Code Modernization

Replace deprecated MATLAB functions and anti-patterns with modern equivalents. This skill is the resolver — check_matlab_code is the detector.

When to Use

• check_matlab_code or checkcode returns "not recommended" or "to be removed" warnings
• User asks to modernize, migrate, or update old MATLAB code
• Code uses functions listed in the quick reference table below
• After static analysis reveals deprecated API usage
• Writing new code in a domain that has known deprecated patterns

When NOT to Use

• Reviewing code quality broadly — use matlab-review-code (which may then trigger this skill)
• Debugging runtime errors — use matlab-debugging
• Performance profiling — use performance skills (though anti-patterns below overlap)

Quick Reference: Top Deprecated Functions

Critical Anti-Patterns

Never use these in new code:

Modern Design Patterns

Prefer these in all new code:

Table-Based Workflows

data = readtable('sensors.csv');
data.Timestamp = datetime(data.Timestamp);
data.Status = categorical(data.Status);
recentData = data(data.Timestamp > datetime('today') - days(7), :);
summary = groupsummary(recentData, 'SensorID', 'mean', 'Value');

String Arrays (not char arrays)

name = "John";                        % not 'John'
names = ["John", "Jane", "Bob"];      % not {'John','Jane','Bob'}
fullName = firstName + " " + lastName; % not [first,' ',last]
idx = contains(names, "Jo");          % not cellfun + strfind

Arguments Block (not nargin/varargin)

function result = processData(data, options)
    arguments
        data (:,:) double
        options.Method (1,1) string {mustBeMember(options.Method, ["fast","accurate"])} = "fast"
        options.Verbose (1,1) logical = false
    end
end

Vectorization (not loops)

% Instead of: for i=1:n, V(i) = pi/12*(D(i)^2)*H(i); end
V = pi/12 * (D.^2) .* H;

% Instead of: loop with if
Vgood = V(D >= 0);   % logical indexing

Preallocation

result = zeros(1, n);     % numeric
C = cell(1, n);           % cell array
S(n) = struct('f1', []);  % struct array

Key Migrations

File I/O: csvread/xlsread → readmatrix/readtable

% Old                          → Modern
M = csvread('data.csv');       % M = readmatrix('data.csv');
M = dlmread('data.txt','\t'); % M = readmatrix('data.txt','Delimiter','\t');
[n,t,r] = xlsread('f.xlsx');  % T = readtable('f.xlsx');
csvwrite('out.csv', M);       % writematrix(M, 'out.csv');
xlswrite('out.xlsx', data);   % writetable(T, 'out.xlsx');

Deep Learning: trainNetwork → trainnet

% Old: classificationLayer specifies loss implicitly
net = trainNetwork(X, Y, layers, options);

% Modern: specify loss explicitly, no classificationLayer needed
net = trainnet(X, Y, layers, "crossentropy", options);

% Prediction
scores = minibatchpredict(net, XTest);
YPred = scores2label(scores, classNames);

eval → Dynamic Field Names / Function Handles

% Old: eval([varName ' = 42;']);
s.(varName) = 42;

% Old: result = eval(['process_' method '(x)']);
handlers.fast = @processFast;
handlers.slow = @processSlow;
result = handlers.(method)(x);

References

Load these when working in a specific domain:

Conventions

• Always run check_matlab_code first — let static analysis find deprecated usage
• After checkcode, scan the source for patterns checkcode misses: subplot (not flagged), str2num (sometimes not flagged), global variables, growing arrays may only warn about size change
• Fix deprecated patterns before other code quality issues
• When writing new code, use the modern pattern from the start — don't write deprecated code and fix it later
• For functions marked "Removed" — they will cause immediate errors, not just warnings
• When migrating, test the modern replacement against the old behavior to confirm equivalence
• Consult the domain-specific reference file for detailed migration patterns with code examples

Copyright 2026 The MathWorks, Inc.