compilers portgroup has errors if dependency is not yet installed

[joefowler (Joe Fowler)]

Because of the github checksum change (see issue #54839), ifeffit needs to be updated.

The attached change to the Portfile ought to suffice.

[joefowler (Joe Fowler)]

Patch for the Portfile

[ryandesign (Ryan Carsten Schmidt)]

But... you haven't changed the checksums in the patch?

Also, as noted in comment:ticket:54839:9, there's no need to increase the revision.

[ryandesign (Ryan Carsten Schmidt)]

It looks like the checksums in fact haven't changed. As such, the port does not need an update.

[joefowler (Joe Fowler)]

Well, I agree that the checksums haven't changed. This shows that I must be completely confused as to what #54839 is all about. I will comment there and see if the ifeffit package was placed on that list by mistake. Let's see what response I get before we close the present issue.

[l2dy (Zero King)]

It was placed on that list by mistake.

[raimue (Rainer Müller)]

I am unable to fetch and checksum this port, which is why I had put it on #54839:

--->  Fetching distfiles for ifeffit
Error: Internal error: compilers.enforce_fortran: 'pgplot' is not an installed port.
Error: Failed to fetch ifeffit: Internal error: compilers.enforce_fortran: 'pgplot' is not an installed port.

[jmroot (Joshua Root)]

This seems like a bug in the compilers portgroup.

[ryandesign (Ryan Carsten Schmidt)]

David, any ideas? You added the code that produces this error message in r142486.

[dstrubbe (David Strubbe)]

I guess you can get this problem if you do just "port fetch" and don't have the dependent port installed, but I believe it works correctly for "port install" (so it wasn't noticed before). I will investigate how to make the check be compatible with "port fetch" too.

[dstrubbe (David Strubbe)]

At any rate, if you want to just do "port fetch ifeffit", you can install pgplot and then it will work.

[ryandesign (Ryan Carsten Schmidt)]

Replying to dstrubbe:

I will investigate how to make the check be compatible with "port fetch" too.

Seems simple... at the end of the compilers portgroup, you call the enforce actions inside a pre-fetch block. Change that to a pre-configure block. The configure phase is the earliest you can be sure the build and library dependencies are installed. (Runtime dependencies might not be installed by then, so I hope nobody is trying to check runtime dependencies with this.)

[ryandesign (Ryan Carsten Schmidt)]

And also investigate what happens with binaries, since they won't have a configure phase. There's already a comment in the portgroup that maybe you need to also do stuff at pre-archivefetch.

[dstrubbe (David Strubbe)]

In c6cf5aafe46f614dc733f14c51a12564a1e985aa/macports-ports (master):

compilers portgroup: Move some checks to pre-configure and pre-archivefetch. Fixes #54890.

[dstrubbe (David Strubbe)]

In 74f6c824bf28eb2dab3c060c4de82c87f69c2f1b/macports-ports (master):

mpi portgroup: Improve enforce_variants for depspec; move checks to configure and archivefetch.

Use of mpi.enforce_variant could be done with a depspec rather than a port name (e.g. path:/opt/local/lib/libplumedKernel.dylib:plumed) but the resulting error messages were not clear; now it will give the name of an actual port in messages. The pre-fetch checks on dependencies should be done in pre-configure since only then are the dependencies guaranteed to be installed; a similar check should be done in pre-archivefetch for binaries. (Similar to #54890 for compilers portgroup.)