Evaluating expression of large objects from the debug console is extremely slow

I'm seeing an issue that apparently hasn't been reported yet. I have a fairly large dictionary in memory and when I want to get its length from the debug console it takes tens of seconds. Obviously, getting the repr of the dict takes that much time but I only want to get its length.

Code to reproduce:

from unittest import TestCase
from collections import defaultdict
import numpy as np
from types import SimpleNamespace

class MyTest(TestCase):
    def test_large_defaultdict(self):
        d = defaultdict(lambda: defaultdict(list))
        for k in 'abcdef':
            for l in 'ghijk':
                for i in range(10000):
                    d[k][l].append(SimpleNamespace(
                        key = k,
                        cycle = np.arange(20),
                        value = np.zeros(20)
                    ))

        pass

Set breakpoint on pass and try to evaluate len(d) from the debug console after the breakpoint is hit. This warning is printed in the console:

pydevd warning: Computing repr of d (defaultdict) was slow (took 42.36s)
Customize report timeout by setting the `PYDEVD_WARN_SLOW_RESOLVE_TIMEOUT` environment variable to a higher timeout (default is: 0.5s)

Yes, it is slow but why is repr called to print the number of items?

Start with the test.py reproduction and the debug console evaluation of len(d), then trace the debugger's value-resolution path that emits the slow repr warning. Compare evaluation of len(d) with direct object inspection and identify where representation is unnecessarily computed. Done means large-object length evaluation no longer incurs the repr delay and the behavior is covered by a regression test.