Restricted Mix umbrella recursion

[yrashk]

In some cases, I find it very useful if we could limit umbrella recursion to a subset of apps. For example, for me, a very pressing issue is selective testing as running tests on all apps all the time is a time killer. If we had a way to specify a regexp that needs to be run against every project found in the apps_path (if the regexp is specified), it would instantly solve the problem. The patch is trivial, if everybody agrees we need this, I can submit a PR.

cc @ericmj

[yrashk]

Another option is a way to specify a glob expression instead of a regexp — so that one can specify "{myapp1}" instead of the default of "*"

[josevalim]

How would that solve the general problem? Do you mean that you always want to skip some apps? Not only for testing, but also for compilation?

[yrashk]

Yes, this would hep skipping apps — there are different cases. Besides the aforementioned case, I also would love to skip one app when I am not in the test mode (I have an app called mysystem_test that only provides functionality for tests, and I really don't want it in the compile cycle for the :dev/:prod env).

Can I submit a PR if it is agreed?

[josevalim]

I am wondering how the solution will look like even to support your own use cases. You want some tasks to go over some apps, other tasks to go over other apps, and so forth. How would the configuration look like?

[yrashk]

Oh, I see, filtering that depends on the task?

[josevalim]

It seems to be what you are asking. :) I wonder if this is easier solved with custom tasks or custom umbrella app configurations.

[yrashk]

I'll be fine if I can define not just apps_path: "apps" but apps: %w(app1 app2)

[ericmj]

Maybe we can solve it with environment variables: $ MIX_APPS=app1,app2 mix test. Or would you prefer to have it setup in mix.exs?

[josevalim]

I like the environment variables idea. Maybe support some sort of value to remove an app too. Just a proposal:

• MIX_APPS=~app1 (all minus app1)
• MIX_APPS=~app1,~app2 (all minus app1 and app2)

[yrashk]

I would then strongly urge to have an apps key instead so you can compute
it however you want from OS env or not — that is MUCH more flexible

On Fri, Sep 27, 2013 at 1:53 PM, José Valim notifications@github.comwrote:

I like the environment variables idea. Maybe support some sort of value to
remove an app too. Just a proposal:

• MIX_APPS=~app1 (all minus app1)
• MIX_APPS=~app1,~app2 (all minus app1 and app2)

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/1730#issuecomment-25239844
.

[josevalim]

@yrashk sounds fair to me. The only issue I see with that is that if you are using two umbrella apps with their custom implementation of MIX_APPS environment lookup, they'd conflict with each other unless you give them different names. Also, every time you have those particular umbrella apps as dependencies, you would be calculating those values from environment, which may make compilation slower.

Supporting by default an environment variable would help solve those problems (because we would check the MIX_APPS var just for the first umbrella app).

Just putting the trade-offs out there, I think both solutions would be great. Wdyt @ericmj?

[ericmj]

@josevalim Could we not check the apps key for just the first umbrella app?

I'm not sure what I think, MIX_APPS is less flexible but it is also a nice convention. Maybe the mix new --umbrella template could generate an apps: System.get_env("MIX_APPS") option so we use the env var by default.

[yrashk]

Good idea.

On Fri, Sep 27, 2013 at 2:37 PM, Eric Meadows-Jönsson <
notifications@github.com> wrote:

@josevalim https://github.com/josevalim Could we not check the apps key
for just the first umbrella app?

I'm not sure what I think, MIX_APPS is less flexible but it is also a nice
convention. Maybe the mix new --umbrella template could generate an apps:
System.get_env("MIX_APPS") option so we use the env var by default.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/1730#issuecomment-25241957
.