pytest: *BROKEN* in test_xpay_limited_max_accepted_htlcs

[MSimeZ]

How to replicate

On CLN v26.04rc2, using MacOS M5 Tahoe 26.4, run uv run pytest -vvvv tests/test_xpay.py::test_xpay_limited_max_accepted_htlcs
The test is passed but with the following error:

ERROR tests/test_xpay.py::test_xpay_limited_max_accepted_htlcs - ValueError: 
Node errors:
 - lightningd-1: had BROKEN or That's weird messages
Global errors:

LOG:

['lightningd-1 2026-04-07T11:20:51.379Z **BROKEN** lightningd: Two equal candidates 0xbdb130cb0 and 0xbdb130cb0, channels 0xbdb2eb028 and 0xbdb2eb028, rr_number 4 and 4', 'lightningd-1 2026-04-07T11:20:51.379Z **BROKEN** lightningd: Two equal candidates 0xbdb130cb0 and 0xbdb130cb0, channels 0xbdb2eb028 and 0xbdb2eb028, rr_number 4 and 4']
Leaving base_dir /tmp/ltests-1x_uo8th intact, it still has test sub-directories with failure details: ['test_xpay_limited_max_accepted_htlcs_1']

[Lagrang3]

Weird. I can't reproduce this on linux.
@sangbida

[sangbida]

I was able to reproduce this on my mac. The test passes but there is an error. From asking asking Claude, I can gather that on macOS cmp_rr_number seems to get cases where the same route hint candidate are passed into that function. It seems like somehow the sorting on macOS allows for non unique values to be passed into that function. Just out of curiosity @Lagrang3 , what is the value of HAVE_QSORT_R_PRIVATE_LAST in config.vars?