Dumb mistakes we make with MATLAB.

O.k., admit it. Who has done this (or something that resulted in the same loss of info) before?

>> laster
??? Undefined function or variable 'laster'.

>> lasterr
ans =
Undefined function or variable 'laster'.

D'Oh! I need a ERRORBEFORELASTERR function.

What's one of your dumb MATLAB mistakes?

[Matt Tearle]

Maybe we need an enhancement request for:

set(0,'IAmAnIdiot',true)

That would go in my startup.m

[jiro]

But what if I type by mistake:

set(0, 'IAmAnIdoit', true)

[Walter Roberson]

Clicking on the above link might get you "page not found". If so, remove the trailing > from the URL.

[Andrew Newell]

My apologies to everyone while I learn how to edit comments. Here is my third try:

This is like David Lodge's "Humiliation Game" http://www.guardian.co.uk/books/2008/jun/13/david.lodge, except that admitting we're dumb doesn't cost us a job (unless we work for Matlab?) and everyone believes us.

[Matt Fig]

Best answer was a toss-up between the top two vote getters.

Answer 1

[the cyclist]

Entering

 >> rand(100000)

instead of

 >> rand(100000,1)

Hm. Why has my whole machine come to a grinding halt?

[Matt Fig]

Done that one way too many times.

[Ashish Uthama]

+1

[Sean de Wolski]

+1

[Jan]

Of course +ones(1)

[Andrew Newell]

Did I ever get bitten by this one today. I had the OS complaining that the startup disk had run out of space for memory (600 GB?!) and telling me to close some applications.

[Bjorn Gustavsson]

My preferred version of this stumble is something along these lines:

idx = find(M(:)<this_or_that);

N = randn(idx);

instead of

N = randn(size(idx));

[Naz]

rand(10000) creates matrix of 10000x10000

rand(10000,1) creates matrix of 10000x1

In the first case you are likely to run out of memory. I think there is a default setting that defines how much memory Matlab can use.

[Steven Lord]

The array size limit preference introduced in release R2015a can help avoid that.

Answer 2

[Andrew Newell]

Trying to edit a command already visible in the Command Window:

>> x=3

x =

     3

Now put the cursor up beside x=3 and try to change it to 4:

>> 4

Oh, right, I should be down here!

[Sean de Wolski]

Yup!

[Paulo Silva]

Yup+1

[Andrew Newell]

assert(Yup=vote)

[Matt Fig]

I do this all the time. I guess I get to thinking I am using a MAPLE worksheet.

[Andrew Newell]

I also try to edit the Preview window in Matlab Answers.

[jiro]

+1 vote on "editing Preview window".

[Sean de Wolski]

Andrew Newell about 2 hours ago

assert(Yup=vote)

Error: The expression to the left of the equals sign is not a valid target for an assignment.

+1 to Preview Window

[Andrew Newell]

It's a greedy algorithm.

Answer 3

[Paulo Silva]

forget to put ; in some vector and array operations and get the command line spammed.

[Matt Fig]

Oh yeah. Been there, done that.

[Sean de Wolski]

I do that all of the time. But just forgetting the ';' ONCE, isn't the bad part. I use the up arrow a lot and so then I rerun the same command w/out the damn ';'. One lost semicolon can stick around for awhile.

Answer 4

[Andreas Goser]

Overlaying command with variables:

plot=1:10;
plot(1:10)
ans =
   1     2     3     4     5     6     7     8     9    10

Or command with commands like creating and own contrast.m function and then using MATLAB's CONTRAST function.

All this typicall leads to errors messages that are not helpful to detect the cause and only if you think of WHICH -ALL, you find it

[Andrew Newell]

And the number of commands grows all the time - especially if you have been visiting the FEX a lot.

[Jan]

x = 1:10; eval('plot=x;'); plot(1:10);

This give different results when run in the command window or inside a function.

Answer 5

[Oliver Woodford]

clear

Oops, I wanted to keep one of those variables.

[Paulo Silva]

I wonder if it's possible to undo such operations?

[Sean de Wolski]

I almost never use clear. I downloaded 'keep' on the FEX and it is far better.

That is of course unless I take the "Nuke it from orbit" approach; for which I wrote a script: 'cll.m' >>clear;close all;clc

Answer 6

[Matt Tearle]

I've been using MATLAB for 15 years or so - I teach people how to use it - and, of course, I stress the important difference between * and .* ...

So guess what once caused me a couple of hours of frustrated debugging of an image reconstruction algorithm...

Physician, heal thyself! (And don't test your algorithm on square images)

Answer 7

[Matt Tearle]

More in keeping with the OP, and another in the "physician heal thyself" category:

[fiddling about at the command line, trying to work out some kinks in an algorithm]

x1 - x2        % looks mostly near zero, but big vector, so:
max(ans)       % oops, I meant...
max(abs(ans))  % no, not *that* ans! ::sigh::

[Paulo Silva]

that one also makes me mad, that's why I almost never use the ans, I prefer to make new variables.

[Andrew Newell]

But the Command History window makes it so easy to patch that boo-boo!

[Matt Tearle]

As I reenter commands I chide myself "thought you'd save those few keystrokes, didn't you..."

[Daniel Armyr]

Yeah, I was tought not to use ans by some teacher way back. Using ans simply gets you into trouble.

Answer 8

[Matt Fig]

Along the lines of the cyclist's popular mistake, I do this one often enough:

x = rand(1:100);  % I meant:  x = rand(1,100);
??? Error using ==> rand
Maximum variable size allowed by the program is exceeded.

[Matt Tearle]

Oh yeah, done that. Not often, but at least once... It's memorable

[Matt Fig]

I do this about twice a year. I would like to blame fat fingers, but the ":" isn't that close to the "," on the keyboard!

[Jan]

@Matt Fig: Use a German keyboard where the key are neighboring. I use it to create "zeros(1:100)" frequently.

Answer 9

[Aurelien Queffurust]

When I say orally to my peers to type

   rehash toolboxcache

in 99 cases out of 100 they type

 rehash toolboxcash

Answer 10

[Jan]

The old EVAL bugs, only a little bit more concealed:

Marker = {'lank', 'rank'};  % Ankle markers
Trajectory = rand(100, 3, 2);
MagicAutoAssign(Marker, Trajectory);
plot(rank);  % FAIL

% ---------------
function MagicAutoAssign(Marker, Trajectory)
for i = 1:numel(Marker)
  assignin('caller', Marker{i}, Trajectory(:, :, i));
end

Now plotting "rank" fails, because Matlab calls the toolbox function RANK instead. But it works, if "rank" is accessed using an index:

plot(1:100, rank(:, 1));   % OK

It was even worse in Matlab < 2008a: Even "RANK" in uppercase letters was forwarded to the function in non-debug mode, but in debug mode the uppercase "RANK" was recognized as variable. So you can create an error, which disappears at debugging - worst case!

Simple conclusion: Never create variables dynamically.

[Rik]

That really is a terrible debug situation...

Answer 11

[Matt Fig]

Not so much with MATLAB, but with MATLAB answers. This one has bitten me several times.

I will edit an answer I had given earlier, in response to more questions or comments from the OP. Often the edit will be fairly lengthy, and so take some time. When I am done I check for spelling errors (usually) then hit the Submit button - at least that is what I should do. Sometimes I hit the Edit link, which causes the page to refresh! All typing lost ..... Aargh!

It would be nice if there simply was no Edit link available when editing (what is it there for anyway?), for dummies like me!

[Andrew Newell]

You're a bad influence, Matt. I had never made this particular mistake - until you wrote about it!

[Matt Fig]

Sorry Andrew, at least I am not alone ;)

Haha, maybe if I bump this post I can infect more regular posters!

If they too complain, then perhaps TMW will remove the Edit button when editing is enabled...

[Matt Fig]

Look at the Edit link! It is calling to you!

[Andrew Newell]

Must ... resist ...

Answer 12

[Andrew Newell]

The case of the misplaced transpose:

v1'*v2

instead of

v1*v2'

[Matt Tearle]

especially when these are very long row vectors... and there's no semicolon at the end of that line.

[Andrew Newell]

At least if there's no semicolon, you notice right away!

Answer 13

[Jan]

c = {'a', 'b', 'c'}
c(2) = ''
>> c = {'a', 'c'}

Ups. I meant C{2} = ''. Another variant:

c = cell(1, 100);
c(:) = ''
>> c = {}

Better: c(:) = {''} to get a cell of empty strings.

Answer 14

[Michael Quinn]

    >> exit

instead of

    >> edit

[Matt Fig]

Interesting. I never use EXIT, so that probably couldn't happen to me!

[Steven Lord]

Go to the Preferences and expand the MATLAB > General > Confirmation Dialogs section. There's an entry, "Confirm before exiting MATLAB", that if you select it will bring up a dialog each time you try to quit MATLAB (normally. If you forcibly kill it from Task Manager, or if MATLAB crashes, it won't be able to show the dialog.)

Answer 15

[Steven]

I often use != instead of the unusual ~=

[Walter Roberson]

http://zork.net/~sneakums/no-perl.svg

[Jan]

I had a "if a~=b" instead of "if a!=b" in a C-mex function. It even produced the correct result - usually.

Answer 16

[Knut]

for i = 1:3
 disp(exp(i*pi));
end

Instead of:

for i = 1:3
 disp(exp(1i*pi));
end

Answer 17

[Adam]

Since I happened across this old thread, one that continually catches me out since I work a lot with plotting of complex signals is to the effect of

figure; plot( mySignal )
hilbSignal = hilbert( mySignal )
hold on;
plot( hilbSignal );

Doh! Pretty spider's web, but not quite what I was looking for. One day I will remember I have complex data and that I actually want either the real or imaginary part! In this case I keep just thinking of the hilbert transform result as the imaginary part only that it adds.

Answer 18

[Andrew Newell]

Rejoice - now you can do it object-oriented style!

>> ME = MException
??? Error using ==> MException
Not enough input arguments

>> ME = MException.last

ME = 

MException

Properties:
  identifier: 'MATLAB:minrhs'
     message: 'Not enough input arguments.'
       cause: {0x1 cell}
       stack: [0x1 struct]

Methods

[Matt Fig]

Except I was using MATLAB 7.01!

[jiro]

Even with this, you can make the same mistake:

>> MException.lst

??? No appropriate method, property, or field lst for class MException.

>> MException.last

ans =

MException

Properties:

identifier: 'MATLAB:noSuchMethodOrField'

message: 'No appropriate method, property, or field lst for class MException.'

cause: {}

stack: [0x1 struct]

Methods

Answer 19

[Andrew Newell]

While in the debugger, typing quit instead of dbquit kills Matlab!

(My earlier version of this answer may have been too cryptic.)

Answer 20

[Jan]

Working:

n = datenum(datestr(now, 29), 29)

Failing (Matlab 2009a):

n = datenum(datestr(now, 30), 30)
>> Failed to lookup month of year.

DATENUM(S, F) needs a string as format specifier F. If F is numerical, it is interpreted as pivot year. With F=29 DATENUM works on accident only.

Answer 21

[Jan]

I've tried to start an instance of another Matlab version, but the current directory was a toolbox folder by accident:

In Matlab 6.5, Windows:

cd(fullfile(matlabroot, 'toolbox\matlab\strfun');
system('C:\Programme\Matlab\R2009a\bin\win32\MATLAB.exe &')

==> Matlab 2009a starts with a lot of warnings about finding M-files, which are builtin functions. Then the script STRNCMP (the help text only!) is attempted to execute as a function. Because this fails, the Matlab path contains just \toolbox\local. Without toolbox function the diagnosis was tedious.

I had a similar break down, after overloading STRCMP with a damaged version.

[Matt Fig]

That is one mistake I have not made.

[Jan]

I'm glad to hear that. It is a mess and a silly waste of time. Another version of such faults:

M-script with the help text only in one folder:

\MFiles\Folder\MyFun.m

Compiled function for different platforms in specific folders, which are included in the path dynamically:

\MFiles\ForMatlab6\MyFun.dll

\MFiles\ForMatlab7\MyFun.mexw32

Now calling MyFun from the command line or another M-function fails, if \MFiles\Folder\ is the current folder: "Attempt to call the script MyFun as a function". Solution: CD to a neutral folder.

Answer 22

[Jan]

I wanted to add the optimization flag /arch:SSE2 in my MEX call in Matlab 2009a:

mex -O OPTIMFLAGS="$OPTIMFLAGS /arch:SSE2" func.c

For a batch processing I need the functional form (or EVAL?!):

mex('-O', 'OPTIMFLAGS="$OPTIMFLAGS /arch:SSE2"', 'func.c')

>> ERROR: flag /arch:SSE2 is not recognized.

Ugly parser! The over-complicated BAT-> M-> PERL-> Compiler pipeline of the MEX command tries to be intelligent. This works:

mex('-O', 'OPTIMFLAGS="$OPTIMFLAGS', '/arch:SSE2"', 'func.c')

The string must be split inside the double quoted section, although these double quotes are used to avoid the splitting!

Answer 23

[Jan]

Check if a path name contains a certain folder name anywhere

if any(findstr(PathName, '/Sub/')) ...

Because FINDSTR searches the shorter string in the longer one, the condition is true for the path name '/' also.

Modern Matlab version use STRFIND(String, Pattern) and afaik FINDSTR will be deprecated. But the program containing the example was developped under Matlab 5.3. Fixing the bug by changing FINDSTR to STRFIND is not trivial, if the strings are created dynamically, e.g. FINDSTR(a{i}, ['@', b{j}]).

At least in Matlab 2009a the toolbox functions contains a lot of these pitfalls, e.g. Signal\PMEM: "if ~isempty(findstr(flag, 'CORR')), ...", which triggers for 'OR' also.

Answer 24

[Ben Mitch]

Moving into the Figure window, but in a similar spirit to the rand(100000) thing, I use this when I need an excuse to take a break...

data = randn(2, 1e6);
plot(data)

when I meant this...

plot(data')

Answer 25

[Grzegorz Knor]

Sometimes I make a typo in structure field name, for example:

my_struct.field1 = 1

and then:

my_struct.filed1 = 2

It produces:

my_struct = 

      field1: 1
      filed1: 2

instead of:

my_struct = 

      field1: 2

[Walter Roberson]

Too bad we cannot "lock down" structures to prevent new fields from being added.

[Jan]

@Walter: We can. I'm using a selftest function, which scans the source code and checks all symbols, which contain a dot. The structs are created in a dedicted function, and no other function is allowed to add further fields.

I care for a minimal edit-distance, e.g. I avoid S.mvc and S.mcv, which would be too susceptible for typos. Of course I never create fields by complicated EVAL commands or LOAD structs directly to the workspace.

Using OO methods would be a more convenient method to limit the definition of fields. But even for functional programs, OO strategies are important for writing reliable software.

Answer 26

[Jan]

The process-first-non-singelton feature can be smarter than the user:

m = randi(4);
n = randi(4);
M = rand(m, n);
ColumnSum = sum(M);  % DANGEROUS

If m is 1, the sum is calulated alog the 2nd dimension. Better:

ColumnSum = sum(M, 1);  % Secure

NOTE: For min, max and std the 2nd input is not the dimension!

max([1, 2; 3, 4], 2)      % >> [2, 2; 3, 4], elementwise maximun
max([1, 2; 3, 4], [], 2)  % >> [2; 4], maximun along 2nd dimension

[the cyclist]

The squeeze() command has similar dangers, when there are unanticipated singleton dimensions. In my experience, it is almost always possible, and wiser, to use permute() rather than squeeze().

[Jan]

permute()? I'd expect reshape() to do the equivalent operation without the smart choice of the dimension.

Answer 27

[Shatrughan]

have anyone tried typing "Why" in the command prompt? MATLAB gives funny responses..just carry on typing for fun.. :)

[Image Analyst]

I've listed some other Easter Eggs in http://www.mathworks.com/matlabcentral/answers/2001-what-matlab-easter-eggs-do-you-know

Answer 28

[Jan]

Although I never use clear all (see Answers: Bad CLEAR ALL), it appears in user-supplied code frequently. If all is defined as a variable, the behaviour might be unexpected:

clear all
b   = 1;
all = 2;
clear all
whos
%   Name      Size    Bytes  Class     Attributes
%   b         1x1         8  double

If "all" is a variable, it is cleared, but the other variables are not touched.

[EDITED] The "clear all" behaviour is a typical example for the bad programming practize of using magic numbers. Here the string 'all' is magic, because it triggers a totally amnesia, if there is no variable of this name. It would be smarter to use invalid symbols for special commands, e.g. "$all". See Wiki: magic numbers and Wiki: magic strings.

[Daniel Shub]

From now on I might just have to answer homework questions with code that has variable names of all and clear (and maybe builtin and str2func just to up the challenge).

[Jan]

And I should take the time to publish my functions unclear() and totallyunclear() in the FEX.

[Daniel Shub]

I am not sure I see the use of magic strings in this case as particularly bad. Mostly because I don't use clear in my programming. In fact, clear is one of the few (doc, help, and edit being the others that come to mind) functions that I rarely call in its functional form.

[Jan]

The magic strings are not bound to the functional form.

"S = 'all'; clear(S);" behaves equivalently. See my next answer concerning magic strings in labels of UIMENUs.

I do not suffer from the magic string "all" for the CLEAR command, because I never overload the builtin function ALL and avoid a CLEAR('all') at all.

But the same problem concerns "clear('variables')", "clear('global')", "clear('functions')", "clear('import')" and "clear('classes')". Clearing an M-file of the same name is not possible and using the name "import" as a variable is not unlikely.

For the newer flag "-regexp" TMW has been so smart to use the minus to _unmagicalize_ the string. The same for "which <> -all" or "save -mat". The usage of magic constants without an escape character can and will lead to bugs or to large and ugly constructions to catch the exceptions.

[Walter Roberson]

Magic strings are a MathWorks mistake, not a dumb mistake on our part.

[Daniel Shub]

@Jan, my point was that I don't use the CLEAR command in any of my programs. The magic strings in UIMENU is more problematic.

[Jan]

@Walter: The magic strings let us make mistakes, when we forget to consider them. E.g. I forgot the escape character in a context menu, where I can move an object to its "default" position. It is in the nature of context menus, that an automatic unit-test cannot find this bug, but only an active user.

I'm sure most programmers omit a test for {'default', 'remove', 'factory'}, when they assign the string of buttons or menus dynamically.

[Eric Sampson]

To be honest I think a lot of the functions that were created in early MATLAB versions are the ones that tend to have syntax issues like this... Since they've been around so long it would be hard for TMW to break backwards compatibility. As TMW hired more professional programmers over the years, instead of mathematicians/scientists doubling as programmers, things have improved IMHO, like your example about '-regexp'. Of course this is all personal opinion YMMV etc, and it's not intended to be a slight at all to TMW old-timers :)

[Eric Sampson]

P.S. I've also recently requested that TMW consider making the use of 'clear all' an M-lint/Analyzer warning, because it has side effects that trip people up (like clearing breakpoints & globals, flushing the M-file cache, etc). If you like the idea, call/email support and lend your vote to this enhancement request! :)

Answer 29

[Jan]

The strings 'default', 'remove', and 'factory' are not allowed as uimenu labels, see Ref: uimenu_props. To get the string 'default' as label, '\default' must be used.

This is another example for the bad magic strings. Imagine you want to create a bunch of uimenu s automatically using a list of words. Then the exception handling for these three words will be ugly and increase the program size without a real benefit.

[EDITED] The named strings have a special meaning for other handle graphics objects also, e.g. UICONTROLs:

uicontrol('String', 'default')

This doe not display 'default' in a button, but an empty string, because this is the default value:

get(0, 'defaultUIControlString')

To display 'default', 'factory', or 'remove', a leading backslash is required, see doc: setting-default-property-values:

uicontrol('String', '\default')

It would be smarter to use the backslash for the special commands, but this cannot be changed without breaking the backward compatibility.

[Jan]

What is required to display the string '\default' in an UICONTROL???

[Walter Roberson]

Maybe \\default ?

Answer 30

[Geoff]

Corrupting years of experience in other languages...

After using MatLab for several months now, I find that when I go back to writing C code I catch myself forgetting to put parentheses on my if statements and using single-quotes for my strings.

[Geoff]

Oh yeah I knew there was another... Writing comments with %. Grrrr

[Jan]

And MATLAB with lower-case a, t, a, b. Can anybody explain, why MATLAB is written with an upper-case L frequently? I did this myself also. Has it been the official spelling in Matlab 4?

See Stevens confusion of ~= and != above.

[Daniel Shub]

I believe the name at one point meant Matrix Laboratory

[Sean de Wolski]

It still does!

Answer 31

[Kye Taylor]

>> close al
Error using close (line 111)
Specified window does not exist.

Dammit!

Answer 32

[Jan]

I needed a short and confusing command for a demonstration:

eval = 1:255
eval eval

Error: "eval" was previously used as a variable, conflicting with its use here as the name of a function or command.

This works:

eval('eval')
>> 100   117    96   107

Answer 33

[Michael Richards]

fig = figure('CloseReqFcn',@error_prone_fcn)

presses x button

error

alt-f4

error

tries close matlab

error

task manager > end task

error

[Steven Lord]

Specifying 'force' with the close command will force the figure to close, bypassing its CloseRequestFcn.

fig = figure('CloseRequestFcn',@(varargin) error('No closing this figure!'));
close(fig) % Will not close the figure
close(fig, 'force') % Will close the figure
fig % This is now a handle to a deleted figure