Person/codekiln/GitHub/logseq-gardener/Analysis/Fable/26/09/12/0933 ET Where my plan agrees with Astra's proposal and where it departs

• Where my plan agrees with Astra's proposal and where it departs

  • Person/codekiln/GitHub/logseq-gardener/Project/Proposal/Astra and Person/codekiln/GitHub/logseq-gardener/Project/Proposal/Fable describe the same engine and most of the same clients. This page lists what I would take from Astra's proposal as written and the places where my plan makes a different call, so that each one can be decided on its own.

  • Taken as written

    • garden impact, under the name garden diff, as the first feature after reading and navigation, for the reasons on Person/codekiln/GitHub/logseq-gardener/Analysis/Fable/26/09/12/0933 ET Call the impact command garden diff and run it where the link checker runs.
    • The overlay of unsaved editor buffers over the saved graph, and separate views per checkout.
    • Typed resolve results that separate a match, an ambiguous name, no match, and an index still warming.
    • The publisher's feature-support report, and its rule that an unsupported query looks different from an empty one.
    • Editing commands that keep untouched bytes, check the source revision before writing, and record enough to finish or undo a multi-file rename.
    • A persistent index that short-lived commands reuse, which Person/codekiln/GitHub/logseq-gardener/Analysis/Fable/26/09/12/0933 ET Agents calling the CLI justify a persistent cache on a small garden now agrees with.

  • Where I would decide differently

    • Publishing comes earlier

      • Astra sequences publishing after garden impact and the editing commands. I would run Tine's static export against the encode garden in the first week, because the published site's problem is concrete today and the trial costs an hour, and then schedule garden export after the reference graph and before mutations.

    • Concurrency starts in git

      • Astra proposes a shared editing session in which clients exchange operations with explicit block identities. Concurrent edits in this repository meet at merge time in git, so I would build the block-aware merge driver first, per Person/codekiln/GitHub/logseq-gardener/Analysis/Fable/26/09/12/0658 ET A Logseq-aware git merge driver before a CRDT, and open the session experiment when a merge the driver cannot settle shows up in practice.

    • Pick the tool's name before anyone installs the binary

      • Astra's proposal keeps garden throughout. The collisions in Person/codekiln/GitHub/logseq-gardener/Analysis/Fable/26/09/12/0658 ET One tool carries four names and two of them collide stand, and renaming gets expensive once someone has installed the binary.

    • Sharing parsed text across worktrees waits

      • Astra keys shared parse artifacts by content hash. I would give each checkout its own cache directory and add sharing when a worktree's first command is measured slow, per Person/codekiln/GitHub/logseq-gardener/Analysis/Fable/26/09/12/0933 ET Give each checkout its own cache and share parsed text later.

    • Name the tool that runs each parser test

      • Astra evaluates parsers against syntax, the resulting graph, and text preservation. My plan names the tool that runs each test: lsdoc's differential check, a dump from Logseq's own graph-parser under nbb-logseq, and a byte-identical round trip through tine-core's serializer, per Person/codekiln/GitHub/logseq-gardener/Analysis/Fable/26/09/12/0933 ET Use Logseq's own graph-parser as the oracle and require a byte-identical round trip.