muniversal-1.0: symlinked worksrcpath break correct operation (only last arch built and installed)

[Ionic (Mihai Moldovan)]

If worksrcpath is really a symlink to another directory, muniversal 1.0 copies the symlink instead of the referencing directory tree. Further operations are all faulty.

One example of this happening is boehmgc:

lrwxr-xr-x   1 macports   19 Mar  5 04:02 bdwgc-8.0.2 -> ivmai-bdwgc-88a0d3d
lrwxr-xr-x   1 macports   19 Mar  5 04:03 bdwgc-8.0.2-i386 -> ivmai-bdwgc-88a0d3d
lrwxr-xr-x   1 macports   19 Mar  5 04:02 bdwgc-8.0.2-x86_64 -> ivmai-bdwgc-88a0d3d
drwxr-xr-x   3 root      102 Mar  5 04:04 destroot
drwxr-xr-x   7 root      238 Mar  5 04:04 destroot-i386
drwxr-xr-x   7 root      238 Mar  5 04:04 destroot-intel
drwxr-xr-x   2 root       68 Mar  5 04:04 destroot-powerpc
drwxr-xr-x   7 root      238 Mar  5 04:04 destroot-x86_64
drwxr-xr-x 147 macports 4.9K Mar  5 04:27 ivmai-bdwgc-88a0d3d

I have no idea how many other ports are affected by this issue.

CCing Marcus and Ryan because this PG seems to be pretty important to them.

Working on a patch right now.

[Ionic (Mihai Moldovan)]

Joshua pointed out that this behavior is relatively new as of 7921b2e05e9a4c9cda6efedee496affb305dcc07/macports-base.

git grep -E 'copy \$\{?worksrc' revealed a few other ports that might behave erratic after this change: LanguageTool, OmegaT, sphinx4, asdf-binary-locations, autobook, poco, yarn, zinc, infobot, android-platform-tools, apache-solr3, apache-solr4, derby-server, dualist, glassfishv3, launch4j, mahout, microemu, QiII, dart-sdk, fsharp, kotlin, mozart-doc, pypy-bootstrap, scala2.10, scala2.11, scala2.12, scala2.7, scala2.8, scala2.9, amavisd-new, spambnc, ProofGeneral, atlas, live555, mythweb.27, mythweb.28, xpn, pear-TCPDF, php5-simpletest, py-gpilab-core, dcw-gmt, gshhg-gmt, rangs-gshhs-ncarg, sqlmap, extractopinion, naturaldocs, unidic-mecab, ZendFramework2, drush, phppgadmin, tiki, websvn, wordpress, BWidget, gtk-osx-tiger

I may have missed a few.

All these ports copy ${worksrcpath} directly to some other location and will break in unexpected ways if it is a symlink. Ports that copy subcontents of ${worksrcpath} are not affected, though, which is the majority of ports.

[Ionic (Mihai Moldovan)]

Adding original author of the base patch.

[Ionic (Mihai Moldovan)]

Updated to include a basic loop detection mechanism. Actually just added a depth limit of 50 iterations.

[slewsys (Andrew L. Moore)]

Is it possible to just replace

copy ${worksrcpath} ${worksrcpath}-${arch}

with

exec cp -HR ${worksrcpath} ${worksrcpath}-${arch}

[Ionic (Mihai Moldovan)]

Mh, theoretically yes, but it always feels weird to execute system commands (and depend upon these) if native functionality exists.

Using that would be an easy replacement for the (maybe/theoretically) broken ports I mentioned, though.

[Mihai Moldovan <ionic@…>]

In b9564a04c393d6187025d461fa3a4251781f39ad/macports-ports (master):

muniversal-1.0: if worksrcpath is a symlink, copy the directory tree referred to instead of the symlink.

Fixes: #58163