Fixing bugs in the LTS version 1.6.5

Specific Domains Numerics
1

I wonder how realistic is for package developers to assume that bugs reported for the LTS version 1.6.5, which are already fixed in 1.7.1, will be also fixed for the LTS version. Personally, I am facing the dilemma that, since I included the LTS version as compulsory test platform for four packages on which I am working, I am not sure if this is a wise option for me, because failures due to bugs in 1.6.5 can still happen.

Concretely, I reported in #40492 a failure of eigvals for a certain complex matrix for Julia 1.6.0, and the issue was closed because it was fixed in 1.7.1! But this failure continues to exist also in the LTS version 1.6.5 and for other complex matrices encountered in my tests. So, the question is what perspectives exist to have this error corrected in 1.6.5 in a reasonably short time. Otherwise, the only option for me is to drop the compatibility of my packages with the LTS. I wonder if this is the only option for package developers. I would be glad to know opinions on this aspect.

2

If someone points out what was the fix that made it work in v1.7, then it can be backported, unless the fix involved the introduction of a new feature. But someone needs to dig out the fix and report it.

3

I think the fix for 1.7.1 was not related to my concrete bug report, but it was at another level (for example, linking with another version of LAPACK). But I am not sure for this.

4

If you mean the version of OpenBLAS shipped by Julia, I’m under the impression that upgrading it is considered akin to introducing new features, so unlikely to be backported.

5

I think I found the issue where the convergence problems of generalized complex eigenvalues have been discussed in the LAPACK/Julia context
#477 (see also #475). Backport issues to Julia 1.6 are discussed in #39216 in relation to Upgrade to OpenBLAS 0.3.13, but apparently the idea has been rejected. However I had the impression that backporting would be still possible.