Hi,

i was wondering - when will float be supported by Value?
(couldn't find it in the roadmap)

in the interrim - is there a bazar or patch thingy one could point me to?

thanks in advance

--m

having a flot in memory - writing it out through value/json , then reading it again incurs a conversion and its not the same number - ( i guess on a personal level i am probably puzzled how this is not bothering you - but thats beside the point)

Hi,

I could see some benefit for supporting float in Value too. (I do signal processing with floats, so they frequently end up in various locations in my code.)
Having Value supporting float and EditFloat/EditFloatSpin added would allow cleaner code with dialogs.

Currently I have to round the float value to a clean double with roundr() to avoid excessive decimal places in EditDouble/EditDoubleSpin display. Also, when reading the value out from EditDouble/EditDoubleSpin, I will need to cast first to (double) and only thereafter to (float).

As an example, I have filtering frequencies controlled with:

void SetHPF(float fc);
float GetHPF();

Filling the EditDoubleSpin:

hpf<<=roundr(GetHPF(),3);

Hi,

The following code portrays the float rounding issue:

#include <Core/Core.h>

using namespace Upp;

CONSOLE_APP_MAIN
{
	Value v(0.002f);
	Cout() << v << "\n";
	Cout() << FormatDouble(v) << "\n";
	Cout() << FormatG(v,7) << "\n";
}

The result of running the above follows:

0.0020000000949949
0.0020000000949949
0.002
<--- Finished, press [ENTER] to close the window --->

I do not know how to solve this cleanly. In any case a regular user seeing 0.002000000095 in a field where he expects to see 0.002, will not be happy about it. For years I have used roundr() all over the code to fix this up, but having a solution hidden in Core would be awesome!

Best regards,

Tom

OK, here's an idea to handle the EditDouble/EditDoubleSpin rounding issue. Adding something like:

hpf.Pattern("%.7g");

to the constructor of the dialog. This will yield possibly sufficient rounding to the float value being represented as double to avoid excessive decimals.

Hi,

The following code portrays the float rounding issue:

#include <Core/Core.h>

using namespace Upp;

CONSOLE_APP_MAIN
{
	Value v(0.002f);
	Cout() << v << "\n";
	Cout() << FormatDouble(v) << "\n";
	Cout() << FormatG(v,7) << "\n";
}

The result of running the above follows:

0.0020000000949949
0.0020000000949949
0.002
<--- Finished, press [ENTER] to close the window --->

I do not know how to solve this cleanly. In any case a regular user seeing 0.002000000095 in a field where he expects to see 0.002, will not be happy about it. For years I have used roundr() all over the code to fix this up, but having a solution hidden in Core would be awesome!

Best regards,

Tom

	double x = 0.002;
	DDUMP(x);
	float fx = x;
	x = fx;
	DDUMP(x);

That said, all of this made me think that if we knew that it is float in Value, we might be able to apply for something like FormatG(v,7) instead when displaying it (e.g. in EditDouble).

const dword FLOAT_V  = 13;

Hi Mirek,
Quote:

That said, all of this made me think that if we knew that it is float in Value, we might be able to apply for something like FormatG(v,7) instead when displaying it (e.g. in EditDouble).

Would that involve introducing:

const dword FLOAT_V  = 13;

and 'DOUBLE_V' -like processing for it, except when putting it out in FormatG() and friends?

This is in some ways pretty much what I attempted to do with the above "float Value prototype" above, but that is really bad as I do not really understand all the details of Value. You may wish to take a look anyway -- and then trash it.

Best regards,

Tom

No problem... I trust you will soon come up with some astonishingly smart three line solution for the task!

Best regards,

Tom

Tom1 wrote on Sat, 02 April 2022 13:31

No problem... I trust you will soon come up with some astonishingly smart three line solution for the task!

Best regards,

Tom

Nothing too smart for now, but I have added ConvertFloat and EditFloat.

The jury is still out for float in Value.... do not want to do that now. It feels like while it fixes some issues (e.g. Value->JSON conversion when the values went through float precision), there might be many caveats.

operator float() const           { return Is(DOUBLE_V) ? (float)GetSmallRaw<double>() : (float)GetOtherDouble(); }

typedef WithSpin<float, EditFloat>			 EditFloatSpin;
typedef WithSpin<float, EditFloatNotNull>	 EditFloatNotNullSpin;

// And then some magic to make those two working and visible in layout editor.

Hi Mirek,

Sorry to return to this subject, but could you consider adding Null support for float?

I had something like this in mind... (in Core/Defs.h):

const int    INT_NULL           =    INT_MIN;
const int64  INT64_NULL         =    INT64_MIN;

constexpr double DOUBLE_NULL    =    -std::numeric_limits<double>::infinity();
constexpr float FLOAT_NULL      =    -std::numeric_limits<float>::infinity();

class Nuller {
public:
	operator int() const                { return INT_NULL; }
	operator int64() const              { return INT64_NULL; }
	operator double() const             { return DOUBLE_NULL; }
	operator float() const              { return FLOAT_NULL; }
	operator bool() const               { return false; }

	Nuller() {}
};

extern const Nuller Null;

template <class T> void SetNull(T& x) { x = Null; }

template <class T> bool IsNull(const T& x)       { return x.IsNullInstance(); }

template<> inline bool  IsNull(const int& i)     { return i == INT_NULL; }
template<> inline bool  IsNull(const int64& i)   { return i == INT64_NULL; }
template<> inline bool  IsNull(const double& r)  { return !(std::abs(r) < std::numeric_limits<double>::infinity()); }
template<> inline bool  IsNull(const float& r)   { return !(std::abs(r) < std::numeric_limits<float>::infinity()); }
template<> inline bool  IsNull(const bool& r  )  { return false; }

Although, I'm not entirely sure, if this is completely correct way to do it.

Best regards,

Tom

PS. EDIT: I think the above should work mostly. Only the "Cout() << FLOAT_NULL;" prints -inf instead of empty field, which is the default for "Cout() << DOUBLE_NULL;":

CONSOLE_APP_MAIN{
	double d=Null;
	float f=Null;
	double a=f;
	float b=d;
	Cout() << "d = " << d << "\n";
	Cout() << "f = " << f << "\n";
	Cout() << "a = " << a << "\n";
	Cout() << "b = " << b << "\n";
	Cout() << "a == f : " << (bool)(a==f) << "\n";
	Cout() << "b == d : " << (bool)(b==d) << "\n";
	Cout() << "IsNull(d) = " << IsNull(d) << "\n";
	Cout() << "IsNull(f) = " << IsNull(f) << "\n";
	Cout() << "IsNull(a) = " << IsNull(a) << "\n";
	Cout() << "IsNull(b) = " << IsNull(b) << "\n";
}

Hi,

What has changed is that I have really started to enjoy the new EditFloatSpin!

As a result, initializing and reading my float variables to/from EditFloatSpin (as used in e.g. filter parameters) is clean and easy. However, I found that initializing my EditFloatSpin to empty field (e.g. when such filtering is not currently used) requires my float variable to be at Null:

EditFloatSpin hpfedit;
...
float hpf=(float)(double)Null;
...
hpfedit <<= hpf;

Alternatively, some value of hpf (e.g.< 0) could be interpreted as empty filtering and I could do hpfedit.Clear(); if such is the case.

However, I prefer nice, short and clean code, so I would like to write just:

float hpf=Null;

I.e. without any type cast. So, in effect, this is just to make the code cleaner.

Best regards,

Tom

Why dont you just use

double hpf;

?

constexpr float Nullf = -std::numeric_limits<float>::infinity();
inline bool IsNull(const float& r)   { return !(std::abs(r) < std::numeric_limits<float>::infinity()); }
inline void SetNull(float& x) { x = Nullf; }

Quote:

Why dont you just use

double hpf;

?

Well, all my signal processing code runs on 32-bit floats, and therefore, the various coefficients/parameters are also floats. It is straight forward to keep it up in the user interface too. If I used doubles in the GUI, I would end up converting forth and back all those parameters. Never wish to go back there, now that I have EditFloat and EditFloatSpin!

Anyway, if you feel seriously reluctant to add Null support for float, I can live with it: I figured out a way to do it outside of Core almost cleanly:

constexpr float Nullf = -std::numeric_limits<float>::infinity();
inline bool IsNull(const float& r)   { return !(std::abs(r) < std::numeric_limits<float>::infinity()); }
inline void SetNull(float& x) { x = Nullf; }

Still, it would be nicer inside Core... After all, it would introduce only three lines of new code in Core/Defs.h.

Best regards,

Tom

Hello jimlef,

Do you mean this?

	RLOG(Format(
		"Pi(2): %.02f\n"
		"PI(4): %.04f\n"
		"PI(8): %.08f\n",
		roundr(M_PI, 2),
		roundr(M_PI, 4),
		roundr(M_PI, 8)
		));

	double tax = 0.0800000000123;
	String stax = FormatG(tax, 3);
	RLOG("Formatted tax: " << stax);
	RLOG("Scanned   tax: " << ScanDouble(stax));

	Json jtax_r("tax-rounded", roundr(tax, 2));
	Json jtax_n("tax", tax);

	RDUMP(jtax_r);
	RDUMP(jtax_n);

Pi(2): 3.10
PI(4): 3.1420
PI(8): 3.14159270

Formatted tax: 0.08
Scanned   tax: 0.08
jtax_r = {"tax-rounded":0.08}
jtax_n = {"tax":0.0800000000123}

[Edit: Code updated.]

Best regards,
Oblivion

- I can use Value with float as easily as with double without explicit type casts anywhere

- Assigning a float to Value reads back from Value exactly the same as it went in

- If EditFloat* is again replaced by EditDouble*, the behavior is exactly the same as it is now with EditFloat*

- float supports Null

Maybe the precision hint you are suggesting could be automatically initialized in Value(double) / Value(float) constructors and assignment operators =(double) / =(float) to suit the assigned data type?

	double x = 3.14159;
	Value json;
	json("pi") = Precision(x, 2); // Precision does not exist yet
	DDUMP(AsJSON(json));

{"pi":3.14}

Just revisting this issue...

Tom1 wrote on Fri, 03 June 2022 12:03

- I can use Value with float as easily as with double without explicit type casts anywhere

Yes.

Quote:

- Assigning a float to Value reads back from Value exactly the same as it went in

If "reads back" means text output, then yes.

Quote:

- If EditFloat* is again replaced by EditDouble*, the behavior is exactly the same as it is now with EditFloat*

Should work.

Quote:

- float supports Null

No. Is that a problem?

Quote:

Maybe the precision hint you are suggesting could be automatically initialized in Value(double) / Value(float) constructors and assignment operators =(double) / =(float) to suit the assigned data type?

That was exactly the plan.

I think advantage of this solution is that it allows you to specify arbitrary precision for any double-holding Value. E.g.

	double x = 3.14159;
	Value json;
	json("pi") = Precision(x, 2); // Precision does not exist yet
	DDUMP(AsJSON(json));

{"pi":3.14}

Still an idea though.

float a = 1.234562f;
float b = 1.234562f;
Value c = b;
b = c;
if(a == b) Cout() << "Great, it works!\r\n"
else Cout() << "No, it does not read back the same...\r\n";

mirek wrote on Wed, 04 October 2023 10:55

Just revisting this issue...

Tom1 wrote on Fri, 03 June 2022 12:03

- I can use Value with float as easily as with double without explicit type casts anywhere

Yes.

Quote:

- Assigning a float to Value reads back from Value exactly the same as it went in

If "reads back" means text output, then yes.

Quote:

- If EditFloat* is again replaced by EditDouble*, the behavior is exactly the same as it is now with EditFloat*

Should work.

Quote:

- float supports Null

No. Is that a problem?

Quote:

Maybe the precision hint you are suggesting could be automatically initialized in Value(double) / Value(float) constructors and assignment operators =(double) / =(float) to suit the assigned data type?

That was exactly the plan.

I think advantage of this solution is that it allows you to specify arbitrary precision for any double-holding Value. E.g.

	double x = 3.14159;
	Value json;
	json("pi") = Precision(x, 2); // Precision does not exist yet
	DDUMP(AsJSON(json));

{"pi":3.14}

Still an idea though.

Hi Mirek,

Thanks for the update on this subject.

For me 'reads back exactly the same' means binary equality:

float a = 1.234562f;
float b = 1.234562f;
Value c = b;
b = c;
if(a == b) Cout() << "Great, it works!\r\n"
else Cout() << "No, it does not read back the same...\r\n";

As for float Null, yes, please include the changes shown in:

https://www.ultimatepp.org/forums/index.php?t=msg&th=120 82&goto=59874&#msg_59874

This works (and I believe it should) right now. Am I missing something?

Quote:

As for float Null, yes, please include the changes shown in:

https://www.ultimatepp.org/forums/index.php?t=msg&th=120 82&goto=59874&#msg_59874

I am reluctant adding yet another Null...

Mirek

	float f = (float)(double)Null;
	...
	Cout() << IsNull((double)f) << "\r\n";

	float f = Null;
	...
	Cout() << IsNull(f) << "\r\n";

mirek wrote on Fri, 06 October 2023 13:51

This works (and I believe it should) right now. Am I missing something?

Quote:

As for float Null, yes, please include the changes shown in:

https://www.ultimatepp.org/forums/index.php?t=msg&th=120 82&goto=59874&#msg_59874

I am reluctant adding yet another Null...

Mirek

Hi Mirek,

No, you're not missing anything here. My code was there just to demonstrate what I mean by 'reads back exactly the same'.

Is there a specific reason why you wish to avoid adding another Null to support float?

OK, float support (very hesitantly and conditionally) added.

https://github.com/ultimatepp/ultimatepp/blob/cb32981c04d35f 061c0d700dac26e0fe546d36a7/autotest/Float/main.cpp

Some other tests I should add?

mirek wrote on Mon, 30 October 2023 11:34

OK, float support (very hesitantly and conditionally) added.

https://github.com/ultimatepp/ultimatepp/blob/cb32981c04d35f 061c0d700dac26e0fe546d36a7/autotest/Float/main.cpp

Some other tests I should add?

Hi Mirek,

Thank you very, very much! Actually, the float Value support is even wider than I would have imagined... but very much appreciated.

No other tests come to mind.

BTW: What are the conditions and consequences of your hesitation you're referring to above?

EDIT: PS. In Core/Value.cpp, is it intentional or a mistake to have different return typecast for float in GetOtherInt(), GetOtherInt64(), GetOtherBool() and GetOtherDouble()?