Migrating to a unified gRPCServer package developed by both myself and s-celles (#72)

Author: csvance

.dockerignore

@@ -1,6 +1,5 @@
 # Build context is the repository root. Keep what the image needs (Project.toml, Manifest.toml,
-# packages/, the lib/ runtime submodule, docker/, config/) and drop everything else, especially the
-# large submodule .git dirs.
+# packages/, docker/, config/) and drop everything else.
 
 **/.git
 **/.gitignore

.github/workflows/_test.yml

@@ -1,5 +1,5 @@
 # Reusable workflow: run one CPU test suite. Called per-package by ci.yml so the pinned action
-# SHAs, submodule checkout, and depot caching live in one place. Not triggered on its own.
+# SHAs, checkout, and depot caching live in one place. Not triggered on its own.
 name: test suite
 
 on:
@@ -31,10 +31,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 60   # generous for the first run's Reactant CPU artifact download + compile
     steps:
-      # The packages source the vendored forks from lib/* via path, so submodules are required.
       - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
-        with:
-          submodules: recursive
       - uses: julia-actions/setup-julia@fa02766e078afaaf09b14210362cee14137e6a32 # v3.0.2
         with:
           version: ${{ inputs.julia-version }}

.github/workflows/ci.yml

@@ -41,15 +41,14 @@ jobs:
         with:
           # `common` (global files + Core) is folded into every suite, since Core is a dependency
           # of all of them and the global files affect everything. Each suite then adds its own
-          # package dir and the vendored lib/* it depends on (submodule bumps show as a change to
-          # the gitlink path `lib/X.jl`). Filters err toward over-inclusion so a real dependency
+          # package dir it depends on (a gRPCServer rev bump shows as a change to that package's
+          # Project.toml [sources]). Filters err toward over-inclusion so a real dependency
           # change can never produce a false green.
           filters: |
             common: &common
               - '.github/workflows/ci.yml'
               - '.github/workflows/_test.yml'
               - 'Project.toml'
-              - '.gitmodules'
               - 'packages/ReactantServerCore/**'
             core:
               - *common
@@ -59,13 +58,9 @@ jobs:
             gateway:
               - *common
               - 'packages/ReactantServerGateway/**'
-              - 'lib/gRPCServer.jl'
-              - 'lib/gRPCServer.jl/**'
             server:
               - *common
               - 'packages/ReactantServer/**'
-              - 'lib/gRPCServer.jl'
-              - 'lib/gRPCServer.jl/**'
             node:
               - *common
               - 'packages/ReactantServerNode/**'
@@ -74,8 +69,6 @@ jobs:
               - *common
               - 'packages/ReactantServerExport/**'
               - 'packages/ReactantServer/**'
-              - 'lib/gRPCServer.jl'
-              - 'lib/gRPCServer.jl/**'
             # The PyTorch round-trip provisions torch/jax/tensorflow via conda, so it runs only when
             # the PyTorch surface itself changes: the extension, its conda pins, its own test project,
             # or the shared test runner (which carries the torch-specific CPU-wheel and triton
@@ -166,8 +159,6 @@ jobs:
       actions: write   # julia-actions/cache prunes its old caches
     steps:
       - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
-        with:
-          submodules: recursive
       - uses: julia-actions/setup-julia@fa02766e078afaaf09b14210362cee14137e6a32 # v3.0.2
         with:
           version: '1.12'

.github/workflows/docs.yml

@@ -17,10 +17,6 @@ jobs:
       contents: write
     steps:
       - uses: actions/checkout@v4
-        with:
-          # The docs environment resolves ReactantServer's vendored gRPC
-          # backend from lib/gRPCServer.jl, a git submodule.
-          submodules: recursive
       - uses: julia-actions/setup-julia@v2
         with:
           version: '1.12'

.gitignore

@@ -21,4 +21,7 @@ Manifest-v*.toml
 # (large, ~230 MB) and renders the site into docs/build/.
 /docs/node_modules/
 # Subagent working artifacts.
-/.pi-subagents/
+/.pi-subagents/
+
+# Local (uncommitted) agent skills, installed from ~/Git/gRPCServer.jl/skills.
+/.pi/

.gitmodules

@@ -3,10 +3,10 @@
 Developer notes for working on ReactantServer.jl. For using the server, start with
 [Getting Started](docs/src/manual/getting_started.md).
 
-After cloning, populate the vendored submodules and instantiate the workspace:
+After cloning, instantiate the workspace (gRPCServer resolves from the `s-celles-merge` branch of
+github.com/csvance/gRPCServer.jl via the workspace `[sources]`):
 
 ```
-git submodule update --init --recursive
 julia --project=. -e 'using Pkg; Pkg.instantiate()'
 ```
 

CONTRIBUTING.md

@@ -17,7 +17,7 @@ JULIA      ?= julia
 
 all: help
 
-## image: build the reactantserver node image (needs lib/gRPCServer.jl + a local Manifest.toml; see docker/README.md)
+## image: build the reactantserver node image (needs a local Manifest.toml; gRPCServer resolves from GitHub, see docker/README.md)
 image:
 	$(ENGINE) build -f docker/Dockerfile -t $(NODE_IMAGE) .
 

Makefile

@@ -1,7 +1,7 @@
 # Workspace root for the ReactantServer monorepo. This environment is not itself a package;
 # it lists the member packages under [workspace] (they share the single root Manifest) and
-# declares the path [sources] for the vendored lib/* submodules and the inter-package
-# ReactantServerCore dependency, so every member resolves them consistently.
+# declares the [sources] for gRPCServer (the s-celles-merge branch of the upstream repo) and
+# the inter-package ReactantServerCore dependency, so every member resolves them consistently.
 
 # packages/ReactantServerExport is intentionally NOT listed here: it is the offline export
 # tooling, kept out of the shared workspace manifest so its Lux/PythonCall weakdeps never get
@@ -22,12 +22,13 @@ Reactant = "3c362404-f566-11ee-1572-e11a4b42c853"
 ReactantServerCore = "387a5f05-f7b4-4d57-bca9-cd3358e52bdc"
 YAML = "ddb6d928-2868-570f-bddf-ab3f9cf99eb6"
 gRPCClient = "aaca4a50-36af-4a1d-b878-4c443f2061ad"
-gRPCServer = "b7f0dee4-5e6a-42e7-93b0-9c098bad2317"
+gRPCServer = "608c6337-0d7d-447f-bb69-0f5674ee3959"
 
 [sources]
 ReactantServerCore = {path = "packages/ReactantServerCore"}
-gRPCServer = {path = "lib/gRPCServer.jl"}
-gRPCClient = {url = "https://github.com/JuliaIO/gRPCClient.jl", rev = "v1.1.0-rc2"}
+gRPCServer = {url = "https://github.com/csvance/gRPCServer.jl", rev = "s-celles-merge"}
 
 [compat]
+gRPCClient = "1.1.0"
+gRPCServer = "0.3"
 julia = "1.12"

Project.toml

@@ -60,7 +60,6 @@ The node runs natively (no containers): run the supervisor over a directory of m
 scales to all visible GPUs.
 
 ```
-git submodule update --init lib/gRPCServer.jl   # the one vendored fork the build needs
 REACTANT_GPU=cuda REACTANT_GPU_VERSION=13.1 julia --project=. -e 'using Pkg; Pkg.instantiate()'
 
 CUDA_VISIBLE_DEVICES=0,1,2,3 INFERENCE_SERVER_MODEL_DIRS=/path/to/bundles \
@@ -114,8 +113,9 @@ pulls in the heavy Reactant/XLA stack:
 
 Offline model export lives in `packages/ReactantServerExport` (a Reactant tracing frontend plus a
 PythonCall-triggered PyTorch extension); it is deliberately **not** a workspace member, so its
-Lux/PythonCall weakdeps stay out of the server images. The vendored `gRPCServer` fork is the only
-remaining git submodule under `lib/`.
+Lux/PythonCall weakdeps stay out of the server images. `gRPCServer` is sourced from the
+`s-celles-merge` branch of `github.com/csvance/gRPCServer.jl` through the workspace `[sources]`;
+there are no git submodules.
 
 ## Acknowledgments