The eclass calls java-config-2 in many places to get build information. java-config-2 respects the ROOT environment variable, which is set by portage, so the calls that are looking for things like the active VM fail. The calls should probably just explicitly set ROOT to / or EPREFIX or something. * Package: dev-java/icedtea-3.3.0 * Repository: gentoo * Maintainer: gnu_andrew@member.fsf.org chewi@gentoo.org,java@gentoo.org,proxy-maint@gentoo.org * USE: abi_x86_64 alsa amd64 cups elibc_glibc gtk jbootstrap kernel_linux pch source sunec userland_GNU webstart * FEATURES: compressdebug preserve-libs sandbox splitdebug userpriv usersandbox * Checking for at least 8500 MiB disk space at "/var/tmp/portage/dev-java/icedtea-3.3.0/temp" ... [ ok ] * java-pkg_build-vm-from-handle: No vm found for handles: * icedtea-8 icedtea-bin-8 * icedtea-7 icedtea-bin-7 * java-pkg_switch-vm: No VM found for handles: * icedtea-8 icedtea-bin-8 * icedtea-7 icedtea-bin-7 * ERROR: dev-java/icedtea-3.3.0::gentoo failed (setup phase): * java-pkg_switch-vm: Failed to determine VM for building
Also, while it wouldn't fix this bug, shouldn't these checks be done in a src_* phase? A build VM shouldn't be required for a binary package install.
(In reply to William Throwe from comment #1) > Also, while it wouldn't fix this bug, shouldn't these checks be done in a > src_* phase? A build VM shouldn't be required for a binary package install. When "${EAPI}" not in ("0", "1", "2", "3"), the following check can be used in eclass: if [[ "${MERGE_TYPE}" != "binary" ]]; then (And doing these checks also in pkg_pretend() would be better.)
