Known Issues

Binary installation issues

Cross Platform

• This release may not work when installed without the Conda environments provided by the documented newinstall.sh, lsstsw, or Docker installation methods. The documented Python dependencies may differ slightly from those in the release, causing issues with the validation demo.

Red Hat (and clones) specific

• The DM codebase is not compatible with the Intel Math Kernel Library (MKL). Attempting to use MKL may generate incorrect numerical results.

  If you have installed using Conda, you should install the nomkl package, and remove mkl and mkl service. [DM-5105]

Source installation issues

Cross Platform

• Compiling some packages—in particular afw—require large amounts of RAM to compile. This is compounded as the system will automatically attempt to parallelize the build, and can cause the build to run extremely slowly or fail altogether. On machines with less than 8 GB of RAM, disable parallelization by setting EUPSPKG_NJOBS=1 in your environment before running eups distrib.
• The lsst_dm_stack_demo package may produce slightly different numeric results if executed on operating systems other than the officially supported versions of macOS and CentOS, or if using a different Python environment.

Red Hat (and clones) specific

• The DM codebase is not compatible with the Intel Math Kernel Library (MKL). Attempting to use MKL will cause the test suite to fail, automatically aborting the installation. [DM-5105]
• Ensure that the gcc compiler version supports C++14 as specified in the Pre-requisites.

RHEL 7.*

• No specific issues.

RHEL 6.*

• curl looks for certificates in /etc/pki/tls/certs/ca-bundle.crt rather than /etc/ssl/certs/ca-certificates.crt. The solution is to copy ca-certificates.crt to ca-bundle.crt.

macOS specific

• Macs must use the clang compiler, not gcc. [DM-3405]

macOS 10.13 (Sierra) and OS X 10.11 (El Capitan)

• MPICH version 3.2, as currently distributed with the stack, fails regularly and unpredictably with a segmentation fault on macOS systems. MPICH is used by the ctrl_pool task distribution framework, and hence the pipe_drivers top-level scripts package which provides the following executables:

  • coaddDriver.py
  • constructBias.py
  • constructDark.py
  • constructFlat.py
  • constructFringe.py
  • multiBandDriver.py
  • singleFrameDriver.py

  It should be possible to run these commands by restricting them to a single CPU core (i.e., --batch-type=smp --cores=1).

  This issue will be resolved by upgrading to version 3.3 of MPICH when it becomes available. [DM-7588]

Older systems

• Some old installations of XCode on Macs create a /Developer directory.  This can interfere with installation.