Transpose array-like in herm_conj

added 1 commit

55b02cfd - transpose array-like in herm_conj

added 2 commits

c2561aea - 1 commit from branch master
6537a9f2 - transpose array-like in herm_conj

Thanks, this should do the job. However, I believe that the following points should be addressed before merging:

It would be nice if there was a test that fails with the old code and showcases the problem. That could be a separate commit followed up by the fix.

Also, it would be interesting to check how much slower hamiltonian_submatrix becomes for the common case where values are functions. With the new code, an exception will be thrown and caught for each time herm_conj is called (that's for 50% of the values on average) which may result in a significant slowdown. If this is indeed the case, we should seek an alternative approach.

In addition, since this is a user-visible change, I suggest adding it to the changelog file below "Changes in Kwant 1.4.2". The subsection could be called "Changes after the release of Kwant 1.4.2" and we would then change it when we release.

Moreover, since the issue is already present in stable, I believe we should fix it there. (Independently of whether we will release a new stable version of the 1.4.x series.)

Also, it would be interesting to check how much slower hamiltonian_submatrix becomes for the common case where values are functions.

It will not affect it, because herm_conj is used only within HermConjOfFunc.__call__() (herm_conj() is not called on a function based on a conditional check if callable(value)), and there exception also will appear only if returned value is not an array, not a number and can't be casted to a tinyarray, the same as in the case of non-callable value.

It would be nice if there was a test that fails with the old code and showcases the problem.

Modified test_builder.py::test_hermitian_conjugation to capture this, decided to leave it as a single atomic commit.

I suggest adding it to the changelog file below "Changes in Kwant 1.4.2".

Done.

Moreover, since the issue is already present in stable, I believe we should fix it there.

I suggest just cherry-picking it there after getting merged into master, together with f8331485 to fix CI failures.

added 2 commits

a8947abf - 1 commit from branch master
99387575 - transpose array-like in herm_conj

Compare with previous version

added 1 commit

3096740b - transpose array-like in herm_conj

Compare with previous version

added 2 commits

d980cc00 - test conjugation of array-like types
0fd078e5 - transpose array-like in herm_conj

Compare with previous version

changed the description

changed target branch from master to stable

I rebased to stable and separated test and fix commits. The pipeline will fail, however, before f8331485 is not applied. I still suggest to cherry-pick it and then rebase the MR on top of it.