Modulo Operator %%
I would welcome such an operator as well. I find myself implementing a mod
function from time to time, expressing it in terms of the remainder operator.
As for syntax, I don't see %%
posing any syntactical ambiguities, so I'll
second it.
agreed, let's make a proposal
I would prefer the syntax be ‘a mod b’ consistent with my wishlist item:
esdiscuss.org/topic/new-operator, esdiscuss.org/topic/still-waiting-for-integer-division
In to semantics:
www.microsoft.com/en-us/research/wp-content/uploads/2016/02/divmodnote.pdf
From: es-discuss <es-discuss-bounces at mozilla.org> On Behalf Of Cyril Auburtin
Sent: Tuesday, August 13, 2019 5:07 AM Cc: es-discuss <es-discuss at mozilla.org>
Subject: Re: Modulo Operator %%
agreed, let's make a proposal
On Tue, Aug 13, 2019 at 12:06 AM kdex <kdex at kdex.de<mailto:kdex at kdex.de>> wrote:
I would welcome such an operator as well. I find myself implementing a mod
function from time to time, expressing it in terms of the remainder operator.
As for syntax, I don't see %%
posing any syntactical ambiguities, so I'll
second it.
On Monday, August 12, 2019 10:00:09 PM CEST Matthew Morgan wrote:
JS needs a modulo operator. It currently has the remainder operator
%
which works in most cases except for negative values. I believe the the%%
would work great and be easy to remember.let x = (-13) %% 64; is equivalent to let x = ((-13 % 64) + 64) % 64;_______________________________________________
es-discuss mailing list es-discuss at mozilla.org<mailto:es-discuss at mozilla.org>
Related:
esdiscuss.org/notes/2018-01-24#13vd-operator-overloading-for-stage-1
I don’t see anything newer than this
From: es-discuss <es-discuss-bounces at mozilla.org> On Behalf Of Cyril Auburtin
Sent: Tuesday, August 13, 2019 5:07 AM Cc: es-discuss <es-discuss at mozilla.org>
Subject: Re: Modulo Operator %%
agreed, let's make a proposal
On Tue, Aug 13, 2019 at 12:06 AM kdex <kdex at kdex.de<mailto:kdex at kdex.de>> wrote:
I would welcome such an operator as well. I find myself implementing a mod
function from time to time, expressing it in terms of the remainder operator.
As for syntax, I don't see %%
posing any syntactical ambiguities, so I'll
second it.
On Monday, August 12, 2019 10:00:09 PM CEST Matthew Morgan wrote:
JS needs a modulo operator. It currently has the remainder operator
%
which works in most cases except for negative values. I believe the the%%
would work great and be easy to remember.let x = (-13) %% 64; is equivalent to let x = ((-13 % 64) + 64) % 64;_______________________________________________
es-discuss mailing list es-discuss at mozilla.org<mailto:es-discuss at mozilla.org>
On 8/13/19 7:27 AM, Michael Haufe wrote:
I would prefer the syntax be ‘a mod b’ consistent with my wishlist item:
This can bring up various syntactic troubles. What does the following do?
let mod +3
Is it calling the mod operator on the variable named "let" and +3? Or is it defining a variable named "mod" with no initializer, followed by an expression?
Waldemar
On 8/13/19 7:27 AM, Michael Haufe wrote:
I would prefer the syntax be ‘a mod b’ consistent with my wishlist item:
On 8/13/19 9:12 PM, Waldemar Horwat wrote:
This can bring up various syntactic troubles. What does the following do?
let mod +3
Is it calling the mod operator on the variable named "let" and +3? Or is it defining a variable named "mod" with no initializer, followed by an expression?
I can't declare 'let' or 'var' as variable names, but even if I could (Say non-strict mode or ES3) that form would be a VariableDeclaration followed by an ExpressionStatement.
The proposed grammar extension is:
MultiplicativeOperator: one of * / % div mod
Michael
For syntactic precedent, CoffeeScript and R both have %%
for a
modulus returning the sign of the divisor, where a %% b
is
equivalent to (a % b + b) % b
. So JS wouldn't be the first here.
Most languages Wikipedia lists in their comparison (en.wikipedia.org/wiki/Modulo_operation) use a function for this kind of modulus operation, not an operator. That itself seems worth noting.
Isiah Meadows contact at isiahmeadows.com, www.isiahmeadows.com
On 8/13/19 8:32 PM, Michael Haufe wrote:
On 8/13/19 7:27 AM, Michael Haufe wrote:
I would prefer the syntax be ‘a mod b’ consistent with my wishlist item:
On 8/13/19 9:12 PM, Waldemar Horwat wrote:
This can bring up various syntactic troubles. What does the following do?
let mod +3
Is it calling the mod operator on the variable named "let" and +3? Or is it defining a variable named "mod" with no initializer, followed by an expression?
I can't declare 'let' or 'var' as variable names, but even if I could (Say non-strict mode or ES3) that form would be a VariableDeclaration followed by an ExpressionStatement.
The proposed grammar extension is:
MultiplicativeOperator: one of * / % div mod
And I'm saying that's potentially problematic because it changes the meaning of existing programs that happen to use "mod" as a variable name. The above is one example that would turn a let statement into a mod expression. Here's another example:
x = 4 mod(foo)
Waldemar
Le 12 août 2019 à 22:00, Matthew Morgan <mmm211 at zips.uakron.edu> a écrit :
JS needs a modulo operator. It currently has the remainder operator
%
which works in most cases except for negative values. I believe the the%%
would work great and be easy to remember.let x = (-13) %% 64; is equivalent to
let x = ((-13 % 64) + 64) % 64;
Is there a strong advantage of an %%
operator over a Math.mod()
function? There is the precedent of the **
operator implemented as alternative of Math.pow()
few years ago. It would be interesting to hear the feedback of those that use regularly powers, whether the benefit was clear (personally, I almost never use either Math.pow()
or **
, so that I can’t say anything).
At least one disadvantage of an operator over a function, is that you have to think about precedence. The problem is exacerbated in JS, because (following some other languages) the unary minus has an uncanny high precedence level, confusingly very different than the one of the binary minus; so that, after having designed **
, it was realised at the last minute that -a**b
would be dumbly interpreted as (-a)**b
instead of -(a**b)
or 0-a**b
, as anybody who would be likely to actually use the operator would expect. (That particular issue was resolved in a hurry by making the parenthesis-left form a syntax error.)
FWIW another disadvantage is that operators cannot be polyfilled, so it'll
take forever for those not using transpilers to adopt these, while having a
Math,mod
would work right away
Is there any way we can add function/operator equivalence to the language? Perhaps some kind of 'operators' global object with symbol fields matching operator string to functions/constraints?
Dammit babies, you've got to be kind.
We could circumvent this by making %%
analogous to **
. That is, we could
provide a function form Math.mod
as well as an infix operator %%
.
On 8/14/19 7:50 PM, Waldemar Horwat wrote:
And I'm saying that's potentially problematic because it changes the meaning of existing programs that happen to use "mod" as a variable name. The above is one example that would turn a let statement into a mod expression. Here's another example:
x = 4 mod(foo)
Potentially yes and surely there is a yacc definition where one could check to be certain? Regardless, let's assume there is or that workarounds to guarantee infixity are not worth the complication ([no LineTerminator here] usage).
We know that syntax is the last bastion of language luddites, so it's best not to linger on something which was not my main concern. I am more interested in maintaining the duality of the operators over how they are represented (within reason). Thus, if '%%' is what is preferable, then '\' would be the partner.
Michael
Thursday, August 15, 2019 2:47 AM, Andrea Giammarchi wrote:
FWIW another disadvantage is that operators cannot be polyfilled, so it'll take forever for those not using transpilers to adopt these, while having a
Math,mod
would work right away
With such an approach there is risk of another ‘smooshgate’ [1][2]. There is nothing stopping those developers from using a function anyway to bridge the gap if they can’t or won’t use a compiler. This is already the current state of affairs.
[1] developers.google.com/web/updates/2018/03/smooshgate [2] adamsilver.io/articles/the-disadvantages-of-javascript-polyfills
Michael
An operator is far more concise than a function call, and is likely to see greater use. It also aligns better with peoples' intuition on what the "modulus" is, avoiding subtle bugs like in `isOdd = x => x % 2 ===
1(example from https://en.wikipedia.org/wiki/Modulo_operation#Common_pitfalls - try passing a negative to it). And given this one is high value (see above) and *very* low cost (it can literally desugar to
(x % y + y) %
y`), I feel it does meet that bar.
It would be interesting to hear the feedback of those that use regularly powers, whether the benefit was clear (personally, I almost never use either
Math.pow()
or**
, so that I can’t say anything).
It has enough benefit I've seen CoffeeScript users default to %%
and
only using %
when they explicitly want the dividend-dependent
semantics. And engines with a native %%
, if they can detect the
operands are always non-negative, can optimize it to %
pretty
easily. It's better enough that you'd likely start seeing some
partially legitimate FUD spread about the standard %
.
One other added benefit of using divisor-dependent modulo is that x %% (2**n)
, where x
and n
are integers and n >= 0
, could always
be safely rewritten to x & (2**n - 1)
while still preserving
semantics, but x % (2**n)
does not have this property. For
example:
-1 %% (2**1)
→-1 %% 1
→1
-1 & (2**1 - 1)
→-1 & 1
→1
-1 % (2**1)
→-1 % 2
→-1
BTW, I literally tested all three of these in Chrome's devtools
console, using my x %% y
→ (x % y + y) % y
desugaring.
As for a native implementation and the spec, I'd recommend just doing
copysign(fmod(x, y), y)
instead to retain precision.
At least one disadvantage of an operator over a function, is that you have to think about precedence. The problem is exacerbated in JS, because (following some other languages) the unary minus has an uncanny high precedence level, confusingly very different than the one of the binary minus; so that, after having designed
**
, it was realised at the last minute that-a**b
would be dumbly interpreted as(-a)**b
instead of-(a**b)
or0-a**b
, as anybody who would be likely to actually use the operator would expect. (That particular issue was resolved in a hurry by making the parenthesis-left form a syntax error.)
I doubt this would happen with %%
. It's similar enough to the
existing %
in concept that most would expect it to have the same
precedence. With **
, there was a very unique issue with it: there
were people actually reading it both ways, and even a language
(Python) that interprets -a ** b
and -a**b
differently in light
of that (as (-a) ** b
and -(a ** b)
respectively). That's not a
concern at all with most operators, so it doesn't apply to most new
operator proposals.
Isiah Meadows contact at isiahmeadows.com, www.isiahmeadows.com
To me there's no risk, as MooTools, Prototype, and Scriptacolous are both
things of the past, and never implemented Math.mod ... so, with that
approach, custom transpiling functions are more dangerous, as somebody
might have implemented %%
already for other purposes, and we break Babel
outcome adding new syntax anyway ... the smoosh accident, is the equivalent
of custom Babel utilities these days.
Look at TypeScript and the private class fields, if you want to compare new syntax instead
Fair points, but since **
has its Math.pow
counter part, why wouldn't
%%
have Math.mod
as counterpart too? At least it looks like there's
room for both, if standardized, as the behavior and description would
likely be mostly the same (precedence a part)
Static functions don't have the same risk as prototype functions;
Math.mod
would make sense to add.
One suggestion, though, would be to try to add the API method first, and look at usage for awhile before trying to add the syntax.
BTW, I just wrote up a more precise, formalized proposal over here: isiahmeadows/proposal-divisor-dependent-modulo, and I'd be more than willing to work with a TC39 champion on it. I personally prefer syntax (pretty strongly), but I'm not beholden to it.
I do feel the semantics are simple enough it'd be okay to lower it to
syntax, and it naturally just glides right in. I find it very odd
that some languages use a simple operator %
or relatively short
function for remainder keeping the sign of the dividend but relegate
the version keeping the sign of the divisor (the more useful and
intuitive of them) to a much more verbose function call. Of all
Wikipedia lists in en.wikipedia.org/wiki/Modulo_operation,
here's the four that do this currently - all but one expose an
operator for the first:
- Fortran:
mod
for dividend-dependent,modulo
for divisor-dependent - Julia:
%
/rem
for dividend-dependent,mod
for divisor-dependent - Java:
%
for dividend-dependent,Math.floorMod
for divisor-dependent - XBase++:
%
for dividend-dependent,Mod
for divisor-dependent
And it's worth noting most other languages (including some end user-oriented ones) that show a syntactic preference to one or the other expose a simpler one where the sign matches the divisor, a more complicated one where the sign matches the dividend. For a variety of examples:
- Ruby:
%
/modulo
for divisor-dependent,remainder
for dividend-dependent - SML:
mod
for divisor-dependent,Int.rem
for dividend-dependent - Elm:
modBy
for divisor-dependent,remainderBy
for dividend-dependent - Euphoria:
mod
for divisor-dependent,remainder
for dividend-dependent - Python:
%
for divisor-dependent,math.fmod
for dividend-dependent - Smalltalk:
\\
for divisor-dependent,rem:
for dividend-dependent
And of course, many don't even expose a type of modulo where the sign matches the divisor. For some examples:
- APL
- LibreOffice/Excel
- Lua
- Perl
- Mathematica
- PL/I
- TCL
There's also Dart, a relatively new language which defaults to non-negative always.
This relatively long list of languages, despite C's heritage and semantics being inherited in much of them, makes me question using a function for this, and there would need to be a lot of FUD to get people to use the function more than the operator.
So this is why I would prefer an operator as opposed to syntax for this.
Isiah Meadows contact at isiahmeadows.com, www.isiahmeadows.com
Isiah, thank you for composing this proposal.
+1 for at_least having Math.mod()
And here's a reason why:
The code mod = (x,y) => ( ( x % y ) + y ) % y
is not only slow but
inaccurate. For example, for mod( -249825144445569.97, -2022673516699079.8)
gives -249825144445569.75
(Chrome Version
77.0.3865.35) instead of -249825144445569.97
I then switched to mod = (x,y) => x - Math.floor( x / y ) * y
which
gives an accurate answer, and is faster.
But, on a hunch that flow control would be faster than arithmetic, I
devised this:
mod = (x,y) => {
const z = x % y;
return (y >= 0 && z >= 0 ) || ( y < 0 && z <= 0 ) ? z : y + z;
}
And that is sometimes the fastest approach -- assuming I haven't screwed
up the maths; cf jsbench.me/lrjze7oqv9/2 Although it does seem to
depend on the numbers - with the Math.floor()
approach sometimes coming
out faster.
And this all neglects BigInts, Infinites, NaNs, and signed zeroes.
So this isn't a trivial function to polyfill. And, for that reason alone,
it should be in the standard. (And if we get it in Math, and convince
people to use it, then the pressure for a more ergonomic solution will
increase.)
Peter
Reply inline.
On Friday, August 16, 2019 5:50:40 PM CEST peter miller wrote:
Hi,
+1 for at_least having
Math.mod()
And here's a reason why:
The code
mod = (x,y) => ( ( x % y ) + y ) % y
is not only slow but inaccurate. For example, formod( -249825144445569.97, -2022673516699079.8)
gives-249825144445569.75
(Chrome Version 77.0.3865.35) instead of-249825144445569.97
I then switched to
mod = (x,y) => x - Math.floor( x / y ) * y
which gives an accurate answer, and is faster.But, on a hunch that flow control would be faster than arithmetic, I devised this:
mod = (x,y) => { const z = x % y; return (y >= 0 && z >= 0 ) || ( y < 0 && z <= 0 ) ? z : y + z; }
And that is sometimes the fastest approach -- assuming I haven't screwed up the maths; cf jsbench.me/lrjze7oqv9/2 Although it does seem to depend on the numbers - with the
Math.floor()
approach sometimes coming out faster.And this all neglects BigInts, Infinites, NaNs, and signed zeroes.
Most, if not all, functions on the Math
prototype can't handle BigInt
, but
that's a separate issue.
While it would be nice if Math.mod
supported that, another Math
proposal
could spec that along with fixes for all the other methods.
For now, it should be good enough if we had a modulo operation at all.
Code that uses % is often surprisingly buggy. For example even a simple function such as this is broken if n is negative:
function isOdd(n) { return n % 2 === 1; }
isOdd(-3) // false
The same is not true in other programming languages. The same in Python works just fine:
def isOdd(n): return n % 2 == 1
isOdd(-3) // true
The advantage of an operator over a function is that it allows us to say to people who are new to JS: "Always use %% instead of % unless you have a good reason". Just the same as we say "Always use === instead of == unless you have a good reason".
BTW, that very example is something I cover in my proposal's introduction. It's also part of why I want an operator - this is one of the few times where spreading FUD is a good idea IMHO. isiahmeadows/proposal-divisor-dependent-modulo
Isiah Meadows contact at isiahmeadows.com, www.isiahmeadows.com
Another typical example:
const normalizeHue = hue => (((hue % 360) + 360) % 360) / 360; // phew..
// would become:
const normalizeHue = hue => (hue %% 360) / 360;
Now how does this get to becoming a proposal? Is there a TC39 member that has seen this and will introduce it to the committee?
JS needs a modulo operator. It currently has the remainder operator
%
which works in most cases except for negative values. I believe%%
would work great and be easy to remember.let x = (-13) %% 64;
is equivalent to
let x = ((-13 % 64) + 64) % 64;