _cmsCloneTransformChangingFormats.cold.1

but regardless I can already see the problem: liblcms2mt is indeed mis-compiled. The symbol is marked private rather than public. During 'fink rebuild lcms2mt2', I see:

checking for __attribute__((visibility))... yes
checking whether C compiler accepts -fvisibility=hidden... yes

I'm going to guess you get no/yes?

Correct:

checking for __attribute__((visibility))... no
checking whether C compiler accepts -fvisibility=hidden... yes

My config.log has:

configure:16014: checking for __attribute__((visibility))
configure:16036: gcc -o conftest -g -O2 -I/sw/include -L/sw/lib -Wl,-headerpad_max_install_names conftest.c  >&5
configure:16036: $? = 0
configure:16049: result: yes
configure:16072: checking whether C compiler accepts -fvisibility=hidden
configure:16091: gcc -c -g -O2  -fvisibility=hidden -I/sw/include conftest.c >&5
configure:16091: $? = 0
configure:16100: result: yes

Mine has:

configure:16014: checking for __attribute__((visibility))
configure:16036: gcc -o conftest -g -O2 -I/opt/sw/include -L/opt/sw/lib/0 -L/opt/sw/lib -Wl,-headerpad_max_install_names conftest.c  >&5
ld: warning: directory not found for option '-L/opt/sw/lib/0'
configure:16036: $? = 0
configure:16049: result: no
configure:16072: checking whether C compiler accepts -fvisibility=hidden
configure:16091: gcc -c -g -O2  -fvisibility=hidden -I/opt/sw/include conftest.c >&5
configure:16091: $? = 0
configure:16100: result: yes
configure:16116: : CFLAGS="$CFLAGS"
configure:16119: $? = 0
configure:16144: checking whether byte ordering is bigendian
configure:16159: gcc -c -g -O2 -fvisibility=hidden -I/opt/sw/include conftest.c >&5
configure:16159: $? = 0
configure:16204: gcc -c -g -O2 -fvisibility=hidden -I/opt/sw/include conftest.c >&5
configure:16204: $? = 0
configure:16222: gcc -c -g -O2 -fvisibility=hidden -I/opt/sw/include conftest.c >&5
conftest.c:28:4: error: use of undeclared identifier 'not'
                 not big endian
                 ^
1 error generated.
configure:16222: $? = 1

btw, another line of concern from the lcms2mt2 configure output:

./configure: line 15933: LIB_PLUGINS: command not found

(also, I found it kind of confusing how lcms2mt2 is extracted from the ghostscript sources; I thought there was some sort of circular dependency happening at first, lol...)

[TheSin-]

So looks like the harmless bug that shouldn’t affect anything does affect config scripts :(

try updating fink is inject from latest master and run again I bet it fixes at least that part

[cooljeanius]

So looks like the harmless bug that shouldn’t affect anything does affect config scripts :(

try updating fink is inject from latest master and run again I bet it fixes at least that part

Do you remember which commit specifically fixed it? I'd like to cherry-pick it.

[dmacks]

So looks like the harmless bug that shouldn’t affect anything does affect config scripts :(
try updating fink is inject from latest master and run again I bet it fixes at least that part

Do you remember which commit specifically fixed it? I'd like to cherry-pick it.

These two together:

• fink/fink@35b9167
• fink/fink@02ca76c

[cooljeanius]

So looks like the harmless bug that shouldn’t affect anything does affect config scripts :(
try updating fink is inject from latest master and run again I bet it fixes at least that part

Do you remember which commit specifically fixed it? I'd like to cherry-pick it.

These two together:

• fink/fink@35b9167

• fink/fink@02ca76c

OK thanks, I can confirm that rebuilding with a fink that has these 2 commits makes the issue reported here go away.

[dmacks]

Yay! GNU autoconf-archive's AX_GCC_FUNC_ATTRIBUTE is indeed fragile in this way (and internally documented as such).

[cooljeanius]

So I'm not really sure... while I may have worked around the failure locally, I'm wondering if there might be anything else to fix here that's worth keeping the bug open for?