Inconsistent behavior when extending a application generator versus generating an app using the CLI

[beeman]

Current Behavior

In this project we extend the Next and React generators to create apps.

In a default Nx 18 and 19 project (with Crystal) the app that gets generated by our extended generators, the content of project.json is verbose as can be seen in this generated app.

However, if we generate a second app using the nx CLI in the same workspace, the content of project.json is (mostly) using the inferred targets as can be seen here.

The curious thing is that this is happening without having set useInferencePlugins in nx.json or having the env var set.

Expected Behavior

Considering the workspace does not have useInferencePlugins set, I would expect the extended generator to behave the same as when we're using the CLI.

GitHub Repo

https://github.com/beeman/solana-dapp-react-version

Steps to Reproduce

1. Create a new app using pnpx create-solana-dapp@next
2. Pick the React option (for consistency with the example, same happens with Next though).
3. Verify that web/project.json is verbose
4. Create a second app: pnpm nx g @nx/react:app second --e2e-test-runner none --routing false
5. Verify that second/project.json is inferred.

Nx Report

❯ pnpm nx report

 NX   Failed to process project graph. Run "nx reset" to fix this. Please report the issue if you keep seeing it.

Pass --verbose to see the stacktraces.

❯ pnpm nx reset

 NX   Resetting the Nx workspace cache and stopping the Nx Daemon.

This might take a few minutes.


 NX   Daemon Server - Stopped


 NX   Successfully reset the Nx workspace.

❯ pnpm nx report

 NX   Failed to process project graph. Run "nx reset" to fix this. Please report the issue if you keep seeing it.

Pass --verbose to see the stacktraces.

Failure Logs

No response

Package Manager Version

pnpm 9.0.5

Operating System

• macOS
• Linux
• Windows
• Other (Please specify)

Additional Information

This also happened with Nx18 but I decided to wait till Nx19 to make sure it wasn't fixed there.

Somehow, nx report fails in my demo project, I'll create a separate issue for that (#23296).

[Coly010]

Hey @beeman!

Perhaps there is a misunderstanding here on Inferred Projects with new workspaces.
By default, inferred targets is turned on.

The useInferencePlugins and env var are options provided to allow people to specifically opt-out of inferred targets. Therefore when our first party plugins (such as @nx/react) are run, they will attempt to use inferred targets, even if other projects in the workspace are using defined projects in project.json files.

I'm going to convert this issue to a discussion because this is correct behaviour but perhaps it should be discussed more.

[beeman]

Hi @Coly010, thanks for the quick reply!

I indeed think I don't understand how it's supposed to work.

As you can see here, we are using first-party plugins to generate these things. We wrap them, but the actual generator is the first party plugin.

I would expect this to work just like how it's when I invoke nx g @nx/react:app.

The important part is that I expect this to work predictably and consistently. Currently, it's none.

I don't want to force people into disabling inferred plugins as I think it makes sense for people to get a generated project that's as similar as if they would have run the commands themself (create-nx-workspace, install @nx/react, then generate the app). That's why I don't set the useInferencePlugins flag, nor set the env var. Still the 2 options behave differently.

Somehow, this is not the case when I extend, but it is the case when I manually invoke it which causes my confusion.