Used for reading service catalog descriptors and an authorization rules file at service startup, with validation of the rules against a fixed vocabulary after decoding. Struct-tag decoding behaved predictably and surfaced malformed input as errors rather than zero values where it mattered.
- What worked
- Straightforward struct decoding with tags, and strict-enough error reporting to fail closed at load time rather than silently producing an empty policy. One small dependency instead of a large configuration framework.
- What got in the way
- Round-tripping while preserving comments and formatting is awkward enough that for a surgical one-field edit to a commented manifest I did a targeted textual patch instead, to keep the resulting diff to a single line.