Fix 30/360 US, 30/365 and Act/365 Canadian day counters diverging from QuantLib

[gaoflow]

Three day counters return different values from the QuantLib functions they port. Numbers below are from QuantLib 1.43 against QLNet develop.

case	QLNet	QuantLib 1.43
Thirty360(USA).dayCount(2007-02-28, 2007-03-31)	31	30
Thirty360(USA).dayCount(2008-02-29, 2008-08-31)	181	180
Thirty365().dayCount(2007-01-31, 2007-02-01)	0	1
Thirty365().dayCount(2007-01-01, 2007-01-31)	30	29
Actual365Fixed(Canadian), 2-year reference period	DivideByZeroException	domain error

UsImpl runs the last-of-February adjustments after the 31 -> 30 checks, so D1 is still 28/29 when dd2 == 31 && dd1 >= 30 is tested and D2 never collapses to 30. thirty360.cpp US_Impl::dayCount puts the February block first, under the comment // NOTE: the order of checks is important, and the class docstring here already describes that behaviour. Thirty365 is missing the two ISO 20022 31 -> 30 adjustments that thirty365.cpp applies, which is why a one-day interval can accrue zero. CA_Impl is missing QL_REQUIRE(frequency != 0, ...) from actual365fixed.cpp.

I compared every counter in Time/DayCounters against QuantLib 1.43 over 48,205 date pairs each (month ends, 31sts, Feb 28/29, leap and non-leap years). Only these three diverged: Bond Basis/ISMA, European/Eurobond, Italian, ISDA/German with and without a termination date, NASD, Act/360, Act/364, Act/366, Act/365.25, Act/365F Standard and NoLeap, Act/Act ISDA/AFB/ISMA, Simple, 1/1 and Business/252 all matched already, and all 27 match now.

Adds testThirty360_USA and testThirty360_NASD (NASD was already correct, just untested), extends testThirty365 with 31st endpoints, and adds the Canadian guard case. Expected values are QuantLib output. Suite: 533 passed, 0 failed.

[Copilot]

Pull request overview

This PR aligns QLNet’s 30/360 (USA), 30/365, and Actual/365 (Fixed) Canadian day counters with QuantLib’s behavior, addressing previously observed divergences and adding targeted regression tests to lock in the corrected results.

Changes:

• Fixes 30/360 (USA) end-of-February handling by applying the last-of-February adjustments before the 31 -> 30 checks (order-dependent).
• Implements the ISO 20022 31 -> 30 endpoint adjustments in 30/365 day counting.
• Adds a guard in Actual/365 (Fixed) Canadian to reject reference periods longer than one year (avoids zero-frequency division) and extends the test suite accordingly.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
tests/QLNet.Tests/T_DayCounters.cs	Adds/extends regression tests for 30/365 31st endpoints, 30/360 (USA), 30/360 (NASD), and the Canadian long reference-period guard.
src/QLNet/Time/DayCounters/Thirty365.cs	Adds ISO 20022 31 -> 30 adjustments to match QuantLib day-count behavior for 30/365.
src/QLNet/Time/DayCounters/Thirty360.cs	Reorders and refines 30/360 (USA) adjustments so end-of-February logic is applied in the correct order.
src/QLNet/Time/DayCounters/Actual365Fixed.cs	Adds a frequency != 0 requirement for the Canadian convention to prevent divide-by-zero for long reference periods.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[amaggiulli]

👍