# Empty Arrays with Flow of Control and Logical Operators 17

Posted by **Loren Shure**,

After reading last week's post on calculating with empty arrays, one of my colleagues mentioned some other behaviors with empty arrays that have tripped
him up in the past. Today I will discuss how empty arrays work in the contexts of flow of control expressions (both conditional
and looping, i.e., `if` and `while`) and short-circuit operators (i.e., && and | |).

### Contents

### Empty Arrays in Flow of Control

Let me first start with plain empty arrays in flow of control situations. For example, what will this code do?

E = []; if E disp('Empty is true') else disp('Empty is false') end

Readers who remember my comment on last week's blog will correctly guess that the empty expression for the `if` statement is treated as `false`. Why? The way I think about it is this. If I am looking for locations of some condition in an array, and I don't find them,
I end up with an empty output. This very output is the kind of expression I am likely to want to use, somehow, in an `if` statement. Let's try it to be sure.

E = []; if E disp('Empty is true') else disp('Empty is false') end

Empty is false

The situation gets a bit more complicated if there is a logical expression for the `if` or `while` statement that has an empty array as one of its elements. Let me show you what I mean. Paraphrasing the documentation,

There are some conditions however under which while evaluates as true on an empty array. Two examples of this are

A = []; while all(A), doSomething, end while 1|A, doSomething, end

Let's see what's going on in each of these examples. In the first one, the function `all` is being called with an empty input. According to the second reference below (on empty arrays), the function `all` is one of the functions that returns a nonzero value for empty input. Let's see.

allE = all(E) allEislogical = islogical(allE)

allE = 1 allEislogical = 1

The way I think about this is that there are no `false` values in `E`, hence the `true` result.

### Empty Arrays with Logical Operators

The second expression involves an elementwise logical operator ( | ). In this case, the first part of the expression, `1`, is true, so the second part, after the elementwise `or`, is never evaluated. So the fact that an empty result returns `false` never comes into play here. Why? Because & and | operators short-circuit *when and only when they are in the context of if or while expressions*. Otherwise, the elementwise operators do

**not**short-circuit.

In contrast, the logical operators, && and | |, always short-circuit, regardless of context.

### Short-circuit Logical Operators (| | and &&)

The next important idea to remember is that the short-circuit logical operators expect scalars as the inputs for the expressions. This means that an empty array, not being a scalar, may cause you some grief if you are unprepared for that situation. Let me show you what I mean. Compare the following 2 code snippets.

true || E

ans = 1

try E || true catch ME disp(ME.message) end

Operands to the || and && operators must be convertible to logical scalar values.

In the second snippet, the expression E || true

produced an error, because E isn't a scalar value. Once the error occurs, the second operand is never evaluated. Contrast
that with the snippet, where the first input evaluates to `true`. Short-circuiting then takes over and the second operand, which would cause an error in this context, is never evaluated.

### Examples

Here are a few more code examples to help you see the patterns. Try to figure out the answers before reading the results.

if [] disp('hello') else disp('bye') end

bye

true | []

ans = []

[] | true

ans = []

true || []

ans = 1

try [] || true catch ME disp(ME.message) end

Operands to the || and && operators must be convertible to logical scalar values.

if true | [] disp('hello') else disp('bye') end

hello

if [] | true disp('hello') else disp('bye') end

bye

if true || [] disp('hello') else disp('bye') end

hello

try if [] || true disp('hello') else disp('bye') end catch ME disp(ME.message) end

Operands to the || and && operators must be convertible to logical scalar values.

### References

Here are a bunch of references to the MATLAB documentation where all of this information is covered.

### Empty Thoughts?

The behaviors with empties in MATLAB are, I believe, consistent and useful. Nonetheless, the behaviors have lots of details to master and can be confusing. If you have any thoughts on the matter, please respond here.

Get
the MATLAB code

Published with MATLAB® 7.9

## 17 CommentsOldest to Newest

I would love to hear people’s thought on the practice of relying on short-circuit behavior instead or careful isempty checking. For example, if you had a

if (A || B), dosomething, end;

in your code, and B can potentially be empty, would you consider it safe to leave as is? What about readable? There could of course be more involved expressions, where you would have to look carefully to make sure an empty array would not cause an error.

Thanks,

-n

I understand “if []” evaluating to true. That’s good. I also see why the logical operators do what they do. But this is confusing to me:

For any array that is not empty, if ALL returns true, ANY does too. The empty array seems to be the only exception!

Chris-

I understand and sympathize with your thoughts on any/all and empty. I think of all and any looking for exceptions, in some way. So, for all, if there are no false values, as there are not with empty, then the result is true. For any, I think with the opposite polarity. If I can find any true values then it’s true but empty has no true values. It boils down to empty have neither trues nor falses, and any and all using that to decide truth.

–Loren

Crosslink from StackOverflow: http://stackoverflow.com/questions/1710299/corner-cases-unexpected-and-unusual-matlab

Naor,

I think you are conflating two issues: (1) the use of short-circuit operators, and (2) the use of non-logicals in conditional expressions.

The first item seems unobjectionable. Short-circuit operators are extremely useful and can make the code more expressive.

On the second item (non-logical scalars), I tend to agree with you. It seems better — in terms of maintainability and clarity — to say:

instead of

even though they are equivalent.

Loren, thanks for your explanation in response to Cris’s question. I think it makes sense. For ALL, it’s probably more efficient to look for any occurrence of false. Once it finds at least one false, then it can return false. For ANY, it’s more efficient to look for any occurrence of true. Once it finds at least one true, then it can return true.

Loren

While I understand the outcome of all([]) is true as designated by TMW, I must say that there is an objection that can be raised from traditional formal logic considerations. If, given

E = [];

we can think of the statement:

all(E)

as being translated, “All elements of E are nonzero,” as the documentation suggests, then we have the problem of existential import.

For example, a corresponding statement which may be more easily considered is the following,

“All circle-squarers are non-mathematicians.”

We can say that this statement is true only when we consider the mental concept of a circle-squarer, because our minds can imagine a circle-squarer even if it is contradictory to do so. But this is not true when we consider the tacit term ‘real’ in the statement. As in:

“All (real) circle-squarers are non-mathematicians.”

This statement MUST be false because, in traditional formal logical terms, the term ‘circle-squarer’ fails to designate.

Now back to E. We immediately see that the term ‘elements of E’ fails to designate, and thus the statement, “All elements of E are nonzero,” must be false. I think this is why people have a hard time rationalizing TMW designation of all(E) as true.

It should also be noted that in modern formal logic, such statements have no requirement for existential import. If TMW is using modern formal logic, which it probably should, then ALL statements which contain a term that does not designate are trivially true. I.e., both

“All elements of E are non-zero.”

and

“All elements of E are zero.”

are both true. Thus there is no reasoned defense of TMW designation which cannot also be made to support the opposite designation. It is an arbitrary choice that had to be made one way or the other by TMW; it is better memorized than rationalized.

At least that is how I see it. ;-)

Matt/Loren,

Defining all([])=true has the additional advantage that all([A;B])=(all(A)&all(B)) holds when either array is empty.

The older versions used to generate an error for “if []” didn’t they? (although “if nan”, “if {}” and “if i” still do!) I guess that why I still prefer to write “if ~isempty(x) && all(x(:))” when it’s needed.

Ben-

Another good way to look at the behavior! Thanks.

–Loren

Ben

if {} errors because the expression has to evaluate to something numeric, even if empty. Same for i. Neither one can be converted to a logical array.

–Loren

It’s not exactly on topic but what I miss in the try-catch syntax is a flow as in the if-elseif-end-syntax. Thus I would like to see a try-elsetry-catch block. I think it’s a very natural thing to have, no?

Best,

Marcel

Marcel-

Please feel free to use the link at the right to make this enhancement suggestion. I personally don’t mind instead using another try-catch inside the catch block. I find the nesting clarifies my thinking.

–Loren

This behavior has tripped me up quite a few times now.

Although the rules make sense individually, they are inconsistent*.

What trips me up is when I use an if statement and add a condition later. So I start with

and then decide that C is a condition too, so I turn it into

only to have things crash if A is empty – which was previously not a problem.

I find it counterintuitive that adding code concerning C changes the way A is treated. This makes error-checking a lot more difficult, until you learn to take countermeasures.

I appreciate being able to use empty as false: if the variable has no value, it definitely doesn’t have the right one. This is very intuitive for me. However, I’m now considering teaching myself not to use this shortcut (and adding a short-circuited ~isempty instead), because it might trip me up later, unless I can learn to remember that when adding a condition to an if or while, I also need to add a short-circuited isempty…

I’d really like for the short-circuiting operators to evaluate in the same way that if and while do; not just for empties but also for arrays and such. Is there a particular reason that they require scalar values? Could we get ||| and &&& (for instance) to short-circuit for arrays (including empties)?

* I must admit that it’s quite probable that there is no consistent way to handle this, other than never accepting [] where a logical is expected. This would be more robust for the user (smaller chance of a small change causing large problems), but less convenient. So maybe all of this simply can’t be helped.

I am unable to resolve the below error. I am unable to figure out where the problem is becuase; Iam not using any || and && in my program nor combinations of if and while statements.

ERROR:

??? Operands to the || and && operators must be convertible to logical scalar values.

Error in ==> fzero at 333

elseif ~isfinite(fx) || ~isreal(fx)

Error in ==> BRALHX_Condenser_Equations at 69

Fdel=fzero(Fun,1.4766e-007,options);

It will really helpful if I understand more about this error with respect to fzero function.

Cheers

Vinz

Vinz-

You might try dbstop if error so you can see the link of code and the values where the issue is exposed.

If that doesn’t help, please contact technical support (link on right of blog) with enough information to reproduce the issue so they can help you.

–Loren

Hi

I have a problem with using these short-circuit operators.

Say, x = NaN; x(:,:,2) = NaN; x(:,:,3) = NaN;

y = 1; y(:,:,2) = 2; y(:,:,3) = 3;

if isnan(x(:,:,:)) == 0 && isnan(y(:,:,:)) == 0

% Execute functions

end

I want to execute certain functions only if I have at least one non-NaN value in x and at least one NaN value in y. This means that the loop should end in the above example because all values of x is NaN.

But I get this error: ‘Operands to the || and && operators must be convertible to logical scalar values’

Am I doing something wrong here?

Thanks

Sundar-

isnan is an elementwise function, so it’s returning arrays the same size and as x and y, not scalars, which is why && is complaining. You may want to judiciously use the function any with ~isnan and collapse the right dimensions to get a scalar before doing the &&.

–Loren