Reject target-only release cache sharing
This commit is contained in:
parent
b86491d82d
commit
41c91785c7
3 changed files with 42 additions and 23 deletions
|
|
@ -58,15 +58,21 @@ That package run exposed a release-forge ownership fault. One exact commit was
|
|||
split across manifest-specific Cargo target directories, so `epiphany-core`,
|
||||
`epiphany-openai-runtime`, and `epiphany-tool-mcp-runtime` rebuilt overlapping
|
||||
dependency graphs independently. The third manifest alone took 8m45s. The
|
||||
working cut gives the exact commit one shared target root while retaining
|
||||
separate lockfile validation, sequential builds, detached clean source, and
|
||||
byte-level release witnessing. All fourteen packaged-release tests pass.
|
||||
first cut gave the exact commit one shared target root while retaining separate
|
||||
lockfile validation, sequential builds, detached clean source, and byte-level
|
||||
release witnessing. All fourteen packaged-release tests passed, but the exact
|
||||
timed b86491d8 run disproved the real hypothesis: after the first owner finished,
|
||||
the second manifest immediately rebuilt common crates because the repository has
|
||||
no root workspace and its independent lock/resolution graph changes Cargo
|
||||
fingerprints. The shared-target mechanism is rejected and is being removed.
|
||||
|
||||
## Next action
|
||||
|
||||
Commit and push the release build-graph collapse, package that exact commit,
|
||||
and compare elapsed time against the three-root b8c76fd0 baseline. Publish the
|
||||
result only into a clean successor generation. Without overrides, prove:
|
||||
Finish recording the b86491d8 rejected-path timing, then replace the nine
|
||||
independent Epiphany package roots with one root workspace, one lockfile,
|
||||
root-owned patches, and one Cargo invocation for the required release binaries.
|
||||
Package and time that exact replacement. Publish the result only into a clean
|
||||
successor generation. Without overrides, prove:
|
||||
|
||||
1. Self derives proposal-bound Modeling and canonical Imagination.
|
||||
2. Self selects and launches canonical Imagination.
|
||||
|
|
|
|||
Reference in a new issue