broken ports due to cxx_stdlib mismatch

[EJFielding (Eric Fielding)]

I updated my MacPorts to 2.5.0 using the usual port selfupdate and then ran port upgrade outdated. The upgrade seemed to go well, but at the end it checked for broken ports and found three but did not try to do anythng to fix them. I tried running "port rev-upgrade" but that also did not fix the broken ports. These are the broken ports:

port rev-upgrade
--->  Scanning binaries for linking errors
--->  No broken files found.
--->  Found 3 broken ports:
     TeXShop4 @4.01
     py27-scipy @1.1.0 +gcc49
     hdf5 @1.10.2 +cxx+gcc7+hl

I tried uninstalling and installing again the hdf5 port, but it still stayed broken even after rebuilding. I can't tell if this is a problem with the new port version or with the individual ports.

[jmroot (Joshua Root)]

If you run rev-upgrade with the -d option it will show you why the ports are considered broken. It's likely to be the new check for whether ports are linked with the C++ standard library that they say they are.

[EJFielding (Eric Fielding)]

Thanks for the hint. You are right. Those three ports seem to be linked to the libstdc++ library. There were a lot of other debug output lines that did not seem relevant.

DEBUG: Ignoring loadcommand containing @rpath in /opt/local/lib/rustlib/x86_64-apple-darwin/lib/libterm-08da937915c30bc0.dylib
DEBUG: Ignoring loadcommand containing @rpath in /opt/local/lib/rustlib/x86_64-apple-darwin/lib/libtest-6a429b3fcf5acfff.dylib
DEBUG: Ignoring loadcommand containing @rpath in /opt/local/lib/rustlib/x86_64-apple-darwin/lib/libtest-6a429b3fcf5acfff.dylib
DEBUG: skipping ppc in /opt/local/share/cmake-3.11/Modules/CPack.OSXScriptLauncher.in since this system can't run it anyway
--->  No broken files found.
TeXShop4 is using libstdc++ (this installation is configured to use libc++)
py27-scipy is using libstdc++ (this installation is configured to use libc++)
hdf5 is using libstdc++ (this installation is configured to use libc++)
--->  Found 3 broken ports:
     TeXShop4 @4.01 
     py27-scipy @1.1.0 +gcc49
     hdf5 @1.10.2 +cxx+gcc7+hl
DEBUG: Checking time since last reclaim run

So is this something that I need to fix or do I ignore it?

[jmroot (Joshua Root)]

It's something that the port maintainers need to fix. However the ports are no more broken than they were before.

[Schamschula (Marius Schamschula)]

TeXShop4 fixed in ​https://github.com/macports/macports-ports/commit/a7ad34625d6fd7c54a99f338618ecca87c71a91f

SuiteSparse @4.2.1_4 is also broken.

[michaelld (Michael Dickens)]

SuiteSparse needs some work to see if a modern version works with Octave. I'll try to work on that along with this issue "soonish" (for SS & SciPy).

[jmroot (Joshua Root)]

FTR a rev-bump alone won't fix this. If you change configure.cxx_stdlib to match what the port is already doing, you don't need to rev bump. If you fix the port to use the specified stdlib, you do need to rev bump because the installed files change. The latter is preferable, but the former is acceptable as long as the port doesn't use or provide any C++ libraries from/to other things.

[mf2k (Frank Schima)]

@EJFielding: A little off-topic, but why are you using an ancient version of gcc (gcc49) with py27-scipy? You should use the same gcc variant for all of your scientific ports. Use gcc7 like you did for hdf5.

[jmroot (Joshua Root)]

How many of these are still broken after rev-upgrading with 2.5.1?

[EJFielding (Eric Fielding)]

Two are still broken

 port rev-upgrade
--->  Scanning binaries for linking errors
--->  No broken files found.
--->  Found 2 broken ports:
     hdf5 @1.10.2 +cxx+gcc7+hl
     py27-scipy @1.1.0 +gcc7

Replying to jmroot:

How many of these are still broken after rev-upgrading with 2.5.1?

[jmroot (Joshua Root)]

Even after letting rev-upgrade reinstall them?

[EJFielding (Eric Fielding)]

On my 10.12.6 Mac, port rev-upgrade does not attempt to reinstall the broken ports under 2.5.0 or 2.5.1.

[jmroot (Joshua Root)]

You probably have revupgrade_mode set to report in your macports.conf then. You'll either have to change it to rebuild or reinstall the affected ports manually, since the new base version can't fix the metadata for ports that have already been installed.

[EJFielding (Eric Fielding)]

You were right, "revupgrademode" was set to "report" but I have not changed my macports.conf file that I can remember and it has a modification date of January 2017. It was rebuilding before. Maybe the previous versions were not paying attention to that line.

[majoc-at-astro (majoc-at-astro)]

For the record, still there for us on 10.13.4, built from the 2.5.1 MacPorts tarball:

### (229 of 264: min-test-set.sh) port install hdf5 +cxx +gcc7 +fortran +openmpi (Mon Jun  4 10:54:01 BST 2018)
[snipissimo]
--->  Building hdf5
--->  Staging hdf5 into destroot
--->  Deactivating hdf5 @1.10.2_0+cxx+fortran+gcc7+hl+openmpi
--->  Cleaning hdf5
--->  Uninstalling hdf5 @1.10.2_0+cxx+fortran+gcc7+hl+openmpi
--->  Cleaning hdf5
--->  Computing dependencies for hdf5
--->  Installing hdf5 @1.10.2_0+cxx+fortran+gcc7+hl+openmpi
--->  Activating hdf5 @1.10.2_0+cxx+fortran+gcc7+hl+openmpi
--->  Cleaning hdf5
--->  Scanning binaries for linking errors
--->  No broken files found.
Error: Port hdf5 is still broken after rebuilding it more than 3 times.

[jmroot (Joshua Root)]

OK, so hdf5 at least still needs to be fixed. With those variants it's building with mpicxx-openmpi-gcc7. The mpi portgroup is probably where configure.cxx_stdlib needs to be set appropriately.

[EJFielding (Eric Fielding)]

My py27-scipy and hdf5 ports were rebuilt successfully after I changed my macports.conf. Thanks!

I have "hdf5 @1.10.2_0+cxx+gcc7+hl" so I am not using openmpi in my hdf5 installation.

[jmroot (Joshua Root)]

I'll close this ticket then. I've opened a more specific one for mpi: #56608