Commit Graph

8 Commits

Author SHA1 Message Date
Reactor Scram
64f0427ef4 ci(gui-client): hide the Linux GUI deb since it's not ready yet (#4258)
It's still in the CI artifacts for easy testing, but there's no point
letting users see it since it's in the middle of the process split
re-architect
2024-03-21 23:49:34 +00:00
Reactor Scram
b0904e382a chore: add crate for privileged Linux tunnel process (#4229)
Refs #3713 

```[tasklist]
### Before merging
- [ ] Is 'firezone-client-tunnel' okay for the binary name?
- [ ] Using a library and building it as two binaries is correct, right? `cargo run -p firezone-client-tunnel` takes 1 second. `cargo run -p firezone-gui-client --bin firezone-client-tunnel` takes 1m42s because it builds all the GUI deps.
```
2024-03-21 14:06:55 +00:00
Reactor Scram
e05cbbe0a0 build(gui-client/linux): include an empty firezone-tunnel binary with the Tauri deb package (#4220)
I thought this was going to use `cargo-deb` but it was actually easy
with the Tauri deb bundling we already use.

```[tasklist]
### Before merging
- [x] Make sure every file in the Tauri deb is also in our deb (e.g. icons)
```
2024-03-20 14:11:41 +00:00
Reactor Scram
651ea3ae00 build(gui-client/linux): make sure debug symbols get uploaded for the Linux GUI client (#4217)
- Split up CI artifacts into "exe", "pkg", and "syms" so it's easy to
check they're being uploaded. This shouldn't affect published artifacts
- Set `strip = "none"` which seems to be necessary to get the debug
symbols in Linux, although they still end up in the exe and not the dwp
file 🤔 don't know why
- Test Linux stacktrace in CI

Stacktrace examples:
- On Linux we at least get function names, but we aren't getting line
numbers for some reason
https://github.com/firezone/firezone/actions/runs/8350493514/job/22857032124#step:10:268
- On Windows we also get line numbers, as before
https://github.com/firezone/firezone/actions/runs/8350493514/job/22857033367#step:11:351

I didn't test downloading the files and doing a stacktrace locally, but
I have batched that up for whenever I do a big manual test of the
CD-produced release artifacts:
https://github.com/firezone/firezone/issues/3887
2024-03-19 22:18:03 +00:00
Reactor Scram
74026d8b13 build(gui-client): disable AppImage bundling (#4216)
AppImages won't work with process splitting. (#3713)

As far as I can tell, they just produce one binary. Internally they use
FUSE or something to mount a squashfs image, but that image won't be
able to hook into systemd and run with root permissions and everything.
I don't think it's practical, and Tauri's AppImage bundling doesn't have
the features for it.

Even their deb bundler doesn't have any way to specify a path for a
daemon to be installed. The sidecar feature only seems intended for the
GUI app to call, not anything else on the system.

(There is such a thing as installing AppImages, but I don't think it's
worth pursuing - We should just do debs)
2024-03-19 17:26:25 +00:00
Jamil
9cab250696 chore(windows): Sign internal exe using beforeBundleCommand (#3994)
Refs #3230 

It looks like we need to sign the internal exe before it gets bundled
too. We can use `beforeBundleCommand` to do so.

Soon, Tauri should have native support for this exact scenario:
https://github.com/tauri-apps/tauri/pull/8718
2024-03-06 16:00:54 +00:00
Jamil
19e833262f chore(windows): Sign windows exe too (#3992)
Fixes #3230
2024-03-05 22:35:24 -08:00
Reactor Scram
fd31152106 refactor(ci): enable Linux do-nothing GUI builds (but not tests) in CI/CD, extract scripts for that (#3735)
Builds a do-nothing `return 0` Linux client to make sure the CI/CD
scripts are set up and producing AppImage / deb bundles as expected.


![image](https://github.com/firezone/firezone/assets/13400041/7d2d8f02-adde-4b1b-89ec-02aaf112ac48)

---------

Signed-off-by: Reactor Scram <ReactorScram@users.noreply.github.com>
2024-02-23 17:57:39 +00:00