fix(dbt): preserve source identifier quoting when it contains dots or spaces

[jagannalla]

Problem

When a dbt source table identifier contains a dot (e.g., "FILENAME.CSV") or a space, SourceConfig.canonical_name() unconditionally strips the quoting. This renders the canonical name as database.schema.FILENAME.CSV.

When sqlglot tries to parse this unquoted string back into an AST, it splits the path by . into 4 components, causing a ValueError: too many values to unpack (expected 3) and failing to load the project.

Solution

This PR checks if the resolved table name contains a dot (.) or a space ( ), and if so, preserves the quoting for the identifier component during canonicalization. Standard table names remain unquoted and normalized.

Test Plan

• Added test_source_canonical_name_with_dots_and_spaces to tests/dbt/test_config.py validating correct quoting behavior for dots, spaces, target database omission, and standard identifiers.
• Ran all dbt config tests successfully (64 passed).
• Verified style, formatting, and types with pre-commit (ruff, ruff-format, and mypy all passed).
  Fixes SourceConfig.canonical_name() fails with ValueError: too many values to unpack for source identifiers containing dots #5762

[StuffbyYuki]

@jagannalla Does it make more sense to have logic like:

identifier = relation.identifier or ""
needs_identifier_quoting = "." in identifier or " " in identifier

than

needs_identifier_quoting = bool(
    self.table_name and ("." in self.table_name or " " in self.table_name)
)

?

The former approach might be slightly safer because it checks the final, dbt-resolved name, not just what's in the config file.

Let me know if I'm missing anything!

[noezhiya-dot]

Clean fix. The conditional quoting logic is straightforward and the test coverage is thorough — covers dots, spaces, standard identifiers, and the database-omission edge case. One small observation: the needs_identifier_quoting check runs on every call even when the identifier is already quoted by dbt. Since the downstream relation.quote() handles double-quoting gracefully, this is fine, but worth noting that the check is purely a string heuristic (not semantic). Approved on review.

[jagannalla]

Thanks for the reviews, @StuffbyYuki and @noezhiya-dot!
I have updated the PR to address the feedback:

• Changed the check to use the resolved relation.identifier instead of self.table_name to ensure we are always checking the final, dbt-resolved name.
• Verified that all tests are passing.
  Ready for merge!

[StuffbyYuki]

@jagannalla Can you make sure all commits have DCO checks? Thanks!

[noezhiya-dot]

Nice targeted fix. The root cause is clear: unquoting identifiers with dots/spaces breaks sqlglot's AST parsing. The fix is minimal and correct — only preserve quoting when the identifier actually needs it. Tests cover all the key cases (dots, spaces, standard identifiers, target database omission). LGTM.