In one place an evaluation was pulled into the integration point loop.
That might cause a tiny bit more computation. On the other hand the
optimizer might take care of that. On the plus side we don't map data by
hand anymore in these places, and the evaluations are more local than
before.

localDOF() seems to be incompatible with the current approach for
staggered coupling. But I have the impression that there are some flaws
with the current approach to staggered coupling.

This is required by MFront models for evaluation of the bulk modulus.

in add_secondary_variable

 of ThermoRichardsMechanics_3D_ThermoElastic_Stress_Analysis

  pipx run jupytext --to markdown Tests/Data/Mechanics/Linear/DiscWithHole/Linear_Disc_with_hole.ipynb

This saves significant (up to 1/2) memory of the integration point data.

Further reduces memory allocations and slightly faster too.

Slightly better performance, saves up to half of the space
needed for each integration point.

Lars Bilke authored 2 years ago and Dmitri Naumov committed 2 years ago

Now fmt is used as a regular package. Before it was bundled with
spdlog.