The one Pylint check Ruff cannot take, and the case for pulling it out of CI

The reason R0801 is the check that stayed behind has less to do with Pylint's speed than with the question the rule asks. An unused-import rule can answer from a single file or syntax tree; duplicate detection has to compare source across a corpus, because the copy may sit anywhere else in the tree ^[6]. So the standard fix for a slow lint stage does not apply. Partitioning files among workers can change the clusters that get detected, which is why some real-world projects have already separated Pylint's duplicate-code checker into its own single-process CI job ^[7]. Once you have done that, you own a serial stage whose cost tracks the size of the repository, and you are installing a general-purpose linter to feed it.

Invert the maintainer's published ratios and you get the version worth quoting: on those three repositories Arid 2.0 finished in roughly 0.40 to 0.52 percent of the isolated Pylint 4.0.6 time ^[15]. The gap between the fastest and slowest of the three ratios is about 31 percent ^[16], narrow enough to read as implementation rather than repository quirk. He is explicit that this is not an argument that Pylint is 200x slower overall, since Pylint runs many checks Arid does not attempt ^[11], and he says flatly that "written in Rust" is not a benchmark, which is why the comparison is pinned and maintained ^[12].

What the ratios cannot tell you is whether this is your problem. The comparison is reported as multipliers, with no absolute wall clock for either tool ^[17]. Two hundred times faster on a stage that takes three seconds is invisible; on a stage that takes six minutes it is the reason your pipeline feels slow. The only number that settles it is your own timing of `pylint --disable=all --enable=duplicate-code .` ^[4], and the author agrees: if a dedicated R0801 pass is fast enough for your repository, you already have the solution ^[18].

The unresolved risk is agreement between the two implementations. Both have to decide what counts as similar, and Pylint exposes controls for ignoring comments, docstrings, imports and function signatures when it builds the representation it compares ^[5]. Two tools can make those decisions differently and report different clusters. The material supplied is about elapsed time, not about output matching on the same source tree ^[19], and it comes from the person who wrote the replacement.

That still leaves a narrow, defensible case. Duplicated logic multiplies the places that have to be found, changed and tested, a point Pylint's own documentation makes ^[13], so switching the check off is not the cheap option it looks like. If R0801 is the only reason Pylint is in the pipeline ^[1], the choice is between one flag and one fewer linter ^[8].