Add non-interactive usage

[bradwilson]

I use Ansible to automate my development machines. Due to limitations imposed by Linux distributions, I am currently using dotnet-install to install specific SDKs and runtimes. This comes with some significant downsides; namely, there is no uninstallation and/or upgrade scenarios supported by dotnet-install. dnvm seems like a natural fit.

However, I cannot automate dnvm as it stands today because it mandates interactivity. To make the switch, I would require non-interactive invocations:

• Ability to install the tool (don't prompt me for anything, ideally just give me a fixed URL to grab the latest dnvm binary to drop into a ~/bin folder)
• Ability to install an SDK
• Ability to install a runtime
• Use exit code 0 to indicate success, and non-0 to indicate failure

I would of course use dnvm in its interactive mode to perform upgrades as necessary (in much the same way I use Ansible to get software initially installed when building the dev environment, but rely on apt update && apt upgrade to keep it up to date).

Non-requirements:

• Machine parseable output is not required for my usage, though other users who wanted to write something like an Ansible module might eventually want the ability to do queries and operations and retrieve output that is formatted for machine consumption, like JSON. (To be clear: machine parseable output would be fine for my usage, so long as it's easy to read by the human 😄)

[richlander]

I just ran into this issue as well.

I was playing with Cursor Agents. It doesn't have .NET installed in its environment. It installed (the wrong) .NET via the install script. I thought I'd give it some AI-first instructions on how to use dnvm. It would be great if there was a non-interactive approach that works well in this scenario.

I'm happy to contribute to an AI-first README.

I'll probably start with AI instructions for the README. That will inform what we could do with dnvm.

Here's some of the chat log.

The feedback includes a reference to ChatGPT but I'm using Cursor. I asked ChatGPT for some peer-review feedback. It gave me good feedback and I shared that with Cursor as its marching orders.

[nagilson]

I believe dnvm update has -y to accept all by default, does that still require interactivity? Note that I tried to test myself but encountered an issue:

[agocke]

The current status of this issue is that it’s about 50% implemented and 50% tested.

As Noah mentioned, -y should basically do a standard install with no prompts, but it’s lightly tested.

Another scenario is just dropping dnvm into a bin folder. That should also work, but it’s about 0% tested. The big difference there is that dnvm itself won’t live in DNVM_HOME, and I might have some places in the code that assume that.

The last piece is a permanent latest url. This isn’t implemented. The trick is that I don’t run an actual webserver so I can’t do this with redirects. I have to deploy the binaries to the GitHub pages repo. I’m honestly not sure how well that will perform, but hopefully it’s good enough since dnvm is only about 5MB.

[richlander]

For my scenario, dnvm doesn't yet exist on the machine.

[bradwilson]

Even the stable URL is something I can work around by querying GitHub to know what the latest release is, as I use this technique for other downloads (like Docker Compose):

$ curl -s https://api.github.com/repos/dn-vm/dnvm/releases/latest | grep "browser_download_url.*linux-x64.tar.gz\"" | cut -d : -f 2,3 | tr -d \ \"
https://github.com/dn-vm/dnvm/releases/download/v0.9.7/dnvm-0.9.7-linux-x64.tar.gz

No idea if this helps @richlander's scenario for download or not. You can tweak the grep for the bits you're looking for:

Conversion from Linux CLI tools to something else are an exercise left up to the reader. 😁

---

The big thing for me is being able to recreate the equivalent of having gone through the install, without interactivity. If that means setting know environment variables, that's something I could do. Better might be if we could repurpose some of the machinery inside to do a non-interactive version of the install, so that I'm not constantly tracking any changes you need to make to the shell startup in my Ansible script.

[agocke]

@bradwilson When you say "go through the install" -- do you want to just install dnvm or do you also want it to automatically install a .NET SDK? That is, obviously copying dnvm alone will only install dnvm. But running curl ... | sh -s -y will run with the default interactive options, which includes tracking the latest SDK. I'm curious which one would be better for you.

[bradwilson]

I will be installing SDKs, so long as I can specify them and get them installed non interactively. My installation scripts specify a list of versions to install.

[agocke]

@nagilson I have to admit, I'm fairly confused at how you ended up with an empty manifest file. Does this repro for you? Do you have any more details about how that happened?

[bradwilson]

I poked around a bit this afternoon, and it looks like I can get an "automated" self-install:

• Download a known "old" version of DNVM from GitHub
• Uncompress it into ~/.local/share/dnvm
• Run ~/.local/share/dnvm/dnvm update --self which will not only pull the latest, but also extract the env file (since that's an embedded resource, not in the tar file) which will also do the ~/.bashrc work
• Then I just run dnvm track on all the SDK versions I want to track

I'm going to push this into Ansible, but I'm guessing maybe I don't actually need anything, since I can (as I mentioned above) "repurpose some of the machinery inside to do a non-interactive version of the install", since that appears to be what dnvm update --self is doing. 👍🏼

[bradwilson]

Yep, that works. https://github.com/bradwilson/ansible-dev-pc/blob/main/dotnet/core.yaml

Edit: I do need to add logic to see what's already being tracked. You're storing that in a JSON file, which I could read. For now, though, it's good enough to get me started.

Edit: The JSON reading piece is done, so now everything is idempotent.

This bit may look confusing:

manifest_json: "{{ lookup('file','~/.local/share/dnvm/dnvmManifest.json') | regex_replace('', '') | from_json }}"

In the actual source file, the left hand string of the regex_replace is actually the Unicode Byte Order Mark (0xFEFF), which GitHub appears to be filtering out in the source view. In VS Code:

BOMs are mostly a "Windows disease", so of course from_json throws an error trying to parse it.

[richlander]

BOMs are a "Windows disease",

• Reconsider including BOM in templates dotnet/sdk#39187
• All release notes should be consistent for BOM and LEs dotnet/core#9957

I appreciate your support in this fight.

[bradwilson]

@richlander I required all files to be BOM free automatically in my CI build script:

https://github.com/xunit/build-tools-v3/blob/3924f254f246ef8ac0eddf4a54f5b012faa74160/models/BuildContext.cs#L173-L204
https://github.com/xunit/build-tools-v3/blob/3924f254f246ef8ac0eddf4a54f5b012faa74160/targets/FormatSource.cs#L21-L31

And I removed BOMs from all my output reports (it was at request of a user, who wanted to post-process them, but I should've known better already): xunit/xunit#3224

[bradwilson]

I appreciate your support in this fight.

@richlander Did you want more than me just putting a thumbs up on the two issues? 😄

[agocke]

I'm doing my own json writing and it should definitely not be writing a BOM, so I have to investigate how that's getting in there.

[agocke]

Oh great, Encoding.UTF8 has a BOM by default...