xtb-6.7.1 fails to rebuild

[sassy-crick]

Describe the bug
Trying to rebuild xtb-6.7.1 fails, despite using the same procedure

To Reproduce
Steps to reproduce the behaviour:

I am part of the EasyBuild community and as such we are working with checksums of the downloaded sources, and also with build-scripts which means builds can be reproduced, independently of where and which OS is being used. However, we recently noticed that one of these recipes fails to be reproduced.
Full detail can be found here
From what we understand, the problem appears to be the pulling down of a substantial amount of dependencies, and it appears it is not the first time that happens (see Issue #23105 and #1291. It might be good practice to change that over to a a more robust solution, maybe something similar what EasyBuild does?

Please provide all input and output file such that we confirm your report.

Expected behaviour
Scientific software must be reproducible and thus installation must be reproducible and not shift over time

Additional context
Full detail can be found here

[foxtran]

See also #1120

[boegel]

@foxtran Any specific thoughts on this?

Is there some high-level information available on how xtb deals with the required dependencies?

It seems like a bunch of things are being downloaded during the installation of xtb, and for some of those dependencies there's no specific tag/version being download (but latest master/main/whatever branch), which makes the installation of a specific version of xtb break over time?

[foxtran]

Any specific thoughts on this?

I would prefer to keep them in my mind 😇

Is there some high-level information available on how xtb deals with the required dependencies?

If you do not need them, just disable with -DWITH_CPCMX=OFF -DWITH_TBLITE=OFF in the case of cmake. Other deps should be compatible with xtb 6.7.1. Unfortunately, there is no info about specific versions of deps used for compiling xtb 6.7.1.

It seems like a bunch of things ...

Exactly. Unfortunately, it is still not fixed even in master branch for CMake... Meson support is a bit better in the master branch: now it is not breaking over time as far as I remember.

[sassy-crick]

Thanks for the feedback @foxtran
So one way forward would be to switch to meson for building if I read that correctly.
As we are also installing CREST and Censo, and as we are using the modules we are building for these builds. I am not sure if switching off tblite here is a good idea as we need it later. Thus, I think we would need to have that switched on.
I will try a meson build and report back. Bear with me please.

[sassy-crick]

I think I found the problem, as mentioned in the PR above: tblite was recently updated to 0.5.0 and that seems to be incompatible with xtb as here it just pulls in the latest version, and not a pinned down one. I think it would be better to use specific versions, not only to avoid building problems but also for reproducibility issues. After all, as somebody who is doing computational chemistry, I want to reproduce my work elsewhere and not on a specific machine. So for me that is a bug.

[foxtran]

Yep. That is also a problem for me... And after a lot of fixes regarding version's pinning, we are still have only one-year-old xtb 6.7.1 with those bugs.

[sassy-crick]

Well, that is the reason why I spent some time and effort to get that at least sorted with EasyBuild.
As a more general comment: I am not sure how robust the current way of doing things for xtb and CREST is. It appears one is using one version of tblite for example, and the other a different one. Using the programs at the same time will then be a bit more fun at one point, due to incompatibility I fear.

[Albkat]

Yep. That is also a problem for me... And after a lot of fixes regarding version's pinning, we are still have only one-year-old xtb 6.7.1 with those bugs.

Currently, we’re working on a new release, but the Windows build has become a nightmare since the deprecation of ifort (and icl).

We plan major changes after version 6.7.2 and, thus, must ensure that everything works at least as well as (and hopefully better than) in the previous releases.

[foxtran]

become a nightmare since the deprecation of ifort (and icl).

Just use ifx.

I can setup CI/CD that will produce Win binaries of xtb.

[sassy-crick]

So far, using ifx and icx i.e. the newer compilers, seems to have caused me little headaches when I am compiling code. I might have been just lucky though, at least on Linux. I cannot speak about Windows. There are a few flags which have changed though.

[Albkat]

become a nightmare since the deprecation of ifort (and icl).

Just use ifx.

I can setup CI/CD that will produce Win binaries of xtb.

I am using ifx.
Thanks, any help in this regard is greatly appreciated