Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.43%. Comparing base (3ca01ec) to head (a71e394).
Report is 1 commits behind head on master.

@@           Coverage Diff           @@
##           master   #15818   +/-   ##
=======================================
  Coverage   98.43%   98.43%           
=======================================
  Files         163      163           
  Lines       21294    21296    +2     
=======================================
+ Hits        20960    20962    +2     
  Misses        334      334           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Code size report:

   bare-arm:    +0 +0.000% 
minimal x86:    +0 +0.000% 
   unix x64:   +40 +0.005% standard[incl +32(data)]
      stm32:   +16 +0.004% PYBV10
     mimxrt:    +8 +0.002% TEENSY40
        rp2:   +16 +0.002% RPI_PICO_W
       samd:    +8 +0.003% ADAFRUIT_ITSYBITSY_M4_EXPRESS

I've modified the set_intersect_int() and avoided linear search too. Now this PR should not reduce the performance of the existing hash-based approach. I also tried to minimize code size as much as possible.

I've also set minimum accepted arg count to 1 in order to make set.intersection() compatibe with CPython in that case.

If this case (having one arg) is the only problem, I can solve it easily without significant code size increase.

It seems a little bit more code is needed to avoid unnecessary copy when n_args == 2 than I expected, which may increase code size. So I think it's better to wait for @dpgeorge review.

It seems a little bit more code is needed to avoid unnecessary copy when n_args == 2 than I expected, which may increase code size

I think it's worth improving things so it doesn't need the unnecessary copy. The new implementation should match the existing implementation (in terms of memory usage) for update == false and n_args == 2, ie something like s.intersect({1, 2}).

And fixing that will actual improve the general case for update == false and n_args >= 2, because it won't need to copy the input set at all.

Please try to do that, and we'll see how much firmware size changes. It shouldn't be too bad.