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_codereports a lifecycle diagnostic — any of these three wordings, matched case-insensitively (STRMATCH is not recommendedis 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 (
argumentsblocks,inputParser/validateattributes→mustBe*validators, repeating args, name-value forwarding) → hand off tomatlab-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 |
clearvars |
— | Workspace | Clears functions from memory; forces recompilation |
datestrnuance:datestr's job was to format a date as text, so its modern replacement isstring(dt)/char(dt)or settingdt.Format— not a baredatetime(...)call. Usedatetimeto replacedatenum/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 |
| Mapping Toolbox (webmap, wmmarker, wmline, geotiffread, mfwdtran, makerefmat) | references/mapping-guidance.md |
Conventions
- Always run
check_matlab_codefirst — 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.Mapis not flagged bycheck_matlab_code. A plain script using it returns zero diagnostics; acontainers.Mapproperty 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 anycontainers.Mapusage, and that handle-sharing warning on a property, as acontainers.Map→dictionarytrigger.subplotis not deprecated or discouraged. It is fully supported and is never flagged bycheck_matlab_code.tiledlayout/nexttileis a preferred alternative for new code, not a required migration — do not tell the user to replace workingsubplotcode.- 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 removeddiagnostic (severityerror) 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
.mfiles 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.