Bugs (fixed):
* The cash per point for many securities is wrong. For example, it is
5 for all BTC securities (which is wrong).
Cause: cpp override config that should be specific to cme is applied
to ALL securities.
Previously, the Subscribe impl only bootstrapped the first observer
(i.e., only in first call to Subscribe).
Bugs (fixed):
* The ladder crashes if there are working orders when the
ticks-per-row setting is changed.
Cause: The ladder viewmodel resubscribes to order updates each time
the ticks per row setting changes (as this determines the size of
the price buckets into which orders are grouped). The IObservable
"Bootstrap" extension fn doesn't properly clean up state leftover
from the previous subcsription when this happens.
This addresses a subtle inconsistency in the pnl window.
The old "start_pnl" and "day_pnl" values (from IPNLUnit) are computed
with respect to the mid market for both long and short positions, while
the new "unrealized" pnl values are computed with respect to the side of
the market needed to close the position (i.e., bid market for long pos,
ask for short). Consequently, the combined pnl columns weren't equal to
the sum of realized and unrealized pnl.
When looking at avg cost, you often really want to know the difference
between it and the current market. These columns provide that information
so you don't need to do the arithmetic in your head.
Bugs (fixed):
* Iceberg orders w/ fractional quantities can get into a state where
they send an infinite stream of orders w/ 0 qty
Note: I also changed the number of decimal places used when rounding
order quantities from 6 to 7 (the level of precision guarenteed by
IEEE 754).
When enabled, the checks fail regularly (usually more than once per
hour). Sometimes many of them happen in rapid succession before sparkd
crashes. I'm disabling them in the hopes that the crosses are benign,
and this will stop the crashes.
* EMNativeOrder
* OrderRouter - only log routing record result if there are errors
* OrderServiceServer
* OrderStore - also avoid calling pb::to_string twice per change (it's
expensive)
* ServiceBridge
Bugs (fixed):
* Sometimes mview shows that a security has working orders when it
doesn't. It appears to be rounding-related as the price and qty of
the phony working orders is NaN and 0, respectively.
Cause: Rounding error accumulates in the order summarizer. If it gets
bad enough, the working quantity (which is computed incrementally as
orders transition to/from a working state) may still be non-zero after
all orders become non-working.