"warning takes sprintf-like arguments directly" not entirely true?

If I write this:

message = sprintf('%d %d %d', [1 2 3])
warning(message)

I get a warning from MATLAB Code Analyzer at the second line, saying "warning takes sprintf-like arguments directly".

However if I follow its recommendation and I type:

warning('%d %d %d', [1 2 3])

I get the error "Formatted arguments cannot be non-scalar numeric matrices.".

Is this the desired behaviour?

[Cedric]

Which version of MATLAB are you using, 2018a as indicated?

[KALYAN ACHARJYA]

[Massimiliano Zanoli]

@Cedric yes, R2018a as indicated. Fresh installation.

@KALYAN ok, now try with:

warning('%d %d %d', [1 2 3])

[Cedric]

I don't have time to investigate much now but I'll have to (sigh). I've been using WARNING, ERROR and ASSERT like SPRINTF, which is supported on my 2017b:

Answer 1

[Image Analyst]

Get rid of the brackets so you have scalars instead of a vector:

>> warning('%d %d %d', [1 2 3])
Error using warning
Formatted arguments must be scalar. 
>> warning('%d %d %d', 1, 2, 3)
Warning: 1 2 3

So it must not like that fact that you're trying to put in one matrix with 3 format specifiers (%d).

[Massimiliano Zanoli]

Thank you for your suggestion. However this does not work when one doesn't know the number of arguments in advance, as was the point with vectors. Previous MATLAB versions would at least repeat the warning for every element in the vector, now one has to for-loop them...

[Paul Wintz]

You can use mat2str to convert arrays to strings.

Answer 2

[Jan]

This is the documented behavior [EDITED] since R2018b:

doc warning

You find:

 warning(msg,A1,...,An)
 A1,...,An — Values
 character vector | string scalar | numeric scalar
 Values that replace the conversion specifiers in msg, specified as a
 character vector, string scalar, or numeric scalar.

[1,2,3] is not a numeric scalar.

So either use sprintf or provide the elements separately:

data  = [1,2,3];
dataC = num2cell(data);
warning('%d %d %d', dataC{:})

While it is documented, one could still discuss, if this is the "desired" behavior. I did not struggle with this feature in the last 20 years.

[Cedric]

We did not struggle with this because it used to be supported. On my 2017b:

 A1,...,An — Numeric or character arrays
 scalar | vector | matrix | multidimensional array

[Jan]

@Cedric: Oh! I've searched in the release notes and did not find a corresponding hint: https://www.mathworks.com/help/matlab/release-notes.html?rntext=warning&startrelease=R2017a&endrelease=R2018b&groupby=release&sortby=descending&searchHighlight=warning.

[Massimiliano Zanoli]

Ok so, if I understand correctly, vectors were supported by warning in 2017 but no longer in 2018, and the code analyzer hint has not been updated accordingly. Alignment between warning and sprintf is not so strong anymore.

[Cedric]

There are several "things" going on here actually.

The usual way of using FPRINTF, SPRINTF, and also WARNING, ERROR, ASSERT, is to pass as many extra args as there are data fields in the formatSpec. The classic way is given by Image Analyst in his answer: you have three data fields defined by %d in the formatSpec '%d %d %d', and therefore you pass three extra args for three values and not a vector of values (which is a single arg):

 warning( '%d %d %d', 1, 2, 3 ) ;

and not

 warning( '%d %d %d', [1,2,3] ) ;

Now F/SPRINTF and formerly WARNING, ERROR, and ASSERT would repeat the formatSpec as many times as needed for matching the size of vectors and arrays of values (when possible), and you could do the following:

You can see this as the program "eating the data with bites of size the number of data fields in the formatSpec".

What is not supported anymore apparently is this behavior, which (after thinking a bit about it) should legitimately not be used in error/warning messages, because it messes up the formatting when you pass inappropriate vectors/arrays and people would certainly misunderstand and make mistakes:

Here you see that the formatSpec with a single data field is repeated nine times (the number of elements in the array), which is neither appropriate nor useful.

[Walter Roberson]

Which is why you might want to sprintf first to construct the message and then warning() it... only for mlint to tell you that you should have used warning() directly even though it no longer supports doing the same thing.

[Massimiliano Zanoli]

The hint is somewhat misleading, especially with this new MATLAB release... Not a big deal, I agree, but I still spent my nice half-an-hour on it... Thank you for your investigations anyway.

[Matthias Roosz]

An old topic - but still annoying for those not switching to new matlab releases every year...

IMHO totally undesired behaviour. I liked using sprintf-like arguments to warning (and error) - as the message hint STILL suggests. All code relying on dealing array values to multiple format specifiers breaks with the change to R2018b or later with no hint to be found in the release notes or the code analyzer :'(

distrib = [0.23 3.1415 123.4];  % dummy values
% warning('bad distrib (1%%/median/99%%): %.1f/%.2f/%.1f', distrib)  % used to work
warning('bad distrib (1%%/median/99%%): %.1f/%.2f/%.1f', distrib(1), distrib(2), distrib(3))  % how is this better?

Warning: bad distrib (1%/median/99%): 0.2/3.14/123.4

distribC = num2cell(distrib);
warning('bad distrib (1%%/median/99%%): %.1f/%.2f/%.1f', distribC{:})  % or this?

Warning: bad distrib (1%/median/99%): 0.2/3.14/123.4

A solution (apart from repeating the variable name holding the data to be printed) that does not require an extra line of code before each warning/error would still be interesting. Maybe we'll get something like python's *args? I've tried this

expand = @(x) x{:};
warning('bad distrib (1%%/median/99%%): %.1f/%.2f/%.1f', expand(num2cell(distrib)))

Warning: bad distrib (1%/median/99%): 0.2/

but no luck because the function only 'sees' one argout, thus only reports the first value...

[Walter Roberson]

No mlint warning for this, at least not in current releases:

distrib = [0.23 3.1415 123.4];  % dummy values
SIT = @(x) struct('d', num2cell(x));
warning('bad distrib (1%%/median/99%%): %.1f/%.2f/%.1f', SIT(distrib).d)

Warning: bad distrib (1%/median/99%): 0.2/3.14/123.4

This relies upon a somewhat new feature, that the return value of a function call may now be indexed with dot notation. Unfortunately I am having difficulty finding the appropriate release notes entry to say which release this started with.