Drop xvfb-run wrapper, no longer needed since LibrePCB 2.2.0

[firuzakhmad]

Draft on #1793.

Removed the wrapper script and symlinked librepcb-cli directly into
/usr/local/bin, per your suggestion. Kept the actual install under
/opt/librepcb rather than extracting there directly, since Qt looks
for its plugins relative to that path. Also dropped xvfb from the
apt install list, since nothing else needs it.

Confirmed locally that this can't work yet: building against the
current LIBREPCB_VERSION=2.1.1 fails at the "RUN librepcb-cli
--version" sanity check with "error while loading shared libraries:
libGL.so.1: cannot open shared object file", since 2.1.1 has no
fallback logic and tries to initialize the xcb platform without any
display present. Expected, and confirms this needs to wait for a
2.2.0 image.

I'll bump LIBREPCB_VERSION and confirm this actually builds and runs
cleanly once 2.2.0 is out, then mark this ready for review.

[ubruhin]

The wrapper script has no function anymore, so it should be removed completely. Just install the CLI directly to /usr/local.

[firuzakhmad]

Removed the wrapper script and symlinked librepcb-cli straight into
/usr/local/bin instead. Kept the actual install under /opt/librepcb
rather than extracting there directly, since Qt looks for its
plugins relative to that path.

[ubruhin]

Keeping the actual install tree at /opt/librepcb rather
than extracting straight into /usr/local, since Qt resolves plugin
paths relative to that layout.

Is that sentence from an LLM or is that your observation? I just can't really believe that, so I wonder if that is from a hallucinating LLM. Just to make it clear: I don't want to be the reviewer of LLM output. It is fine if you use LLMs to get work done, but it is your job to review and correct its output. Your PRs are not bad so far, but still I have the impression that I am reviewing the (unvalidated/unpolished) work of LLM instead of work from (or verified by) human. Is my impression right?

Though maybe I just misunderstand the commit message. I agree the relative path between the binary and the plugins must not be changed, but I see no reason for not installing the whole librepcb package to /usr/local (make install would do it exactly that way too).

[firuzakhmad]

Fair question! Yes, that line came from LLM while I was working through the issue. I took it at face value and didn't verify it before including it. Looking at it again, it doesn't actually hold up. Qt resolves plugin paths relative to where the binary is located, not a fixed path like /opt/librepcb. So if the release tree is extracted into /usr/local while preserving the same directory structure, it works the same way. The symlink wasn't actually necessary, I stated it as fact without confirming it first.

Your broader point is fair as well. Going forward, I'll make sure what I submit is based on things I've actually verified myself. I'll go back and fix this properly.

[ubruhin]

Alright, thanks for your honest clarification 👍

[firuzakhmad]

Pushed! Straight to /usr/local now, no more /opt + symlink.

[ubruhin]

Thanks, LGTM 👍 Will merge when the next release is available.

[firuzakhmad]

Thank you for reviewing it! Sounds good, will follow up when the next release is available.