fix(config): enable file include in scoped queries

[becm]

Avoid duplication of entries already present in included files:

• consistent include of files in Git config queries with scope limit
• check for sufficient detection of present values, assignments always create values in main files
• add test for file includes

Improves consistency regarding include behavior with cached Git config and

Fixes #2342
Fixes #1696
Fixes #1337

[derrickstolee]

@becm Thanks for identifying this key piece that connects to several issues. Could you add a test to demonstrate the behavior change to prevent regressions in the future?

[becm]

@derrickstolee sadly only the 1st step, since:

• the assignment seems to be unconditional
• I did not even manage to get all instances where the --list request is issued.

[becm]

@derrickstolee since the --include stuff is dependent on partnered Git version and file setup, I'd have no idea how to attempt a proper test (existing examples just operate on unrelated mock data).

Will try to implement a test for (abstract) config data with the (hopefully) resolved corner case of configure.

[becm]

Reverted PR to primary goal, will tackle git-credential-manager configure issue in separate task.

@derrickstolee I'll try to add a test for file inclusion (global scope).

[becm]

@mjcheetham any further actions to be taken to accept this?

This change will essentially harmonizes behavior before and after Git 2.54 .
#318 changed the default includes behavior by splitting global and local.
#2268 basically introduced holistic config parsing with --show-scope again.

[becm]

@derrickstolee it seems git-credential-manager only ever uses GitConfigurationType.Raw (in non-test code).

Would it be possible to introduce a further code path for Git 2.26+ without the --no-type argument that only uses/generates raw cache and only depends on --show-scope?

This would be an extension or (for non-ancient Git versions) a replacement for the current PR.

[dscho]

The way I read this change, it constitutes a change of behavior. A good one, granted, but one that has a high potential of affecting existing users' setups (especially complex setups). As such, I'd be much more comfortable to make this behavioral change part of a major version bump. Seeing as v3.x is starting to take concrete shape, I would consider that version to be the natural inflection point to integrate this here PR.

[becm]

@dscho behavior change already happened for 2.8.0 with #2268 and #2296 for certain newer Git versions.

This combination effectively uses --show-scope and full config parsing again (which has further impacts).
So this would be an attempt to harmonize behavior in a 2.8.1 for older Git versions.

[dscho]

behavior change already happened for 2.8.0 with #2268 and #2296 for certain newer Git versions.

@becm sure, but the way I see it was that this was an inadvertent behavioral change. Here, we are aware of the change. That makes a huge difference.