When reconstructing a tag which has the function of setting a variable to the result of an expression (both set and for do this), we should preserve the AstIdentifiers in the expression rather than resolving them because they may be directly referencing a value which can be modified later on. If instead, we set our variable to a serialized version of the object, then if that object is modified, the changes will not be reflected in the set value.

The simplest example of this is shown in one of the new tests:

{% set dict = {} %}

{% set foo = (dict, deferred) %}

{% do dict.update({'a': 'b'}) %}

{{ foo }}

If we don't preserve the identifier for dict, then we end up with:

{% set dict = {} %}

{% set foo = ({}, deferred) %}

{% do dict.update({'a': 'b'}) %}

{{ foo }}

Which would result in foo not having the correct value within it!

I realised that this can also happen if a value is reconstructed and then later one of the elements inside of that value is updated. Such a case is much more difficult to handle. I created a separate issue for that: #987