smartconfig/DESIGN.md
sneak cf13b275b7 Squashed commit of the following:
commit e15edaedd786254cf22c291a63a02f7cff33b119
Author: sneak <sneak@sneak.berlin>
Date:   Mon Jul 21 15:17:12 2025 +0200

    Implement YAML-first interpolation with type preservation

    Previously, interpolations were performed during string manipulation before
    YAML parsing, which caused issues with quoting and escaping. This commit
    fundamentally changes the approach:

    - Parse YAML first, then walk the structure to interpolate values
    - Preserve types: standalone interpolations can return numbers/booleans
    - Mixed content (text with embedded interpolations) always returns strings
    - Users control types through YAML syntax, not our quoting logic
    - Properly handle nested interpolations without quote accumulation

    This gives users explicit control over output types while eliminating
    the complex and error-prone manual quoting logic.
2025-07-21 15:19:28 +02:00

61 lines
2.3 KiB
Markdown

# Design: YAML-First Interpolation
## Problem Statement
The current implementation performs string-based interpolation on the raw YAML file content before parsing. This approach has several issues:
1. **Quoting complexity**: We must handle YAML quoting rules manually, leading to edge cases and bugs
2. **Type loss**: All interpolated values become strings, even if they should be numbers or booleans
3. **YAML special values**: Values like "no", "yes", "true", "false", "on", "off" require special handling
4. **Nested interpolation quoting**: Complex logic needed to avoid double-quoting in nested interpolations
5. **Fragile string manipulation**: Operating on raw YAML text is error-prone
## Proposed Solution
Parse the YAML file first, then walk the parsed structure to perform interpolations. This approach:
1. **Leverages YAML parser**: Let the YAML library handle all parsing, quoting, and escaping
2. **Preserves types**: Can return actual numbers, booleans, etc. from interpolations
3. **Simplifies logic**: No string manipulation or quoting logic needed
4. **Clear semantics**: Users control types through YAML syntax (quoted vs unquoted)
5. **Robust**: YAML serializer handles all edge cases correctly
## Implementation Details
### Constraints
- Interpolation only occurs in YAML values, not keys
- Users must properly quote interpolations if they want string output
- The entire file must be valid YAML before interpolation
### Workflow
1. Parse YAML file into a data structure (map/slice/scalar)
2. Recursively walk the structure:
- For string values: check for `${...}` patterns and interpolate
- For other types: pass through unchanged
3. Handle the special `env` key to inject environment variables
4. Return the modified structure
### Type Handling
Users control the output type through YAML syntax:
```yaml
# String output (user quotes the interpolation)
port: "${ENV:PORT}" # Result: "8080" (string)
# Numeric output (no quotes)
port: ${ENV:PORT} # Result: 8080 (number)
# Boolean output
enabled: ${ENV:ENABLED} # Result: true/false (boolean)
```
### Benefits
1. **Correctness**: YAML parser handles all edge cases
2. **Type safety**: Preserves intended types
3. **Simplicity**: Removes complex string manipulation
4. **Predictability**: Users have explicit control over types
5. **Maintainability**: Less code, fewer edge cases