| name | matlab-modernize-code |
| description | Modernize removed or discouraged MATLAB functions and patterns. Use when the check_matlab_code MCP tool reports that a function "is not recommended", "will be removed in a future release", or "has been removed" — at any severity (info, warning, or error) — when migrating legacy code, or when replacing removed or discouraged APIs with current equivalents. Also applies to discouraged patterns Code Analyzer does not flag (containers.Map, bsxfun, str2num, eval, clear all) and to removed or discouraged functions such as trainNetwork, csvread, xlsread, datenum, guide, svmtrain, ltln2val, webmap.
|
| license | https://www.mathworks.com/content/dam/mathworks/license/pmrl/license.md |
| metadata | {"author":"MathWorks","version":"1.3"} |
Code Modernization
Replace removed or discouraged MATLAB functions and anti-patterns with modern equivalents. This skill is the resolver — the MCP tool check_matlab_code (MATLAB's Code Analyzer) is the detector.
Terms used in this skill:
- Removed — the function is no longer in MATLAB; calling it errors.
- Discouraged — the function still works, but a better API exists. It may or may not be slated for removal in a future release.
When to Use
check_matlab_code reports a lifecycle diagnostic — any of these three
wordings, matched case-insensitively (STRMATCH is not recommended is the
same signal as the lowercase form):
'X' is not recommended... (severity: info)
'X' will be removed in a future release... (severity: warning)
'X' has been removed... (severity: error — the code errors today)
- Code uses a discouraged pattern Code Analyzer does not flag — e.g.
containers.Map, bsxfun, str2num, eval, clear all — scan the source, not just the diagnostics (see Conventions)
- User asks to modernize, migrate, or update old MATLAB code
- Code uses functions listed in the quick reference table below
- After static analysis reveals removed or discouraged API usage
- Writing new code in a domain that has known removed or discouraged patterns
When NOT to Use
This skill swaps removed or discouraged APIs for modern equivalents. It does not:
- Review code quality broadly
- Debug runtime behavior (removed-function errors excepted — those are modernization fixes)
- Optimize already-correct code for speed → hand off to
matlab-optimize-performance
- Write or migrate function input/output validation (
arguments blocks, inputParser/validateattributes → mustBe* validators, repeating args, name-value forwarding) → hand off to matlab-validate-function-arguments
Quick Reference: Removed, Discouraged & Anti-Patterns
| Topic | Recommendation | Since | Category | Rationale |
|---|
csvread / dlmread | readmatrix | R2019a | File I/O | |
csvwrite / dlmwrite | writematrix | R2019a | File I/O | |
xlsread | readtable, readmatrix | R2019a | File I/O | |
xlswrite | writetable, writematrix | R2019a | File I/O | |
datenum / datestr / now | datetime (use string/char to format) | R2022b | Date/Time (Not recommended) | Flagged by Code Analyzer |
eval / evalc / evalin | Dynamic field names s.(name), function handles / feval, cells for sequential names | — | Anti-pattern | Not compiled (slow); can overwrite workspace vars; hard to debug; injection risk |
str2num | str2num(text, Evaluation="restricted") | R2022a | Security | Unrestricted form runs input via eval (injection risk); see references |
uicontrol | uibutton, uidropdown, etc. | R2016a | UI/App | |
guide | appdesigner | R2025a | UI (Removed) | |
strmatch | startsWith, matches | R2019b | Strings | |
clear all |
datestr nuance: datestr's job was to format a date as text, so its modern replacement is
string(dt) / char(dt) or setting dt.Format — not a bare datetime(...) call. Use datetime
to replace datenum/now (the numeric/serial-date path). Neither is removed; both are Code
Analyzer "not recommended" as of R2022b.
s = datestr(t, 'yyyy-mm-dd HH:MM:SS'); % old
s = string(t, 'yyyy-MM-dd HH:mm:ss'); % new — note the specifiers change:
% datetime uses MM=month, mm=minute (datestr had mm=month, MM=minute)
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
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");
eval → Structured Alternatives
Avoid eval/evalc/evalin: the text isn't compiled (slower), can silently overwrite workspace
variables, and is hard to read/debug. Match the intent to a construct — see
references/core-functions-guidance.md for the full set (sequential names, sprintf filenames,
try/catch).
% Old: eval([varName ' = 42;']); -> dynamic field name
s.(varName) = 42;
% Old: result = eval(['process_' method '(x)']); -> handle dispatch table
handlers.fast = @processFast;
handlers.slow = @processSlow;
result = handlers.(method)(x);
% Old: for n=1:10, eval(['A' int2str(n) '=magic(n);']); end -> index a cell
A = cell(10,1);
for n = 1:10, A{n} = magic(n); end
References
Load these when working in a specific domain:
| Load when... | Reference |
|---|
Core MATLAB functions (file I/O, strings, UI); security anti-patterns (eval/str2num), containers.Map→dictionary, bsxfun→implicit expansion | references/core-functions-guidance.md |
| Deep Learning (trainNetwork→trainnet, LayerGraph/SeriesNetwork/DAGNetwork→dlnetwork, classify, activations) | references/deep-learning-guidance.md |
| Signal processing functions | references/signal-processing-guidance.md |
| Audio/video I/O migration (wavread, aviread) | references/audio-video-guidance.md |
| Optimization toolbox (optimset, optimtool) | references/optimization-guidance.md |
| Control systems plot options | references/control-systems-guidance.md |
| Image processing ROI objects | references/image-processing-guidance.md |
| Statistics/ML (svmtrain, dataset, classregtree) | references/statistics-ml-guidance.md |
| Communications System objects | references/communications-guidance.md |
Conventions
- Always run
check_matlab_code first — let static analysis find removed or discouraged usage
- After running it, scan the source for patterns the Code Analyzer may under-flag:
str2num (framed only as a perf tip); containers.Map (see next point).
containers.Map is not flagged by check_matlab_code. A plain script using it returns zero diagnostics; a containers.Map property default surfaces only a handle-default-sharing warning ("all instances share the same object data"). The detect-first workflow will not find this migration — treat any containers.Map usage, and that handle-sharing warning on a property, as a containers.Map→dictionary trigger.
subplot is not deprecated or discouraged. It is fully supported and is never flagged by check_matlab_code. tiledlayout/nexttile is a preferred alternative for new code, not a required migration — do not tell the user to replace working subplot code.
- Fix removed or discouraged patterns before other code quality issues
- When writing new code, use the modern pattern from the start — don't write legacy code and fix it later
- A
'X' has been removed diagnostic (severity error) means the function errors today — treat it as a break-fix modernization, ahead of "not recommended" (info) and "will be removed" (warning) cases
- When migrating, test the modern replacement against the old behavior to confirm equivalence
- Legacy
.m files are often Windows-1252 encoded. After an edit, verify non-ASCII characters (°, µ, ±) survived — a diff showing an unrelated one-line change on a comment or string is usually a re-encoded byte, not your edit.
- Consult the domain-specific reference file for detailed migration patterns with code examples
Copyright 2026 The MathWorks, Inc.