Run test-suite environment setup inside the image #3531
Conversation
Commit c17cc69 introduced an `rpmdb --initdb` call into test-suite make_install() but at that point, it's actually the rpm from the host, which is NOT what we want. Split the image preparation to two stages: install, which runs from the outside and basically just copies data into the image, and setup which runs from inside the image. For this we need some extra container gymnastics. Fixes: rpm-software-management#3530
|
Oops, some debugging leftovers in there... (fixed) |
|
I think there are two ways to do the split:
This patch is a blend of both, i.e. the split seems a bit arbitrary 😅 (Of course, the most important part is the rpmdb commands which are now run inside the container, so that's good.) That said, we can always improve this later so it's not a blocker. |
|
Right, the split-line is a bit fuzzy but basically, only do on the outside what has to be on the outside: copying data into the image. Beyond that whats install and what's setup gets increasingly fuzzy. I left the ldconfig stuff alone because it's kinda part of the "cmake install" step, but could certainly be moved to setup as well. If rpm shipped it's own ld.so.conf.d file, putting it into place would be necessarily part of install. But since we create it here, it can be either 😅 |
|
Hmm, but that said: ldconfig on the host might not be compatible with what's on the inside. So while putting the conf file in place can be considered install, running ldconfig should actually happen inside. |
Oh, indeed. You're right, I didn't realize this initially. We are copying stuff from the host, which is one way to draw the line, and I agree it's probably the most sensible 😄 Thanks for clarifying!
Yep, the ldconfig stuff wasn't necessarily what I had in mind, it was more about the various directories and whatnot that we copy/create, but I stand corrected, see above 😄 |
|
OK, looks good otherwise, thanks! |
|
Oh, I was about to move running ldconfig into setup because it is actually wrong on the install side, but there's always the next PR 😄 |
|
Oops, my merge-this finger was itching too much I guess 😄 But yup, there's always the next one. |
Commit c17cc69 introduced an
rpmdb --initdbcall into test-suite make_install() but at that point, it's actually the rpm from the host, which is NOT what we want.Split the image preparation to two stages: install, which runs from the outside and basically just copies data into the image, and setup which runs from inside the image. For this we need some extra container gymnastics.
Fixes: #3530