PyTorch 1.9.1 (fixed in versions >= 1.10.0) depends on an older onnx
version that utilizes a deprecated function from Google's protobuf
library that was removed in version 3.18.0 (see
https://github.com/protocolbuffers/protobuf/pull/8794/).

Workaround by specifying an explicit version range for this
dependency.

Egidio Falotico mentioned NRP has problems integrating py-torch "into
the platform". Let's see what happens…

Fix(pynn-brainscales): Enforce correct log4cxx version

See merge request technical-coordination/devops/platform/ebrains-spack-builds!26

(cherry picked from commit 1e2fcca8)

fa396948 Fix(pynn-brainscales): Enforce correct log4cxx version

Eric Müller authored 3 years ago and Athanasios Karmas committed 3 years ago

The production env seems to use a newer spack repo state → log4cxx is
newer than currently supported for BrainScaleS → quickfix: pin to old
version.


(cherry picked from commit fa396948)

This is a hack to enforce a reinstallation of meta-brainscales.
(The underlying problem is that pynn-brainscales does not specify all
the dependencies but only via meta-brainscales; and there's already one
installed and spack does not easily install another one ;).)

Add specs in spack.yaml. Update names for pyaescrypt and formencode and specify them as dependencies for tvb-framework.