devel/java-subversion 1.15.0 FreeBSD

devel/subversion: - Bump to 1.15.0 (DISTVERSION). - Replace the opt-in WITH_REPOSITORY_CREATION pw useradd/groupadd logic with framework USERS=/GROUPS=, mirroring devel/git's git_daemon pattern; register the svn account in UIDs/GIDs with HOME=/var/db/svn to match svnserve's own rc.d default. Repository creation itself is dropped as out of scope for a port to manage. Documented in UPDATING, since it affects existing installs with a manually-created svn account. - TOOLS now only gates genuinely niche binaries; drop it from OPTIONS_DEFAULT, add SVNSERVE_WRAPPER instead. svnauthz/ svnauthz-validate/svnbench move into core via a patch-build.conf hunk, matching Debian. svnmover, fsfs-access-map, and x509-parser …

Upstream no longer maintains a separate LTS release line, so this port has no path forward other than tracking the same 1.14.x branch indefinitely while devel/subversion moves ahead. Mark it DEPRECATED and schedule removal via EXPIRATION_DATE. The same deprecation also needs to apply to the WITH_SUBVERSION_VER=LTS build of every slave port (devel/java-/py-/ruby-/p5-subversion, security/subversion-gnome-keyring, www/mod_subversion): they never touch devel/subversion-lts/Makefile directly, they go through devel/subversion/Makefile.addons's own .elif LTS branch, which had no warning of its own. Add a matching DEPRECATED/EXPIRATION_DATE pair there too. Bump PORTREVISION for subversion-lts and PORTREVISION_LTS for every slave port. …

- java-subversion: document the real cause of MAKE_JOBS_UNSAFE (build.conf's Java-target add-deps only thread into the phony aggregate target, not the javac -h compile rule, so -j can run javahl-types-java before javahl-java's NativeResources.class exists). - subversion{,-lts}: drop GPG_AGENT_BUILD_DEPENDS; --with-gpg-agent is a pure configure-time #define with no build-time dependency on the gpg-agent binary, only a runtime one (already covered by GPG_AGENT_RUN_DEPENDS). - subversion{,-lts}: remove the dead .if ${PREFIX} != "/usr" guard around two REINPLACE_CMD calls in post-patch; PREFIX defaults to LOCALBASE and is never literally /usr in a standard ports build, so the guard was always true. - mod_subversion: remove the dead PLIST_SUB+= SED=${SED} token; …

Remove every compiler-conditional/workaround flag from this port family, relying on stock Clang/libc++ behavior instead: - devel/subversion, devel/subversion-lts: drop the amd64-only CFLAGS+=-fpic -DPIC block and USES=compiler. Add files/patch-configure: configure's "additional flags to link C++ libraries" check only special-cases *freebsd10* as needing no extra -lstdc++; every later FreeBSD release falls through to the default case and gets -lstdc++ added, which fails to link since FreeBSD has shipped Clang/libc++ (no libstdc++.so) since 10.0. Broaden the match to *freebsd*. - devel/java-subversion: drop the post-configure block that blanked out LT_CXX_LIBADD by hand when COMPILER_TYPE == clang, superseded by the configure-level fix above; drop the now-unused …

Install svn-javahl.jar under JAVAJARDIR (the ports tree's own convention for JARs) instead of a non-standard lib/svn-javahl/ subdirectory. Fix a --without-siwg-python typo in CONFIGURE_ARGS. Switch USES=java to java:build: this port installs a native library and a JAR, it doesn't invoke java itself, so it has no business pulling a JDK into RUN_DEPENDS -- the caller that actually loads the library and the JAR decides which JVM to run, not this port. Add files/patch-native_JNIUtil.cpp: JNIGlobalInit()'s non-Windows NLS setup hand-rolled a bindtextdomain() call that duplicates svn_nls_init() (the same helper every svn_cmdline_init()-based CLI tool uses); call the real function instead, which also picks up bind_textdomain_codeset() that the duplicated code was missing. …