Fixed Point Money Arithmetic

double cannot represent most decimal fractions exactly, because it stores binary fractions, not decimal ones. 0.1 in a double is actually 0.1000000000000000055511151231257827021181583404541015625. A single addition never shows it. Ten thousand additions of 0.1 to a running total measurably do not equal 1000.0 - the demo prints the exact error alongside the exact result so the drift is not just asserted, it is shown. This matters for anything that accumulates over a long run rather than computing once: realised profit tallied across months of closed trades, a running equity total updated on every tick, a fixed-ratio or optimal-f tracker that recomputes its base from cumulative results. Each individual addition's error is tiny. The point is that they do not cancel out - they compound, in the same direction, for as long as the program runs. CDecimal fixes this by never storing the value as a double at all. Internally it is a 64-bit integer scaled by one hundred million - eight decimal places, comfortably more precision than any broker's price or lot step - so addition and subtraction are exact integer operations with no rounding step whatsoever. Two CDecimal values added a million times produce a result with zero accumulated error, not just a smaller one. Multiplication and division are the honest exception. Scaling a decimal by an arbitrary factor - lots, a percentage, anything that is not itself a fixed- point value - genuinely requires a floating point step, because the factor is not fixed-point. MultiplyBy and DivideBy do this openly, in one place, for one call. That is a fundamentally different situation from thousands of additions each losing a fraction of a bit: one rounding step does not compound the way a running total does. Division by zero has no correct answer, and neither MQL4 nor MQL5 has exceptions to signal one with. Rather than return an infinity or a silently wrapped garbage integer, DivideBy returns the original value unchanged when the divisor is effectively zero - a visible, debuggable no-op instead of a value that quietly stopped meaning anything. The class body is pure integer and double arithmetic - MathRound, MathAbs and DoubleToString all behave identically on both platforms, so the MQL5 header is the same class, unchanged, with only the platform-neutral #property lines adjusted. Notes and limitations - Eight decimal places of precision. That comfortably covers five-digit forex quotes and fractional lot sizes, but this is not an arbitrary-precision decimal type - do not reach for it for anything needing more than eight places after the point. - The raw scaled value is a 64-bit long, giving headroom for any realistic account balance or cumulative total, but it is not unbounded - extremely large intermediate values could in principle overflow, the same as any fixed-size integer type. - MultiplyBy and DivideBy use a double internally by necessity, since the factor is not itself fixed-point. That single step does not compound the way repeated addition would, but it is not magic either - do not chain many MultiplyBy calls expecting the same exactness addition gives you. - DivideBy(0) is a silent no-op returning the unchanged value, not an error. Callers that need to detect a zero divisor should check for it themselves before calling.

Script · MT4 · v1.0 · Free

Tags: Binary Options · MT4 Script · MQL4 Source Code

MQL4 & MQL5 scripts · PolyMaria