RE: help - warning division by zero

[Joe Tusek]

-----Original Message-----
From: Andrew Janke <address@hidden> 
Sent: Friday, 25 October 2019 5:22 PM
To: Joe Tusek <address@hidden>; address@hidden
Subject: Re: help - warning division by zero

Hi Joe,

Convention on this mailing list is to "bottom-post", putting your reply below 
the previous emails. Scroll down to see reply.

On 10/25/19 2:12 AM, Joe Tusek wrote:
> -----Original Message-----
> From: Andrew Janke <address@hidden>
> Sent: Friday, 25 October 2019 5:04 PM
> To: Joe Tusek <address@hidden>; address@hidden
> Subject: Re: help - warning division by zero
> 
> 
> 
> On 10/25/19 1:58 AM, Joe Tusek wrote:
>> Hi,
>>
>>
>> I have code that uses ifelse statements to prevent the possibility of 
>> a division by zero. It appears though that all the conditions of the 
>> ifelse are evaluated as part of the ifelse execution and this results 
>> in a warning about a division by zero, even if the code should 
>> prevent such a case.
>>
>> This leaves a warning of division by zero in the Command Window when 
>> the "executed code" had not resulted in one.
>>
>> The code snippet has been crafted to generate the behaviour
>>
>>>> ifelse (1==1,0,1/0)
>>
>> warning: division by zero
>>
>> ans = 0
>>
>>>> ifelse (1==0,0,1/0)
>>
>> warning: division by zero
>>
>> ans =  Inf
>>
>>>> ifelse (1==0,0,1)
>>
>> ans =  1
>>
>> Should this warning be supressed (so that it does not appear in the 
>> Command Window) when there was no active code that resulted in a 
>> division by zero?
> 
> Hi Joe,
> 
> What is this "ifelse" statement or function that you're using in your example 
> code? Octave doesn't have an "ifelse" statement.
> 
> If this is a function you've implemented yourself, you need to know that 
> Octave is an "eager-evaulation" language: all the input arguments to a 
> function are *always* evaluated before being passed in. So that "1/0"
> that is the 3rd argument to the function always happens. This isn't like a 
> Lisp or whatever where you can have "lazy" evaluation.
> 
> To have something like this "short-circuit" and prevent argument evaluation 
> requires special support at the language level.
> 
> Cheers,
> Andrew
> 

> I hadn't even considered the possibility that it was not an Octave
command. I just thought it was undocumented.
>
> What is the easiest way to find out where it is coming from?
>
> I tried stepping into it but couldn't.
>
> Regards,
> Joe
>


Holy cow, it actually exists! My bad.

> What is the easiest way to find out where it is coming from?

The easiest way to find out where any function or named thing is coming from is 
using the "which" command.

>> which ifelse
'ifelse' is a built-in function from the file libinterp/corefcn/data.cc

>> help ifelse
'ifelse' is a built-in function from the file libinterp/corefcn/data.cc

 -- merge (MASK, TVAL, FVAL)
 -- ifelse (MASK, TVAL, FVAL)
     Merge elements of TRUE_VAL and FALSE_VAL, depending on the value of
     MASK.


So, it actually is an Octave thing. But since it's a _function_, and not a 
special syntax, it doesn't have the short-circuiting behavior you want. That 
is, you might be familiar with the ternary operator "condition ? if-case : 
else-case" from C/C++ and elsewhere.
Unfortunately, Octave's ifelse() doesn't work like that.

If you want real short-circuiting behavior, you'll have to use regular 
"if/else" statements at the Octave/M-code syntax level. You can write them as 
one-liners if you really want:

if 1==1; 0 else 1/0; end


No warning happens when you do it this way.

Cheers,
Andrew


Now it all makes sense. 
I can't use it with that behaviour as it will confuse users about whether the 
division by zero affected the outcome. I will just have to do it longhand . 
Thanks.