Backtraces have sadly been broken from some time, and it was a major
pain to find the source (still searching). This commit removes all the
self rolled code with the backward-cpp library
(https://github.com/bombela/backward-cpp)

This library features a much more comprehensive set of options, pretty
printing and deals with dependencies for us.

* Pass a default config to DataContainer.format, such that it can be
  passed in from the outside
* Move `format_config` into separate file to not create dependency cyles
* Add functionality to suppress or round small numbers to 0, such that
  instead of -1.49012e-08, only 0 is printed

In case of an error, print a version of the clang-format command
that will actually fix the error and not only print to standard out

Andi Braimllari authored Nov 11, 2021 and David Frank committed Nov 11, 2021

For further details check for example the numpy docs in:
https://numpy.org/doc/stable/reference/generated/numpy.fft.fftshift.html

The Makefile is just a convenience to more easily build elsa.

To build all of elsa, just call `make build` from the top level
directory. You can also pass any target to build, to specifically build
this target, e.g. `make build test_DataContainer`.

Other useful commands could `make tests`, to build and run all tests.
`make test <test-name>` to build a specific test case and run it.
And `make watch <test-name>` to build and run a test continuously on
code changes in elsa.

Note that a sub-folder structure will be created of the form
`build/$BUILD_TYPE/$COMPILER`. This enables quick iteration for
different build types and compilers, via passing the options
`BUILD_TYPE` and `CXX`.

To build all docker images, for some reason it's necessary to reduce the
spdlog version from 1.9.0 to 1.8.5. For not known reasons, the build for
the GPU only image fails with spdlog version 1.9.0, but it somehow
succeeds for the current pipeline image. I don't know why that is and
how to fix it, so for now reduce it, that images can still be updated
with new CMake versions and so on.

With this commit, all test binaries use regular console output, and
ctests generates the junit xml report used for nice visual presentation
in GitLab (which was broken since I don't know when, so it should work
again now). Previously, no output was generated, as it was piped to the
test binaries unit report, which made finding errors unnecessary
complex.

Failed tests are rerun again such that the output of these tests is
found at the bottom of the CI log, which should make finding errors a
little easier in the future.

This commit tells CPM to fetch a specific commit of quickvec instead of
the master branch. Then new changes have to be added via a MR in elsa
and therefore the whole pipeline will be run. This should prevent errors
such as #146.

Also tests for CG are adapted, for whatever reason they failed, now they
don't anymore...

andibraimllari authored Sep 12, 2021 and David Frank committed Oct 17, 2021

minElement and maxElement return the min and max element of a
DataContainer, respectivly. For complex vectors, an expection is thrown.

Andi Braimllari authored Oct 11, 2021 and David Frank committed Oct 11, 2021

For a complex DataContainer real() and imag() return the real
and imaginary part of it respectively. The return type is a
real valued DataContainer.

Note, similar to the standard library (std::real and std::imag),
non-complex DataContainers are treated as complex with zero
imaginary component.

Andi Braimllari authored Oct 09, 2021 and David Frank committed Oct 09, 2021

The new trajectory generator creates a circular trajectory,
where a certain portion of the trajectory is missing.

This commit also refactors out common code from both generators
and fixes a minor bug, where the trajectory would overshoot slightly
(add a pose at 360.2 or similar).

Andi Braimllari authored Oct 06, 2021 and David Frank committed Oct 06, 2021

To shortcut scaling of operators, add a multiplication of left
hand sided scalar with a right hand sided LinearOperator.

Now, it's not necessary to create uniform scaling operations
to scale operators.

David Frank authored Jul 29, 2021 and Tobias Lasser committed Jul 29, 2021

This patch adds 3 changes:
- Rely on spdlog::stopwatch instead of the 'raw' standard library clock
- Change the output format a touch to improve readability
- Rename of variables to express intent better

As of version 1.8.0 of spdlog a stopwatch is provided. I thought it
would be nice to justs rely on that one instead of rolling our own (even
if it's not that hard).

The other change is the output format. Instead of printing in the format
'[Timing] [debug] Execution of <caller>: <fn> took <time>ms', I changed
it to '[<caller>] [debug] <fn> took <time>s'. So instead of having a
separate Timing logger, the logger for the class is used. IMO, this
cleans up the log a touch, as my eyes don't have to search in different
columns, but it's minor. Also I use the default output format of the
spdlog::stopwatch, which is in decimal and seconds.

Also I used the change to rename the variables, as it was just a little
off.

Access to a portion of the DataContainer in the form of slices.
Currently, only slicing in the direction of the last dimension are
possible, but this should still be useful for many cases.

The one suprising thing in this implementation, is that the returned
DataContainer, has a descriptor which is of the same dimension as the
original DataContainer but with a "thickness" of 1 in the last
dimension, instead of actually being lower dimensional.

so warnings are swallowed

also install our bundled eigen headers

This is actually mostly exiting because it now supports fmtlib version
8, which is great! Sadly to fully utilize fmt 8.0.0, we need C++20
support. Check the [release
notes](https://github.com/fmtlib/fmt/releases/tag/8.0.0) of fmtlib 8.0.0
to see what awesome things we can do, if we switch to C++20 :^)

When two DataContainers are concatenated, a new one with a
RandomBlocksDescriptor is created, with the two original descriptors
as blocks.

Currently, only concatenation of containers of equal dimension is
allowed.