Comparisons

salsa

salsa is the closest conceptual ancestor:

• incremental recomputation
• dependency tracking
• memoization

In dodeca, salsa persistence was implemented by serializing the salsa database with postcard and writing it to disk.

picante is different in a few key ways:

• tokio-first / async-first: derived queries are async and single-flight.
• facet-based persistence: picante avoids serde and uses facet + facet-postcard.
• no durability tiers: picante tracks dependencies precisely, so it doesn't need them.

Why no durability?

salsa has "durability" levels (Low, Medium, High) that let you hint how often an input changes. High-durability inputs (like config) trigger broader invalidation when changed.

picante doesn't need this because it tracks dependencies precisely:

• Each derived query records exactly which keys it read
• A reverse dependency graph maps each input to its dependents
• When an input changes, only queries that actually depend on that specific key are invalidated

Durability is an optimization to skip dependency checking — the assumption being "if config changed, just rebuild everything." But with precise tracking, you get the same result automatically: config changes only invalidate queries that read config. Template changes only invalidate pages using that template. No manual annotation needed.

Plain memoization

Simple memoization caches values, but without dependency edges you can’t answer:

• “what depends on this input?”
• “what needs to be recomputed?”

picante records explicit dependencies between queries while computing.