Add a hook_drush_sitealias_alter, with documentation.

[greg-1-anderson]

Deprecate 'parent' element in site aliases, for eventual removal (probably in Drush 8).

[greg-1-anderson]

'parent' might be a seldom-used feature. If we think folks aren't using it much, we could just flat-out remove it in Drush 7 and be done with it.

[nerdcorenet]

I am using the 'parent' elemtn quite extensively in drush aliases. It allows us to define a server environment and have all sites residing on that server inherit things like the remote-host and root values.

[esolitos]

Does the 'parent' complicate the code a lot?
Because in our case it's a "must" feature of drush, using multiple remote servers we're using it quite extensively.

[greg-1-anderson]

To follow up on my tweet:

If you are using the 'parent' feature, read the patch and see if the policy file solution works well for your use-case.

The thing we are trying to change here is transitioning to loading alias files from yaml rather than loading them in code.

[davereid]

Acquia uses @parent in all of their Drush alias files for Cloud sites. Might want to get someone from their infrastructure team's input?

[greg-1-anderson]

Probably deprecate this in Drush7, remove in Drush8, I'm guessing, then.

We'll also have to consider if the submitted patch is too noisy for folks. :)

[greg-1-anderson]

I mis-spoke above: there is no real requirement to remove 'parent' if we switch to yaml alias files. We could keep 'parent', and commit just the alter hook. But I know that moshe has wanted to get rid of 'parent' for a while (see #975)

[weitzman]

I really like this PR. It moves the complexity of parents to a policy. It is valuable, independant of the transition to YAML site alias files.

The core issue is that greg is the sole person who works on backend and site aliases. Much of the code others could maintain but some is gnarly and rarely used; I think 'parent' is in that category. Drush project is too crucial to ignore this SPOF problem.

[greg-1-anderson]

Shall we commit it as-is, then, with 'parent' deprecated and noisy, but still functional?

[weitzman]

I think we can turn down the log level a notch and then merge.

[kenorb]

It's usually advised to use array_key_exists() for arrays instead.

[greg-1-anderson]

Deprecated: http://en.wiktionary.org/wiki/deprecated

[greg-1-anderson]

I turned the log level down to 'debug' and merged.

End effect: 'parent' element is now deprecated in Drush 7, although it still works just fine, and the deprecation warnings are only visible in "debug" mode.