UNSET_DOUBLE silently converted to 0.0 in CommissionReport fields

[msdavid]

Bug: UNSET_DOUBLE silently converted to 0.0 in CommissionReport fields

Version

ib_async 2.1.0

Description

In wrapper.py, the commissionReport handler converts IBKR's UNSET_DOUBLE sentinel value (1.7976931348623157e+308, meaning "not reported") to 0.0 for the realizedPNL and yield_ fields of CommissionReport.

This makes it impossible for consumers to distinguish between "IBKR did not report this value" and "the value is exactly zero" -- two semantically different states.

Related issues

• Realized PnL for single positions returning 1e308 erdewit/ib_insync#462 and reqPnlSingle returning ridiculously high value for realizedPnL erdewit/ib_insync#388 documented the inverse problem: UNSET_DOUBLE leaking through as a raw huge float in reqPnLSingle. Those were about PnL, not CommissionReport.
• This bug is about the overcorrection in the opposite direction: converting UNSET_DOUBLE to 0.0 instead of to None or math.nan.

Code reference

wrapper.py approximately lines 848-851 in version 2.1.0:

if report.realizedPNL >= UNSET_DOUBLE:
    report.realizedPNL = 0.0
if report.yield_ >= UNSET_DOUBLE:
    report.yield_ = 0.0

Expected behavior

UNSET_DOUBLE should be converted to math.nan (or None) rather than 0.0, preserving the semantic distinction between "not reported" and "zero."

Impact

Any system that tracks realized P&L or yield from commission reports cannot distinguish a genuine $0.00 realized P&L (e.g., a scratch trade) from a trade where IBKR simply has not yet reported the value. This can cause incorrect cost-basis calculations and misleading reporting.

[mattsta]

I think the logic here is we don't want realizedpnl to ever flash to be "bigass float" that IBKR uses for "not-yet-populated-values" because that really messes up your calculations (some bot seeing: "oh wow! profit is ALL THE MONEY IN THE UNIVERSE! LIQUIDATE PORTFOLIO AT MARKET RATES!")

If these defaults aren't picked up cleanly by the IBDefaults() system then that is where the actual default value overrides should be made (so users themselves can decide if they want default UNSET_DOUBLE or np.nan or float("-inf") or 0.0 or whatever fits their own system logic).