Optimize Blockstore

[lidel]

In spare time we should explore if there is a benefit in replacing IndexedDB-backed block storage with blockstore-opfs (in theory avoids structured clone overhead but the diff could be negligible)

[achingbrain]

The blockstore is only active during the initial content load by the service worker, after that all responses are served from the browser's response cache, TBH we barely need a blockstore at all and can probably swap it out for a BlackHoleBlockstore that doesn't store anything.

[lidel]

Ack, makes sense: we definitely want to avoid block storage creep on IPNS/DNSLink sites.

However, "blackholing" feels like an overcorrection. We still need a functional (but ephemeral) cache for DAG-heavy sites like Wikipedia. Without it, we’d be re-fetching parent HAMT blocks for every single file on a page, which is a big performance hit.

Idea: Do we have / could we add blockstore constructor option to switch from localStorage to sessionStorage?

This would let us use a regular blockstore for the duration of the visit, but ensure everything is automatically evicted as soon as the tab is closed.

[lidel]

Triage note:

• this only impacts mutable /ipns/ namespace (IPNS Names & DNSLinks)
• for now we assume browser-based eviction is good enough
• we can revisit once more pressing things are addressed
  • test case: user browsing /ipns/en.wikipedia-on-ipfs.org/ could end up with a lot of blocks + loading every article benefits from /ipns/en.wikipedia-on-ipfs.org/wiki/ HAMT-intermediate blocks