Add rel_file to ansible_managed template variable

[dmke]

SUMMARY

This add a rel_file format string to the ansible_managed template variable.

Fixes #45654

ISSUE TYPE

• Feature Pull Request

COMPONENT NAME

template?

ADDITIONAL INFORMATION

The relative path resolution is very simplistic: it just strips the inventory_dir from the template_path, which may or may not be enough for this feature.

Here's a sample output of this patch applied to stable-2.9:

$ fgrep ansible_managed ./ansible.cfg
ansible_managed = Ansible managed: {rel_file} - DO NOT EDIT

$ cat playbooks/testbed.yml
---
- hosts: localhost
  roles:
  - testbed

$ cat roles/testbed/tasks/main.yml
---
- name: create file
  template:
    src: test.j2
    dest: /tmp/test.ansible

$ cat roles/testbed/templates/test.j2
{{ ansible_managed }}

$ ansible-playbook playbooks/testbed.yml --diff
PLAY [localhost] *************************************************************

TASK [Gathering Facts] *******************************************************
ok: [localhost]

TASK [testbed : create file] *************************************************
--- before
+++ after: /home/dm/.ansible/tmp/ansible-local-13670DFMU7R/tmpIK7AXO/test.j2
@@ -0,0 +1,1 @@
+# Ansible managed: roles/mm/templates/test.j2 - DO NOT EDIT

changed: [localhost]

PLAY RECAP *******************************************************************
localhost: ok=2 changed=1 unreachable=0 failed=0 skipped=0 rescued=0 ignored=0

TODOs

• Bike shedding: How should the variable be named? rel_file might not be descriptive enough.
• Path resolution: Is the current implementation too naïve? If so, what are the environments I should consider? Is basing this off of inventory_dir the wrong approach?
• Documentation: I have not yet touched the documentation. The rationale from "{rel_file}" in ansible_managed #45654 should be included.
• Test: As far as I can tell, this is not heavily tested. I also haven't touched the tests yet.

I'm not a Python developer by trade, so this small patch might contain non-idiomatic code. I'm open for suggestions.

[bcoca]

a couple of things
a) inventory_dir might not be present (see implicit localhost)
b) inventory is not the source for relative file paths, the 'play' is, if they happen to be the same in your setup that is coincidental, but inventory is not part of template/vars/files search path (it is for host_vars/group_vars, but that is diff system)

You might want to look at the dwim functions and how they use a search path, it is not as simple as you might think, specially when templates/ is optionally added and when roles appear in the mix, all create special cases.

[dmke]

a) inventory_dir might not be present (see implicit localhost)

I was not aware of this. However, if task_vars.get('inventory_dir') returns None, rel_path end up as being the same as path

b) inventory is not the source for relative file paths, the 'play' is, if they happen to be the same in your setup that is coincidental, but inventory is not part of template/vars/files search path

Indeed, I have inventory = ./hosts in my ansible.cfg (with both files checked into the repository).

You might want to look at the dwim functions and how they use a search path

Hm, grep'ing for dwim yields 80+ results, with path_dwim defined in lib/ansible/parsing/dataloader.py. Do you mean that class?

Alternatively, might a simple os.getcwd() suffice as base_dir for this purpose? Presumably, one would invoke ansible-playbook always from the same working directory? I mean, the use case for rel_file is a shared repository, which should be fully self-contained (i.e. containing one or more inventories, the Ansible configuration, playbooks, roles, and plugins, as this minimizes setup-time for each user interacting with that repo).

[bcoca]

yes, dataloader has several dwim (do what i mean) functions for resolving relative paths, depending on context, you end up using one of those

cwd is not correct, you need basedir, since you can cd /var/tmp && ansible-playbook ~/projects/play.yml. We purposefully avoid using cwd when we can to avoid issues like that, we are always relative to either inventory, playbook or role.

[dmke]

Okay, thanks for the info.

Can you recommend a debugger (or anything more sophisticated than print()) to step through the code base for runtime introspection? As I said, I'm not a Python developer, and the Ansible code base is... a bit overwhelming :-)

[bcoca]

epdb is THE debugger, others exist (i do abuse print() but also the q library).

In any case I would use the original argument, which might already be a relative path, if not, compare to the task's base dir and remove the common sides from the left.

[bcoca]

Closing this as other modifications introduced template_fullpath leaving template_path as the 'unresolved' path supplied to the template, this could be relative or not, but results in more or less the same as this modification that relies on having a base_path or not.