[Date Prev][Date Next][Thread Prev][Thread Next]
[Date Index]
[Thread Index]
- Subject: Re: Wishful request for next version: '// 1' should produce integer
- From: Francisco Olarte <folarte@...>
- Date: Fri, 7 Jan 2022 09:54:54 +0100
Hi, just my 2¢...
On Fri, 7 Jan 2022 at 09:00, Lorenzo Donati <lorenzodonatibz@tiscali.it> wrote:
> On 03/01/2022 19:28, Flyer31 Test wrote:
> > if in some furture version integer division // 1 would produce integer
> > result, this would be very nice.
> > Now only int // int produces integer, but float // int produces float,
> > which somehow is a bit "ugly" for the result of an integer division.
For th OP:
In general float-op-any and any-op-float produce floats, in Lua and in
all languages.
After all 2.0+1 produces a float too.
And // is not an integer division, "Floor division (//) is a division
that rounds the quotient towards minus infinity, resulting in the
floor of the division of its operands. "
....
> Just having the oddball case of just "// 1" producing an integer goes
> against the principle of least surprise and probably could slow things
> down: every time the engine needed to perform // it would have to check
> whether or not to convert its result according to its 2nd operand.
This I agree. Although I could tolerate it being optimized away.
> I'd rather prefer to change the semantics of // to always return an
> integer, unless someone convince me the current behavior is more useful.
> After all if you write "//" you probably want an integer, not a float
> with integer value.
Bear in mind // can operate correctly in numbers not representable as
integer, i.e.
> 1E60//1
1e+60
If I want an integer, I would use will use math.tointeger and deal
with overflow stuff. math.floor, suggested by OP, is underdocumented
but it seems very dangerous to use, except for display purposes, as it
varies its return type:
> math.floor(1E60)
1e+60
> math.floor(1.0)
1
I think special casing // will lead to much more confusion than there
is now. Specially if you consider you will have to define it not only
for the 2.5//1 case, but also for the 1//2.5, 1.0//2.5, 2.5//0.1 and
similar stuff, and it will have some new failure modes and corner
cases, like 1E70//1E60 or 1E-60//1E-70.
BTW, the 2.5//0.1 I've been using it as graph_intervals = data_range
// tick_width +1 and similar stuff.
Mixed arithmetic is complex, the "use integer if both are integer,
float otherwise" way is easy to understand. Lua can still surprise you
with things like
> 123456789123456789.*100000000
1.2345678912346e+25
> 123456789123456789*100000000
-7473169249324075776
> 12345678912345678900000000
1.2345678912346e+25
But this are few and less.
Francisco Olarte.