mpi portgroup depends on nonexistent 'mpich-clang34' on Tiger

[jmroot (Joshua Root)]

While it's fine not to support Tiger, this causes mirroring failures. A couple of examples:

% port lint petsc os.major=8
--->  Verifying Portfile for petsc
Error: Unknown dependency: mpich-clang34
--->  1 errors and 0 warnings found.
% port lint scalapack os.major=8
--->  Verifying Portfile for scalapack
Error: Unknown dependency: mpich-clang34
--->  1 errors and 0 warnings found.

​https://build.macports.org/builders/jobs-mirror/builds/281451/steps/mirror/logs/stdio

[kencu (Ken)]

just a plea that before someone does a massive amount of blacklisting of various systems, mpich-default (at least) works just fine for me on Tiger PPC and Intel on up:

$ port -v installed mpich-default
Warning: port definitions are more than two weeks old, consider updating them by running 'port selfupdate'.
The following ports are currently installed:
  mpich-default @3.4.2_0+gcc7 (active) requested_variants='' platform='darwin 8' archs='ppc' date='2021-06-13T09:54:09-0700'

and passes the complete test suite.

So rather than a whole lot more spaghetti I will have to undo, let's get this fixed?

[mascguy (Christopher Nielsen)]

Hmmmm. On line 308 of the portgroup, we're unconditionally disabling all older clang versions, including clang34.

I'll have to investigate...

[mascguy (Christopher Nielsen)]

Of note, mpich-clang34 hasn't been supported since August of 2017:

​https://github.com/macports/macports-ports/commit/3aa83fd52375f4af609eed7d9f6dedbcb25738b7

This doesn't directly relate to the issue, since the portgroup determines what's valid. Nonetheless, it's an interesting bit of historical trivia.

Still digging...

[mascguy (Christopher Nielsen)]

Josh, has this been happening for a long time (a year or more), or did it start within the past few months...?

[jmroot (Joshua Root)]

No idea, I just noticed the mirror failures today.

[mascguy (Christopher Nielsen)]

No worries.

I'm adding supplemental logging in various places, to see if I can track down why this is occurring. (Won't be committing any changes though, this is just for local diagnostic purposes.)

[mascguy (Christopher Nielsen)]

Finally had a chance to fire up my old PowerBook G4, and boot into 10.4.

After updating MacPorts to 2.7.1, syncing, etc, this is what I'm seeing:

$ sw_vers
ProductName:    Mac OS X
ProductVersion: 10.4.11
BuildVersion:   8S165

$ port version
Version: 2.7.1

$ port info --fullname --depends petsc scalapack
fullname: petsc @3.15.0+accelerate+hwloc+mpich
depends_fetch: 
depends_extract: 
depends_patch: 
depends_build: sowing, python39, c2html, apple-gcc42
depends_lib: vecLibFort, mpich-default, hwloc
depends_run: 
depends_test: 
--
fullname: scalapack @2.1.0+accelerate+gfortran+mpich
depends_fetch: 
depends_extract: 
depends_patch: 
depends_build: cmake, gcc7, apple-gcc42
depends_lib: vecLibFort, libgcc, mpich-default
depends_run: 
depends_test: 

$ port lint --nitpick petsc scalapack 
--->  Verifying Portfile for petsc
--->  0 errors and 0 warnings found.
--->  Verifying Portfile for scalapack
--->  0 errors and 0 warnings found.

So I'm not sure where mpich-clang34 is coming from. Though overriding os.major via the command line isn't 100% reliable, so I'm not surprised that the results are different.

Josh, is there anything else I can test from 10.4, to help diagnose this...?

[mascguy (Christopher Nielsen)]

Another question is whether this is only occurring for an Intel-based setup.

My old 2007-era MacBookPro should be able to run 10.4 as well, though I'd need to install the OS.

[mascguy (Christopher Nielsen)]

Josh, do you have any additional thoughts on this? (Not urgent, simply want to help you resolve this.)

[jmroot (Joshua Root)]

No other clues yet except that I noticed there's also a dependency on the nonexistent mpich-clang70 on Leopard:

% port lint petsc os.major=9 
--->  Verifying Portfile for petsc
Error: Unknown dependency: mpich-clang70
--->  1 errors and 0 warnings found.

[jmroot (Joshua Root)]

Looks like mpi.get_default_mpi_compiler might be what's making up nonexistent subports.

[mascguy (Christopher Nielsen)]

Thanks Josh, I'll try to take a closer look at some point soon.

[jmroot (Joshua Root)]

In d0ce9d72a935ad51004fa3183230e11409b08656/macports-ports (master):

mpi pg: don't depend on nonexistent ports

Closes: #63091