Adopt src layout for the agent and server packages - #7
Draft
messa wants to merge 20 commits into
Draft
Conversation
Convert agent and server to standalone uv projects using the hatchling build backend, with metadata moved from setup.cfg into pyproject.toml ([project] table) and committed uv.lock files. Remove setup.py/setup.cfg. Add e2e_tests as a third uv project that pulls in agent and server via editable path sources, and move its test modules into e2e_tests/tests/ to match the agent/server layout. Add per-component Makefiles (check = lint + test) and make the root Makefile delegate via $(MAKE) -C; add a server `image` target that builds the container with podman and smoke-checks the entrypoint. Keep run_e2e_tests as a backwards-compatible alias. Rewrite the Dockerfile and the python-tests/e2e-tests workflows to use uv (astral-sh/setup-uv, uv sync --frozen). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Bundle the MIT LICENSE into the agent and server wheels by symlinking the root LICENSE into each component (hatchling picks it up via PEP 639). - server/Dockerfile: bump base image to python:3.14 and install dependencies in their own cached layer (uv sync --no-install-project) before copying the source, so source changes no longer invalidate the dependency install. - CLAUDE.md: add an "Agent workflow" note to review changes via a background subagent. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Martin Hanzík <martin@hanzik.com>
The debug log in test_rotate_log_file referenced expected_dst_second_file, which is only defined in test_new_log_file_gets_copied. Use expected_dst_file instead: the destination path actually in scope, and the same variable the analogous first wait-loop already logs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a lint target to e2e_tests/Makefile, identical to the agent and server ones, and make it part of the default check (check: lint test). Recurse into e2e_tests from the root lint target so the aggregate lint now covers all three components, matching the existing check and test aggregates. Previously the e2e test code was the only Python in the repo not covered by flake8. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Point the "project readme file" link at leadhub-code/logline instead of the old messa/logline location, matching the homepage URLs already corrected in the pyproject files. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Switch both components from hatchling to uv's own uv_build backend: - requires = ["uv_build>=0.11,<0.12"], build-backend = "uv_build" - [tool.uv.build-backend] module-root = "" for the flat layout, since the package dir lives in the project root (not under src/) and uv_build defaults module-root to "src" - license-files = ["LICENSE"] because uv_build, unlike hatchling, does not auto-glob license files Replace the agent/server LICENSE symlinks with real copies of the repo-root MIT license. uv_build validates license-files while building the project, and the upward ../LICENSE symlink escaped the build root, which broke both the server Docker build (context is server/) and isolated sdist builds. Real copies fix every build path (wheel, sdist, Docker). Copy LICENSE into the server image so uv_build's metadata validation passes during uv sync. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace flake8 with ruff across all three components (agent, server, e2e_tests). The ruff configuration lives in each component's pyproject.toml: line length 150, rule sets E/F/W/I, and isort with two blank lines after the import block. force-sort-within-sections keeps the existing "sort by module name" import style, so reordering churn is small. Update the Makefiles and GitHub Actions workflows to lint with `uvx ruff check .`. The e2e_tests component is now linted as well (previously it had no lint step), which surfaced a latent bug: a debug log in test_rotate_log_file referenced an undefined name (expected_dst_second_file instead of expected_dst_file). Apply ruff's autofixes (import sorting, unused imports, redundant f-string prefix, unused exception variable) and fix the remaining findings by hand: - re-export the entry points via __all__ in the package __init__ modules - reference the imported module in the smoke tests - use `is not None` / `not ...` instead of `!= None` / `== False` - fix the undefined name in the e2e rotate test Document the Python style conventions (line length, two blank lines after imports, preferring `from X import Y`) in CLAUDE.md. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Hoist function-local imports to module level where there is no reason to defer them: - logging name imports in setup_logging/setup_log_file (agent and server) - the logging import in the e2e conftest - the lazy ssl imports in the TLS branches (agent client, server) - the lazy yaml imports in the configuration loaders (agent and server) Also switch hashlib usage to `from hashlib import sha1` to match the `from X import Y` convention. Conditional compatibility shims (try/except ImportError in asyncio_helpers and util), the optional zstandard/zstd fallbacks, and the smoke-test imports are intentionally left local. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The minimum supported Python version is 3.9, where asyncio.run, asyncio.create_task and asyncio.get_running_loop (all 3.7+) and asyncio.to_thread (3.9+) are always available, so the fallback shims are dead code. Delete the agent's asyncio_helpers module and import run/create_task/ to_thread directly from asyncio. Drop the to_thread shim from the server's util module the same way. The zstandard/zstd fallback in util is intentionally kept: it selects between two optional libraries at runtime and is not a Python-version shim. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
In the agent's obfuscate_secrets(), re.ASCII was passed as the 4th positional argument to re.sub(), which is `count`, not `flags`. That made the flag a no-op (256 used as the replacement count) and triggered a DeprecationWarning on Python 3.13+. Pass it as `flags=re.ASCII`. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
datetime.utcnow() returns a naive datetime and is deprecated since Python
3.12. Use datetime.now(timezone.utc) instead. The rendered string is
unchanged: the strftime format ('%Y%m%dT%H%M%SZ') has no %z/%Z field and
the trailing Z is a literal, so the rotated-file suffix stays identical.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The TLS certificate path from the YAML config was resolved relative to the config file's directory, but the key path was used verbatim (i.e. relative to the current working directory). Resolve the key the same way as the certificate and the key password file for consistent behaviour. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The agent ran its event loop with debug=True, which was left over from development. The asyncio debug mode adds runtime overhead (slow-callback logging, extra thread-safety checks, ...) that is undesirable in production. Run the loop with the default settings. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A freshly connected client that never sends its initial command kept a server connection slot (and file descriptor) occupied indefinitely, which could be abused in a slowloris-style fashion. Wrap the receipt of the initial command in a timeout; established connections are still allowed to stay idle for as long as needed, since the agent only sends data when the watched log grows. On the agent side, open_connection() had no timeout either, so a silent or unreachable server would make the connection attempt hang forever. Wrap it in a connect timeout so the follow loop can retry. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The handler used assert statements to validate data coming from the client (required header fields, the data payload type and the expected write offset). Assertions are stripped when Python runs with -O, which would silently disable this validation, and they also produce confusing AssertionErrors. Replace them with explicit checks that raise ProtocolError, and read the offset via .get() so a missing field is reported as a protocol error rather than a KeyError. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a "Pull requests" section that prescribes a single blockquote format for flagging a stacked PR (one based on another PR's branch rather than main), so these notes look the same across all PRs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The TLS certificate and key paths were Path objects when read from the YAML config but plain strings when passed via CLI args. Wrap the CLI values in Path() so tls_cert_file and tls_key_file have a consistent type regardless of source, matching how the agent already handles its cert path. Purely cosmetic: ssl.load_cert_chain accepts both. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move the logline_agent and logline_server packages under a src/ directory and point the uv build backend at it (module-root = "src" instead of ""). This is the conventional src layout: the project root is no longer importable, so tests and tools exercise the installed package rather than accidentally importing straight from the working tree. The server Dockerfile now copies src/ instead of the old top-level package directory. e2e_tests is left unchanged since it ships no package of its own. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
martinhanzik
force-pushed
the
202606_messa_minor_updates
branch
from
June 3, 2026 13:01
b880dba to
11dc1c1
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Move the
logline_agentandlogline_serverpackages under asrc/directory (the conventional src layout), so tests and tools run against the installed package instead of accidentally importing straight from the working tree.Changes
agent/logline_agent/→agent/src/logline_agent/andserver/logline_server/→server/src/logline_server/(viagit mv, so the file history is preserved).module-root = "src"(was"") in bothpyproject.tomlfiles.DockerfiletoCOPY src ./src.e2e_testsis left unchanged — it ships no package of its own.No changes were needed to the Makefiles, the CI workflows, the script entry points, or
uv.lock.Testing
make checkis green for all three components: agent (2), server (12), e2e (7 — including the real TLS round-trip and the--helpsmoke tests). The e2e suite installs both packages and runs their CLIs, so it exercises the build/install from the new layout end to end.make -C server image— podman build plus the entrypoint--helpsmoke check passes.🤖 Generated with Claude Code