summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMarek Szuba <marecki@gentoo.org>2021-07-14 17:03:10 +0100
committerMarek Szuba <marecki@gentoo.org>2021-07-14 17:14:45 +0100
commit472f36bcd300da19c6f6953ece0811b63a3134e0 (patch)
tree5ca6af07a22126c25f33aa8f4cb5c1d9741d9ef8 /profiles/base
parentsys-cluster/ampi: keyword 0_pre20140616 for ~riscv, EAPI 5 -> 8 (diff)
downloadgentoo-472f36bcd300da19c6f6953ece0811b63a3134e0.tar.gz
gentoo-472f36bcd300da19c6f6953ece0811b63a3134e0.tar.bz2
gentoo-472f36bcd300da19c6f6953ece0811b63a3134e0.zip
sci-libs/pastix: add 6.0.3, masked
The ebuild in principle works but there are several features which for various reasons remain problematic (see the package.use.mask entry), and with this being a new major version chances are not all revdeps will be happy with it (hence the overall mask). Testing very much encouraged! Bug: https://bugs.gentoo.org/751034 Signed-off-by: Marek Szuba <marecki@gentoo.org>
Diffstat (limited to 'profiles/base')
-rw-r--r--profiles/base/package.use.mask20
1 files changed, 20 insertions, 0 deletions
diff --git a/profiles/base/package.use.mask b/profiles/base/package.use.mask
index 83513d5d82c1..3bcead9090dd 100644
--- a/profiles/base/package.use.mask
+++ b/profiles/base/package.use.mask
@@ -6,6 +6,26 @@
# This file is only for generic masks. For arch-specific masks (i.e.
# mask everywhere, unmask on arch/*) use arch/base.
+# Marek Szuba <marecki@gentoo.org> (2021-07-14)
+# Several known issues (possibly with older pastix versions as well but
+# in v6 they result in clear errors):
+# * CUDA support needs testing, may or may not work as it is;
+# * All sci-libs/scotch ebuilds currently in the tree seem to always use
+# 32-bit integer representation, even when built USE=int64. Seeing as
+# metis ordering has got more complicated issues (see below), just
+# stick with 32 bits for now;
+# * sci-libs/metis currently always uses whatever integer representation
+# is native to the target architecture even though upstream allows for
+# using 32-bit integers on 64-bit arches, this will have to be implemented
+# so that pastix can pass int64? to it;
+# * With MPI enabled two tests fail (when linked against openmpi, at least)
+# due to incorrect implementation of either MPI support itself or
+# said tests - example_drv_facto0_d_reentrant tries to initialise MPI
+# more than once, test_bvec_gemv_tests calls MPI_Query_thread() after
+# invocation of MPI_FINALIZE;
+# * dev-libs/starpu needs a version bump.
+>=sci-libs/pastix-6.0.0 cuda int64 metis mpi starpu
+
# Alfredo Tupone <tupone@gentoo.org> (2021-06-27)
# lto does not work on dev-lang/gnat-gpl-2021
dev-lang/gnat-gpl:10 lto