Skip to content

build(gateway): pin MCP 5b34b26 and PrismPM main for LexLean release/v0.4 (#7) - #103

Merged
humuhumu33 merged 6 commits into
mainfrom
claude/gateway-pins
Nov 13, 2026
Merged

humuhumu33 merged 6 commits into
mainfrom
claude/gateway-pins

Conversation

@humuhumu33

Copy link
Copy Markdown
Contributor

Moves the gateway's two upstream pins off the PrismPM backport now that UOR-Foundation/mcp#49 has merged (MCP on LexLean release/v0.4 at 5dccf9e5b and on PrismPM main).

  • .github/mcp.rev: MCP c53816e to 5b34b26.
  • .github/prismpm.rev: the backport (2e79f509) to PrismPM main d650d940 (0.4.0). The generated host code and the two copied Foundation files are unchanged between the two, so no source change follows from it.
  • gateway/lexlean.lock relocked: only the compiler value changes, mcp.core is the same tree. gateway/mcp-identities.sh passes against the pin.
  • lexlean = "0.4.0" in the three gateway manifests; gateway/registry.expected records the new package (d172aee0…, reproduced with the registry script's own method).
  • The pinned LexLean has 303 negative fixtures against the register's 298. Two more run check and join F4 (254). Three run production, which no gateway tool exposes, so they are listed in OUT_OF_SCOPE_FIXTURES and compared with nothing. TOOLS.md describes the new pin; at it the fixtures' locks already record the compiler's identity, and 8 of the 298 expected trees differ (290 before).
  • The three fixture projects that record the compiler are relocked and the identity provision.rs asserts follows.

Not done, and why this is a draft: prismpm build of the gateway has to run to produce gateway/generated (the export and its request pins), gateway/Cargo.lock needs the generated host crates, and the recorded vectors may change. It could not finish on the development machine (the host killed the WSL VM three times for virtual memory), so the CI job builds it and publishes the packages as the gateway-build artifact; the next commit takes them from there.

Refs #7.

🤖 Generated with Claude Code

humuhumu33 and others added 6 commits November 13, 2026 10:43
…v0.4

MCP#49 moved MCP to the head of LexLean release/v0.4 (5dccf9e5b) and to
PrismPM main, so the gateway no longer needs the PrismPM backport.

Move .github/mcp.rev to MCP 5b34b26 and .github/prismpm.rev to PrismPM
d650d940 (0.4.0). The gateway is compiled by the LexLean MCP names, so
relock gateway/lexlean.lock (only the compiler value changes; mcp.core is
unchanged), require lexlean 0.4.0 in the three manifests and record its
package in registry.expected. The three fixture projects that record the
compiler are relocked and the identity the provision test asserts follows.

The pin has 303 negative fixtures against the register 298: two more run
check and join F4 (254), three run production, which no tool exposes, and
are listed as out of scope. TOOLS.md describes the new pin.

Not yet done, because it needs prismpm build of the gateway, which could
not finish locally: gateway/generated (the export and its request pins),
gateway/Cargo.lock and the recorded vectors. CI builds the packages and
publishes them as the gateway-build artifact.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The gateway CI build compares the export of prismpm build with
gateway/generated/lexlean_mcp_gateway_core and failed on the stale one. Take
the export from that build: the CI log holds the diff of the build against the
committed package, which reversed onto the committed files gives the new
package. Its digests are the ones package.json and provenance.json record
(src/lib.rs 72788e33, Cargo.toml unchanged, semantic ID ecba9a6d, compiler
635a25ac), and the target digest of request.json follows the compiler.

CI repeats the comparison; this is not a local build.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
gateway/Cargo.lock changes only lexlean: 0.3.0 to 0.4.0 and the checksum of
the package gateway/registry.sh makes from the LexLean pin (d172aee0, the one
registry.expected records). The workspace builds against it with the
generated host crates of the CI build.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The fixture project records a lock of the compiler before the pin move, so
lexlean check answered LLC0102 (exit 2) before it reached the atom the test
is about. Relocked, it gives the one diagnostic TL-07 expects, LLL1004, with
exit 1. Only the compiler value of the lock changes.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
At the head of LexLean release/v0.4 the CLI reproduces expected/build for all
18 examples (the fixture locks record the identity of the compiler built from
the pin), so TL-06 found none off the tree where the list named 12. The list is
empty and the test still fails if an example joins it. TOOLS.md and QUALITY.md
now say AC-04 is shown at this pin.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@humuhumu33
humuhumu33 marked this pull request as ready for review November 13, 2026 18:51
@humuhumu33
humuhumu33 merged commit ece9ef3 into main Nov 13, 2026
11 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant