I started a Wiki page here:

https://github.com/r-lib/lintr/wiki/Writing-robust,-performant-linters

My first thought was to add notes like this or an efficiency/"advanced" section to the custom_linters vignette, but I think a Wiki may be preferable here for reasons noted there.

The position() condition breaks your approach unfortunately.
Maybe we can avoid a top-level //* by constructing an XPath that has /*[position() > 1] as an element?

how about //OP[parent::expr[count(expr) >2]]

untested but hopefully it's clear what I mean -- test against unary operator on parent instead of self

Worth a try.

Not quite so simple (h/t our extensive test suite!), but the basic idea works. The complications for future reference were:

1. parent::expr is not always true, we see <expr_or_assign_or_help> for some EQ_ASSIGN cases
2. count(expr) > 1 is not enough for the EQ_SUB (assignment of function parameters) case, where a missing argument means there's no <expr> on the RHS, while the LHS doesn't get an <expr> because it's SYMBOL_SUB there, so the only <expr> is for the <expr><SYMBOL_FUNCTION_CALL>alist</SYMBOL_FUNCTION_CALL></expr>

Codecov Report

Merging #2025 (35a06ab) into main (c3c10bb) will decrease coverage by 0.01%.
The diff coverage is 100.00%.

❗ Current head 35a06ab differs from pull request most recent head 3ad2ec8. Consider uploading reports for the commit 3ad2ec8 to get more accurate results

@@            Coverage Diff             @@
##             main    #2025      +/-   ##
==========================================
- Coverage   98.55%   98.55%   -0.01%     
==========================================
  Files         113      113              
  Lines        4995     4994       -1     
==========================================
- Hits         4923     4922       -1     
  Misses         72       72              

Do we have a test for something like -1-1 to ensure the XPath refactoring scopes correctly and only finds one lint in the example?