Allow users to provide an explicit process count #78
Conversation
rtobar
force-pushed
the
explicit-process-count
branch
from
0266266 to
0b45b4f
Compare
I agree that this PR has merit on its own, but I'm unsure it would be useful for pre-commit hooks -- the reporting mechanism of Hence I think the better solution in this specific case is to enforce, or strongly reccomend, the serial mode of operation. A |
Instead of always defaulting to using as many processes are CPUs (or no sub-processes if checking less than 8 files), a new -j/--jobs flag allows users to customise how many sub-processes will be used to perform the checks. Defaults to the old behavior (i.e., os.cpu_count() processes), caps values on the lower end to 1 automatically, prevents subprocess spawning if requiring only one process, and aligns with the sphiinx-build semantics. Signed-off-by: Rodrigo Tobar <[email protected]>
rtobar
force-pushed
the
explicit-process-count
branch
from
0b45b4f to
da8c971
Compare
Co-authored-by: Adam Turner <[email protected]>
Co-authored-by: Adam Turner <[email protected]>
Co-authored-by: Adam Turner <[email protected]>
Thanks, I hadn't realised that was the policy for this project (different people gravitate towards different processes), but TBH I didn't check if there were contribution guidelines either. I've applied your suggestions now as-is since they seemed more than reasonable. |
|
"No force pushes" is policy for https://github.com/python/ repos, but we don't actually have a policy either way for this repo/org. And we have all merge options available here: 
|
sphinxlint/__main__.py
Outdated
| help="Run in parallle with N processes, defaults to 'auto', " | ||
| "which sets N to the number of logical CPUs." | ||
| "Values <= 1 are all considered 1.", |
There was a problem hiding this comment.
Is this the same behaviour as sphinx-build for N <= 1?
There was a problem hiding this comment.
Don't really know, but I thought it'd be the less error prone.
I agree with Adam that it doesn't sound so useful for pre-commit hooks. Can you suggest some of the other situations? |
Co-authored-by: Hugo van Kemenade <[email protected]>
Co-authored-by: Hugo van Kemenade <[email protected]>
Co-authored-by: Hugo van Kemenade <[email protected]>
Benchmarking and profiling, for one, which currently is limited to having to do it with up to 8 files, lest you spawn subprocesses. But giving full flexibility to users in general for situations where they might wnat to use a different solution between 1 and |
Useful for pre-commit hooks, as read in the discussion in #76, but probably in other situations too.
Instead of always defaulting to using as many processes are CPUs (or no sub-processes if checking less than 8 files), a new -p/--processes flag allows users to customise how many sub-processes will be used to perform the checks. Defaults to the old behavior (i.e., os.cpu_count() processes), caps values on the lower end to 1 automatically, and prevents subprocess spawning if requiring only one process.