DAO Proposals & Community

View active proposals, submit new ideas, and connect with the SWARMS community.

Make WORKSPACE_DIR handling resilient, and cover it with tests

about 2 hours ago1 comments

tests

structs

telemetry

utils

#2023 made `WORKSPACE_DIR` survive `import swarms`. Now that the value actually reaches the code that uses it, the two places acting on it have to cope with a value they don't control — a user-supplied path can be unwritable, or can name a file. ## `bootup.py` **Stops duplicating the defaulting logic.** `_prepare_workspace()` now defers to `workspace_manager.ensure_workspace_env()` instead of keeping a second copy. The two copies had already drifted: `ensure_workspace_env` clears `get_workspace_dir`'s `lru_cache` when it invents a default, and bootup's copy didn't — so a cached `None` could stick for the life of the process. **Guards `mkdir` on the caller's path.** It ran inside bootup's `try` block, which logs and **re-raises**, so a `WORKSPACE_DIR` that can't be created took down `import swarms` entirely. `exist_ok=True` doesn't help: it suppresses "already exists" only when the existing entry is a *directory*, so a `WORKSPACE_DIR` naming a file raised regardless. On failure it warns and falls back to `{cwd}/agent_workspace`. **The `workspace_manager` import is inside the function deliberately.** `swarms.utils.workspace_manager` pulls in `swarms/utils/__init__`, which calls `initialize_logger` at import. At module scope that runs *before* `bootup()`, so the logger would resolve its directory from the unprepared value. I hit this while writing the tests; the comment is there so it doesn't get "cleaned up" back to the top. ## `loguru_logger.py` **The same unguarded `makedirs` was here too — I introduced it in #2022**, one file away from the identical bug. It now degrades to console-only logging with a warning instead of raising. **Handlers are rebuilt when the log directory moves.** The `_CONFIGURED` guard from #2022 stops the 28 modules tearing down each other's handlers, but it also latched the *first* directory seen — and the first call happens before bootup settles `WORKSPACE_DIR`. After a fallback the logger stayed aimed at the unusable path and **wrote nothing anywhere**. Tracking `_CONFIGURED_DIR` fixes that; a changed directory is the one case where reconfiguring is correct. ## Tests `swarms/utils/loguru_logger.py` had **no coverage at all** before this. | file | tests | fail on master | | --- | --- | --- | | `tests/telemetry/test_bootup.py` | 5 | 2 | | `tests/utils/test_loguru_logger.py` | 13 | 2 | The four that fail on master are exactly the behaviours fixed here. The rest are regression guards for what #2022 and #2023 established: `log_folder` is a label rather than a path, per-module routing, non-`swarms` records ignored, size rotation, write errors swallowed, and `WORKSPACE_DIR` surviving import. `test_bootup.py` uses subprocesses because `WORKSPACE_DIR` is read during import, which happens once per interpreter — monkeypatching in-process would prove nothing. `test_loguru_logger.py` resets the module's process-wide state (`_CONFIGURED`, `_CONFIGURED_DIR`) and restores loguru's handler stack between tests; without that they leak into each other and into neighbouring files. I verified it perturbs nothing: `tests/utils/` + `tests/telemetry/` show the same 24 failures with and without the new file. ## Carried along, not authored here Flagged so a reviewer can scope them separately: | change | note | | --- | --- | | `pyproject.toml` | version `14.0.0` → `14.0.1` | | `auto_agent_builder.py` | adds `batch_run` delegating to `execution_utils.batched_run` | | `structs/concat.py` **deleted** | verified: nothing imports it, and it was never exported from `structs/__init__` | ## Verification - full suite vs master, same test files both sides: **no failures introduced**; the only difference is the four tests that now pass - `import swarms` works with `WORKSPACE_DIR` set, unset, and pointing at an unusable path - an unusable path falls back and logs still land in the fallback (previously: nothing written) - `black --line-length 70 --check` and `ruff check` clean across 277 files

Proposed by kyegomez

DAO Proposals & Community

Discord Community

GitHub Repository

Twitter

Documentation