linearToLogScale and linearToLogScale in include/lmms_math.h do something completely unrelated to their function names

[Cyp]

eˣ = exp(x) = pow(M_E, x) ≠ x^e = pow(x, M_E) = "logToLinearScale"
log(x) ≠ pow(x, 1/M_E) = "linearToLogScale"

They seem to be used only in AutomatableModel.

[JohannesLorenz]

Here are the functions:

lmms/include/lmms_math.h

Lines 197 to 208 in a9e3e70

	static inline float logToLinearScale( float min, float max, float value )
	{
	if( min < 0 )
	{
	const float mmax = qMax( qAbs( min ), qAbs( max ) );
	const float val = value * ( max - min ) + min;
	float result = signedPowf( val / mmax, F_E ) * mmax;
	return isnan( result ) ? 0 : result;
	}
	float result = powf( value, F_E ) * ( max - min ) + min;
	return isnan( result ) ? 0 : result;
	}

lmms/include/lmms_math.h

Lines 212 to 225 in a9e3e70

	static inline float linearToLogScale( float min, float max, float value )
	{
	static const float EXP = 1.0f / F_E;
	const float valueLimited = qBound( min, value, max);
	const float val = ( valueLimited - min ) / ( max - min );
	if( min < 0 )
	{
	const float mmax = qMax( qAbs( min ), qAbs( max ) );
	float result = signedPowf( valueLimited / mmax, EXP ) * mmax;
	return isnan( result ) ? 0 : result;
	}
	float result = powf( val, EXP ) * ( max - min ) + min;
	return isnan( result ) ? 0 : result;
	}

[JohannesLorenz]

static in a header function does make no sense

[PhysSong]

@JohannesLorenz Why do you marked this for 1.2? Shouldn't we take care of compatibility issues?

static in a header function does make no sense

Not always, see this: https://gist.github.com/htfy96/50308afc11678d2e3766a36aa60d5f75

[JohannesLorenz]

@PhysSong It's invalid and it's in stable-1.2. It might even generate wrong sounds.

Though... If we write a conversion function, that might break more than it might fix, and there are currently not many complaints from musicians. I think we can milestone it 1.3.

[PhysSong]

I think we may implement a 'power' scale(x^a) and upgrade the existing weird-and-fake logscale to it. In this way, we can support the old behavior while adding the correct behavior by converting the old 'logscale' to x^(1/e). However, it may confuse users.

[Rossmaxx]

So basically, the powf() parameters should be swapped but wasn't swapped due to a backward compat issue. How much impact will this make?

[Rossmaxx]

If there's a compat break, might as well re milestone it but let someone clarify.

[Rossmaxx]

Oops