ubuntu-drivers debug wrapper #2026
Conversation
|
After out of band discussion, the system_scripts approach isn't what we want to go with. There's a different approach with just using binaries vendored in |
.flake8
Outdated
| @@ -3,3 +3,5 @@ | |||
| builtins = _ | |||
| max-line-length = 88 | |||
| extend-ignore = E203 | |||
| exclude = | |||
| subiquity/system_scripts | |||
There was a problem hiding this comment.
if it's under subiquity I want it passing flake8. It looks like it's just 2 things to fix, noqa is fine for long strings that really do need to be long.
There was a problem hiding this comment.
Ack. I guess this is all moot since I'm going to tear this out anyways, but I was concerned something like
scripts = """\
some text
really long line # noqa
another really long line # noqa
"""would be weird since it technically embeds the comment in the string, but you can just do
scripts = """\
some text
really long line
another really long line
""" # noqait turns out.
| @@ -0,0 +1,69 @@ | |||
| script = """\ | |||
There was a problem hiding this comment.
as already discussed, I don't want to write these scripts out to use, I want them directly in the snap.
Chris-Peterson444
force-pushed
the
fake-ubuntu-drivers-new
branch
from
a1d4b20
to
f168f57
Compare
subiquity/server/ubuntu_drivers.py
Outdated
| self.wrapper_script_source: str = umockdev_wrapper.script | ||
| self.script_path = "/tmp/umockdev_wrapper.py" | ||
|
|
||
| self.dev_config = configs.broadcom |
There was a problem hiding this comment.
This is a bit hardcoded which means that we can't actually use the other config blobs, such as dell_certified_with_nvidia, without source modification. Thoughts?
There was a problem hiding this comment.
I'm thinking with moving away from the system_scripts module, we have a default config like:
self.dev_config = {
"devices": [
{
"modalias": "pci:v000014E4d00004353sv00sd01bc02sc80i00",
"vendor": "0x14E4",
"device": "0x4353",
}
]
}which gets written on boot to a known location (i.e. /tmp/umockdev_config.yaml) and then we just modify that with early-commands or similar before continuing with the install. This is also the inspiration for 1d56545. What do you think?
Adds the *_ORIG variables to the subiquity-server environment so orig_environment will correctly replace mocked variables (e.g., $PATH).
Introduces a special set of binaries we will use to be run in the system environment. This commit introduces subiquity-umockdev-wrapper, which can be used to wrap ubuntu-drivers calls to fake hardware.
Utility to generate an environment for running programs outside of the snap with the addition of the programs vendored in system_scripts.
Chris-Peterson444
force-pushed
the
fake-ubuntu-drivers-new
branch
3 times, most recently
from
f9e9cbc
to
2e09905
Compare
Introduces a new UbuntuDriversInterface class that will wrap all "ubuntu-drivers" calls with the subiquity-umockdev-wrapper script. install_drivers no-op until further work to copy system_scripts to target is done.
Add option to pass "subiquity-fake-pci-devices" on the kernel command line.
For server installs, we specifically pass "--gpgpu" to ubuntu-drivers calls, but this stops broadcom-sta-dkms drivers from being a candidate to install. It's faster to test on server, so we can enable this flag to test installs of braodcom-sta-dkms on server (which normally shouldn't happen).
Add option to pass "subiquity-server-force-no-gpgpu" on the kernel command line.
Wrap the ubuntu-drivers command on subiquity-umockdev-wrapper during install too. Works by copying the wrapper script to /target/usr/bin so it will be found within $PATH by default.
Chris-Peterson444
force-pushed
the
fake-ubuntu-drivers-new
branch
from
2e09905
to
3cfb8a5
Compare
|
Apologies for the noise. Somehow I didn't have pre-commit installed. This is now ready including wrapping the install step, albeit not very elegantly for now. |
Introduces:
subiquity.system_scripts, to store scripts which subiquity can write to the system and call externally.