splitapply does not retain ME stack

When the function that splitapply throws an error, this line number of this error is not include in the output (matlab2024b)

Example:

T=table([1:3]');
G=findgroups(T);
splitapply(@fun, T, G);
function out = fun(varargin)
ERROR
end

This returns:

Error using splitapply>localsplitandapply (line 196)
Unable to apply the function 'fun' to the 1st group of data.

Error in splitapply (line 135)
varargout = localsplitandapply(fun,dataVars,gdiffed,sgnums,gdim,nargout);

Error in templateLoopOverAll (line 231)
splitapply(@fun, T, G);

Caused by:
    Unrecognized function or variable 'ERROR'.

I was expeting that the line number that actually caused that error would be included:

    Error in testfile>fun (line 5)
    ERROR
    
    Error in testfile (line 3)
    fun()

Missing this information makes the debugging of fun very challenging.

How to retrieve this line number?

[Vasco]

I think that this data is removed in splitapply.m line 243.

Answer 1

[Vasco]

The solution is to alter in splitapply.m line 242 from:

    throwAsCaller(addCause(MException(m.Identifier,getString(m)),ME));

to:

    rethrow(addCause(ME, MException(m.Identifier,getString(m))));

Answer 2

[Matt J]

Run with Pause on Errors.

Then the code will stop at the offending line:

[Matt J]

Youc an also invoke Pause on Errors from the command line with,

>> dbstop if error

[Vasco]

dbstop if caught error

That works!

From the menu it does not work:

• This does not work on my installation, it stops at splitapply.m line 196
• That seems not to work when I press: run this section (crtl-enter) -> it does not stop at al

[Walter Roberson]

Try

dbstop if caught error

[Matt J]

dbstop if error should be sufficient in R2024b when the script is run in full. My demo above used that.

The other observations are intended behavior.

[Vasco]

That indeed halts the debugger at this place, but it also halts the debugger in many other places (for example in some of the plotting functions). This makes it hardly usable.

[Matt J]

I use it all the time and have never had a problem with it stopping in plotting functions.

'dbstop if caught error' is usually more prone to stop in places you don't want, and will fail to stop at explicit errors that you do want to intercept.