Snapshots could make named recipes reusable as downloads, reproduction fixtures, or cached generated output. Before adding storage, establish how a compiled profile relates to its declarative catalog and rendering inputs.
Project names, target paths, and configuration are open-ended. Precomputing every variation is not a useful first scope. Start with one named recipe and generate uncommon combinations on demand.
Investigate:
- Which consumer benefits from a compiled profile rather than generation?
- Which inputs identify it: catalog/fragment digest, normalized Selection, effective config, composer version, and generation stage?
- How do parametrized declarations and contributor provenance survive alongside concrete snapshot payloads?
- Does decoding actually beat generation for time, payload size, and memory?
Keep catalog operations authoritative. Two modules updating the same JSX slot or package.json still need semantic composition. VFS sibling deltas require their matching base and cannot simply be stacked.
A useful outcome is one reproducible round trip, explicit invalidation when an input changes, measured costs, and a recommendation to adopt, narrow, or stop. This is research, not approval for a registry service or marketplace.
Related: #249 owns declarative fragments; #252 owns portable artifact completeness and compatibility; #250 owns workspace materialization.
Snapshots could make named recipes reusable as downloads, reproduction fixtures, or cached generated output. Before adding storage, establish how a compiled profile relates to its declarative catalog and rendering inputs.
Project names, target paths, and configuration are open-ended. Precomputing every variation is not a useful first scope. Start with one named recipe and generate uncommon combinations on demand.
Investigate:
Keep catalog operations authoritative. Two modules updating the same JSX slot or
package.jsonstill need semantic composition. VFS sibling deltas require their matching base and cannot simply be stacked.A useful outcome is one reproducible round trip, explicit invalidation when an input changes, measured costs, and a recommendation to adopt, narrow, or stop. This is research, not approval for a registry service or marketplace.
Related: #249 owns declarative fragments; #252 owns portable artifact completeness and compatibility; #250 owns workspace materialization.