Repository navigation
Fix Windows startup and portable memory reporting - #37
Bortlesboat wants to merge 2 commits into
Conversation
Signed-off-by: Andrew Barnes <169967362+Bortlesboat@users.noreply.github.com>
9a6c8b3 to
ccc8aeb
Compare
Corrected the version comments in 580a5f6: the existing checkout pin is |
Signed-off-by: Bortlesboat <169967362+Bortlesboat@users.noreply.github.com>
00f56b5 to
580a5f6
Compare
|
Semgrep, zizmor and DCO passed, but CI 34256977533 ended without executing jobs after the approval wait. The Windows job is important here: importing resource breaks startup on that platform, and the patch uses current RSS from psutil instead of the platform-specific lifetime peak. The separate FastMCP constraint retains the project's supported 2.x import API. Could a maintainer approve/rerun CI for 580a5f6 so the Windows smoke job and retained Linux suite execute? The action-version comments have already been corrected in that head; no empty commit has been added to trigger the gate. |
Fixes #33.
Importing the queue on Windows currently fails because Python's
resourcemodule is unavailable. Usepsutilto report the current process's RSS in MiB and add a Windows CI smoke test that imports the queue and exercises memory reporting.The dependency range also keeps FastMCP on its supported 2.x API. A fresh install otherwise selects FastMCP 4, which no longer provides the
fastmcp.tools.toolimport used by this project.The memory helper now reports current RSS, as its docstring describes, instead of the lifetime peak returned by
ru_maxrss. Memory deltas can therefore become negative when memory is released. Task scheduling and cancellation are unchanged; this does not claim full Windows cancellation support.Validation:
ModuleNotFoundError: resourcewith the new regression test.COLUMNS=512sopsreturned complete arguments for the existing process-identification tests.Manually reviewed the four-file diff, including subprocess isolation, RSS conversion, and the retained Linux workflow.