feat: make axum host configurable via manifest and env vars

[ChristianPavilonis]

Summary

• Add host and port fields to ManifestAdapterDefinition so adapters can declare bind addresses in edgezero.toml under [adapters.<name>.adapter]
• Parse adapter.host from axum.toml with IP validation, defaulting to 127.0.0.1
• run_app() now reads host/port from manifest config with EDGEZERO_HOST / EDGEZERO_PORT env var overrides
• CLI fallback dev server and run_cargo subprocess both respect the same env vars
• Invalid values produce warnings instead of silently falling back to defaults

Configuration Precedence

1. EDGEZERO_HOST / EDGEZERO_PORT environment variables (highest)
2. Manifest value (edgezero.toml or axum.toml)
3. Default: 127.0.0.1:8787

Usage

# edgezero.toml
[adapters.axum.adapter]
host = "0.0.0.0"
port = 3000

# axum.toml
[adapter]
host = "0.0.0.0"
port = 3000

# Or via environment
EDGEZERO_HOST=0.0.0.0 EDGEZERO_PORT=3000 cargo run -p my-adapter

Files Changed

File	Change
crates/edgezero-core/src/manifest.rs	Add host + port to ManifestAdapterDefinition + tests
crates/edgezero-adapter-axum/src/cli.rs	Parse host from axum.toml, propagate via env vars to subprocess
crates/edgezero-adapter-axum/src/dev_server.rs	resolve_addr() with manifest + env var precedence + warnings
crates/edgezero-adapter-axum/src/lib.rs	Export resolve_addr
crates/edgezero-cli/src/dev_server.rs	Fallback path checks env vars with warnings

[aram356]

PR Review

Summary

Well-structured feature that adds configurable host/port binding for the Axum adapter with a clean precedence model (env vars → manifest → defaults). CI gates all pass. One behavioral inconsistency around port 0 validation needs fixing.

😃 Praise

• Clean configuration precedence design with good separation of concerns between CLI, adapter, and manifest layers
• Excellent test isolation: using resolve_addr_from_parts with explicit parameters avoids env var races in parallel test runners
• Thorough test coverage for the happy paths in all three layers (manifest, CLI, adapter)

Findings

Blocking

• 🔧 Port 0 inconsistency: cli.rs rejects port 0 from axum.toml but resolve_addr_from_parts and resolve_dev_addr accept EDGEZERO_PORT=0 and manifest port 0, causing random OS port binding (dev_server.rs:188)

Non-blocking

• 🤔 resolve_addr public API surface: exported publicly but only used internally by run_app — consider pub(crate) or documenting the external use case (lib.rs:22)
• 🤔 host field as String, not IpAddr: accepts any string at parse time, IP validation deferred to adapter runtime (manifest.rs:285)
• 🤔 Duplicated resolution logic: resolve_dev_addr in edgezero-cli duplicates resolve_addr_from_parts logic with different logging (cli/dev_server.rs:88)
• 🤔 Inconsistent error strictness: invalid host in axum.toml → hard error; invalid host in edgezero.toml → warning + fallback. Reasonable per context but worth documenting.
• ⛏ Incidental #[derive(Debug)] on AxumProject — harmless (cli.rs:143)

📌 Out of Scope

• Missing tests for invalid env var fallback paths (e.g. EDGEZERO_HOST="not-an-ip", EDGEZERO_PORT="abc") — good to add but not blocking
• Two config sources (axum.toml vs edgezero.toml) for the same adapter is pre-existing complexity worth documenting

CI Status

• fmt: ✅ PASS
• clippy: ✅ PASS
• tests: ✅ PASS (322 passed)

[aram356]

PR Review

Summary

Well-structured feature that adds configurable host/port binding with clean precedence logic (env > manifest > defaults). The centralised addr.rs module and backward-compatible manifest changes are solid. Main concern is a validation inconsistency between the axum.toml path and the shared resolve_bind_addr path.

😃 Praise

• Extracting resolve_bind_addr as a pure function with comprehensive tests (including IPv6, port 0, partial overrides) is excellent — avoids env var race conditions in parallel tests
• Backward-compatible manifest extension using Option + #[serde(default)] is the right pattern
• Port 0 rejection in addr.rs prevents confusing random-port behavior

Findings

Blocking

• ❓ Two separate validation paths: read_axum_project (axum.toml) hard-errors on invalid IP, while resolve_bind_addr warns and falls back. Same user mistake → different behavior depending on config source. See inline comment on cli.rs.

Non-blocking

• 🤔 pub mod addr API surface: resolve_bind_addr is now part of edgezero_core's public API. Confirm this is intentional vs pub(crate). See inline on lib.rs.
• 🤔 Silent warning drop in CLI fallback: log::warn! in resolve_bind_addr may fire before the logger is initialised in the CLI fallback path, silently swallowing invalid-value warnings. See inline on addr.rs.
• 🏕 Magic 8787 in cli.rs: The default port 8787 is now defined as DEFAULT_PORT in edgezero_core::addr but remains hardcoded in read_axum_project (None => 8787). Consider using the constant or a local const to keep values in sync.
• 🌱 Late host validation: Manifest accepts arbitrary strings for host, validated only at bind time. Defensible for WASM compat but worth a doc comment. See inline on manifest.rs.

CI Status

• fmt: ✅ PASS
• clippy: ✅ PASS
• tests: ✅ PASS (322 tests)

[aram356]

PR Review

Summary

This PR adds configurable host/port for the axum adapter via edgezero.toml, axum.toml, and EDGEZERO_HOST/EDGEZERO_PORT env vars. The shared resolve_bind_addr in core with clear precedence rules is well-factored. However, there is a blocking test failure caused by unsafe env::set_var leaking across parallel tests.

Findings

Blocking

• 🔧 Test failure: env::set_var causes cross-test contamination — read_axum_project_env_overrides_config leaks EDGEZERO_PORT=9999 into parallel tests, causing read_axum_project_accepts_min_valid_port to fail. All tests calling read_axum_project directly are vulnerable. See inline comment for fix. (cli.rs:698-705)

Non-blocking

• 🤔 addr module is unconditionally public in core — pub mod addr is available on WASM targets where bind-address resolution is meaningless. Consider gating behind #[cfg(not(target_arch = "wasm32"))] or leaving as-is if you want the constants available everywhere (lib.rs:3)
• ⛏ Naming inconsistency across resolution functions — resolve_bind_addr (core), resolve_addr (dev_server), resolve_dev_addr (cli/dev_server) — consider aligning names or documenting why each wrapper differs
• 🌱 host could accept hostnames, not just IPs — host = "localhost" would fail with a confusing error; consider improving the error message to say "must be an IP address, not a hostname" (addr.rs:37)
• 📝 Double env-var resolution is by design — the CLI resolves addr then re-exports via env vars to the subprocess, which resolves again; this is safe because the parent exports final values, but a brief comment in run_cargo would help future readers

📌 Out of Scope

• Existing unsafe env::set_var usage elsewhere in the codebase (test_utils.rs, main.rs, adapter.rs, generator.rs) has the same thread-safety risk. A project-wide migration would be valuable but is beyond this PR scope.

CI Status

• fmt: PASS
• clippy: PASS
• tests: FAIL — cli::tests::read_axum_project_accepts_min_valid_port (env var leakage from parallel test)