Add dedicated covariance matrix components and use them

[tmadlener]

BEGINRELEASENOTES

• Introduce edm4hep::CovMatrix[2,3,4,6}f components to represent covariance matrices
  • using std::array<float, N> under the hood
  • Access functionality that works on enums for defining dimensions and some utilities to do the indexing into the underlying storage
• Introduce some enum classes for defining dimensions, e.g. TrackParams or Cartesian
• Use these components instead of the raw std::arrays in datatypes and other components and tie the interpretation of the values to the approriate dimension enum class

Usage example:

#include "edm4hep/TrackerHit3D.h"

void foo(const edm4hep::TrackerHit3D& hit) {
  const auto covXY  = hit.getCovMatrix(edm4hep::Cartesian::x, edm4hep::Cartesian:y);
}

ENDRELEASENOTES

I am not yet sure whether I really like the duplication of all the extra code here. Especially for the components it is effectively always the same and this could in principle be easily templated. However, then we would have to mix jinja and c++ templates, so I am not entirely sure if we want that.

One possibility would be to put that part of the extra code into some ipp file and just #include it in the ExtraCode declaration.

• Check breakages downstream
  • Make upcoming introduction of cov matrix components transparent HEP-FCC/FCCAnalyses#362 to build the stack with and without these changes
  • Transparently introduce CovMatrix components k4EDM4hep2LcioConv#56 in order to not break the tests there

[tmadlener]

The release based CI workflows fail because AIDASoft/podio#553 is not yet in there.

[tmadlener]

dev3 based workflows are broken due to: root-project/root#14964

[tmadlener]

I have also removed the storage details from the docstrings of the covariance matrix members, as that should be (almost) completely be abstracted away by the components.

[tmadlener]

Because this came up at the EDM4hep meeting: This still has the same issues for RDataFrame::Describe as the version with bare std::array. Trying to run it on the example file that gets created during the tests, I get:

TClass::Init:0: RuntimeWarning: no dictionary for class edm4hep::CovMatrix3f is available
TStreamerInfo::BuildOld:0: RuntimeWarning: Cannot convert edm4hep::TrackerHitPlaneData::covMatrix from type: edm4hep::CovMatrix3f to type: array<float,6>, skip element
---------------------------------------------------------------------------
runtime_error                             Traceback (most recent call last)
Cell In[3], line 1
----> 1 df.Describe()

runtime_error: ROOT::RDF::RDFDescription ROOT::RDF::RInterfaceBase::Describe() =>
    runtime_error: TTree leaf TrackerHitPlanes.covMatrix.values[6] has both a leaf count and a static length. This is not supported.

[andresailer]

setCovMatrix(std::array const&) missing

[m-fila]

Shouldn't the setCovMatrix be put in MutableExtraCode instead of ExtraCode?

[tmadlener]

Definitely. Let me check again to have that consistently.

[tmadlener]

After having had another look; The TrackState is actually a component. Hence, it only has ExtraCode and it doesn't need a setCovMatrix because you can just assign to covMatrix directly (without having to go through the setter).

[tmadlener]

I have removed the setCovMatrix(const std::array<float, N>&) from the datatypes and instead added a constructor of the form

template <typename ...Vs>
CovMatrixNf(Vs... vs) : values{vs...} {}

to the covariance matrix components. This makes direct initialization from initializer lists via the datatypes possible, e.g. as in the tests

  auto trackerHit = MutableTrackerHit3D{};
  trackerHit.setCovMatrix({1.f, 2.f, 3.f, 4.f, 5.f, 6.f});
  REQUIRE(trackerHit.getCovMatrix() == std::array{1.f, 2.f, 3.f, 4.f, 5.f, 6.f});

  const std::array arrValues = {6.f, 5.f, 4.f, 3.f, 2.f, 1.f};
  trackerHit.setCovMatrix(arrValues);
  REQUIRE(trackerHit.getCovMatrix() == std::array{6.f, 5.f, 4.f, 3.f, 2.f, 1.f});

This accepts everything that is convertible to a float and emits a narrowing warning if a narrowing conversion happens (if enabled).