R2021a

New Features, Bug Fixes, Compatibility Considerations

Environment

Live Editor Controls: Create dynamic controls in live scripts by linking variables to drop-down items and slider values

When adding a drop-down list to a live script, you can populate the items in it using values stored in a variable. When adding a slider, you can specify the minimum, maximum, and step values using variables.

For example, create a live script and define the variable lastnames that contains a list of last names.

lastnames = ["Houston","Vega","Obrien","Potter","Rivera","Hanson","Fowler","Tran","Briggs"];

Run the live script to create lastnames and add it to the workspace. Then, go to the Live Editor tab, and in the Code section, select Control > Drop Down. In the Items section of the control configuration menu, select lastnames as the Variable.

Drop-down list configuration menu with the lastnames variable selected and the lists of item labels and item values populated with the variable values.

Close the configuration menu to return to the live script. The drop-down list now contains the last names defined in lastnames.

Drop-down list containing a list of 8 last names.

If you add, remove, or edit the values in lastnames, the items in the drop-down list update accordingly.

For more information, see Add Interactive Controls to a Live Script (R2021a).

Live Editor Fonts: Change the name, style, size, and color of fonts programmatically using settings

You can change the name, style, size, and color of titles, headings, text, and code in the Live Editor using settings.

For example, this code changes the color and style of titles in the Live Editor:

s = settings;
s.matlab.fonts.editor.title.Style.PersonalValue = {'bold'};
s.matlab.fonts.editor.title.Color.PersonalValue = [0 0 255 1];

This code increases the size and changes the font of normal text in the Live Editor:

s = settings;
s.matlab.fonts.editor.normal.Size.PersonalValue = 20;
s.matlab.fonts.editor.normal.Name.PersonalValue = 'Calibri';

The Live Editor updates all open live scripts and live functions to show the selected fonts. When you create new live scripts or functions, they use the new fonts as well.

live script with a title that is blue and bold, and a line of normal text in calibri size twenty font

For more information, see matlab.fonts (R2021a).

Live Editor Display: Specify where to display output by default

You can change the default location for output in a new live script. Depending on this preference, new live scripts that you create display their output inline or on the right. To change the default, go to the Home tab, and in the Environment section, select Preferences. Select MATLAB > Editor / Debugger > Display, and then select an option for the Live Editor default view:

Output on right — Output displays to the right of the code. Each output displays next to the line that creates it. This option is ideal when writing code.
Output inline — Output displays inline with the code. Each output displays underneath the line that creates it. This option is ideal for sharing.

Live Editor Functions: Run live functions interactively using the Run button in MATLAB Online

In MATLAB^® Online™, you can now run live functions interactively using the Run button.

To run live functions that require input argument values or any other additional setup, configure the Run button by clicking Run and adding one or more commands. For more information about configuring the Run button, see Configure the Run Button for Functions (R2021a).

Live Editor Bookmarks: Navigate quickly between lines

To navigate quickly between lines, set bookmarks in your live scripts or live functions. Bookmarks are particularly useful in long files when you frequently need to move between sections.

To set a bookmark, put the cursor on the line that you want to set it on. Then, go to the Live Editor tab and, in the Navigate section, click Bookmark. A bookmark icon appears to the left of the line. To clear a bookmark, with the cursor on the line with the bookmark, click Bookmark and select Set/Clear. You also can clear the bookmark by clicking the bookmark icon to the left of the line.

To navigate to a bookmark, go to the Live Editor tab, and in the Navigate section, click Bookmark . Then, select Previous or Next.

For more information about navigating within files, see Go To Location in File (R2021a).

Live Editor Animation Playback Controls: Interactive interface to control animations

Playback controls now appear within the figure window after an animation is done playing. These playback controls provide the ability to replay the animation and explore individual frames without having to re-run the entire live script. Animation playback controls are not supported for animations generated by the movie function.

Live Editor Performance: Improved performance when saving large live scripts or functions

When saving large live scripts or functions, you can continue using the Live Editor sooner in R2021a than in R2020b. While you continue to use the Live Editor, MATLAB saves the file in the background. When MATLAB finishes saving the file, the asterisk (*) next to the file name disappears, indicating that the file is saved.

For example, on a Windows^® 10, Intel^® Xeon^® E5-1650 CPU @ 3.60 GHz test system, if you save a live script containing 35,000 lines of code and then click the title bar of another document open in the Live Editor, MATLAB immediately switches to the other open document. In R2020b, there is a noticeable delay before MATLAB switches to the other open document.

Help Browser: View web documentation by default

Starting in R2021a, when you run MATLAB with an internet connection, the Help browser displays the web documentation by default. When you run MATLAB on a system without an internet connection, or if your internet connection becomes unavailable, the Help browser displays the installed documentation instead.

To change the default documentation location, on the Home tab, in the Environment section, click Preferences. Select MATLAB > Help and change the Documentation Location. For more information, see Help Preferences (R2021a).

Documentation: View MATLAB documentation in French, Italian, and German

A subset of MATLAB documentation in French, Italian, and German is available on the web to licensed MATLAB users. Only a subset of the full documentation is available. For more information, see Translated Documentation (R2021a).

MATLAB Drive: Get the location of your MATLAB Drive root folder programmatically

You can get the location of your MATLAB Drive™ root folder programmatically, using the matlabdrive command, from your desktop or in other MATLAB environments such as MATLAB Online. For example, if you have MATLAB Drive Connector installed on your desktop system, MATLAB returns the location of your MATLAB Drive:

md = matlabdrive

md = 
    'C:\Users\username\MATLAB Drive'

MATLAB Drive: Pause and resume syncing in MATLAB Drive Connector

You can temporarily stop the syncing of your MATLAB Drive files, for example, if you are on a metered or slow internet, by pausing and then resuming syncing.

For more information, see Use MATLAB Connector to Manage Your Files.

MATLAB Drive: MATLAB Drive Connector now available in Chinese and Korean

MATLAB Drive Connector now available in Chinese and Korean.

MATLAB Drive: Change the folder permissions for an invited member of a shared folder in MATLAB Drive Online (May 2021)

After inviting someone to access a shared folder, you can change the folder permissions for that person. For example, if you invite someone to share a folder as a read-only folder, you can change the permissions for that person to allow them to edit the contents of the folder.

For more information, see Share Folders Using MATLAB Drive.

MATLAB Drive: Share folder by invitation to others who already have access to the folder through a view-only link (May 2021)

You can now invite others to access a shared folder even if they already have access to the folder through a view-only link. Inviting someone to access a shared folder allows you to give them edit permissions to the folder.

Functionality being removed or changed

PNG images in documentation are compressed

Behavior change

Starting in R2021a, all PNG images included in the documentation are compressed, reducing their stored size on disk. The compressed images should be visually identical to the original images.

Language and Programming

Name=Value Syntax: Use name=value syntax for passing name-value arguments

MATLAB supports a new syntax for passing name-value arguments. In the new syntax, the name and value arguments are connected by an equal sign, and the name is not enclosed in quotes.

Name=value syntax: plot(x,y,LineWidth=2)

Comma-separated syntax: plot(x,y,"LineWidth",2)

MATLAB continues to support the comma-separated name,value syntax. Existing functions and methods support both syntaxes, and the process for writing functions and methods with name-value arguments is unchanged.

Use the new syntax to help identify name-value arguments for functions and to distinguish names from values in lists of name-value arguments. There are some limitations on where and how the name=value syntax can be used:

The recommended practice is to use only one syntax in any given function call. However, if you do mix name=value and name,value syntaxes in a single call, all name=value arguments must appear after the name,value arguments. For example, plot(x,y,"Color","red",LineWidth=2) is a valid combination, but plot(x,y,Color="red","LineWidth",2) errors.
Similarly, name=value arguments must appear after all positional arguments. Calling myFunction(name=value,posArgument) errors.
The name=value syntax can only be used directly in function calls. They cannot be wrapped in a cell array or additional parentheses. Calling myFunction(a,(name=value)) errors.

Retrieving Display Format: format function can get and set display format

The format function can now output the current Command Window display format. Calling format with an output variable returns a DisplayFormatOptions object that describes the current numeric and line spacing formats:

fmt = format

fmt = 
   DisplayFormatOptions with properties:

         NumericFormat: "shortE"
           LineSpacing: "loose"

You can also use a DisplayFormatOptions object as an input to format to change the display format.

Capturing disp Output: Use the formattedDisplayText function to store disp output as a string

formattedDisplayText captures the output of disp(obj) as a string. The function also enables you to control the formatting of the captured string, including numeric format and line spacing. For example, A is a vector containing three logical values. formattedDisplayText displays the vector elements as the words “true” or “false”:

out = formattedDisplayText(A,"UseTrueFalseForLogical",true)

  out = 
        '   true     false    true
      '

Virtual File Storage: mkdir and rmdir will now be able to create and remove files from VFS directories

Starting in R2021a, rmdir and mkdir are able to create folders in remote locations.

Function Argument Validation: Debugger and profiler is now supported

The MATLAB debugger will now be usable inside of the arguments blocks of functions. While debugging an arguments block the workspace is read only. The MATLAB profiler will now record lines inside of arguments blocks.

Class Diagram Viewer: Create graphical class diagrams to explore class details and share designs

Use the Class Diagram Viewer tool to create graphical class diagrams, including details such as:

Inheritance relationships, including abstract classes, mixins, and multiple inheritances
Lists of properties, methods, and events
Class member access levels

You can use the diagrams to explore complex class designs. You can also use the diagrams to share proposed designs with team members, either as static images or editable diagrams.

Use the associated matlab.diagram.ClassViewer class for command line access to diagrams.

Enumeration Comparisons: Use isequal to compare enumeration members with text data types

Enumeration classes now have a default implementation of the isequal method. You can use isequal to compare enumeration members with character vectors, strings, and cell arrays of character vectors or strings. For more information, see isequal Method (R2021a).

eval function: Context checking to resolve identifiers

MATLAB now resolves identifiers like variable names in an eval statement using additional context. For instance, MATLAB recognizes when a call to eval uses a variable declared in a function.

Compatibility Considerations

With the additional context, MATLAB resolves ambiguities differently than in previous releases. Some code now produces errors or warnings.

Code that warns starting in R2021a will error in a future release.

Example Previous Result R2021 Result Notes

Example	Previous Result	R2021 Result	Notes
function myfun(pi) eval('pi') end >> myfun	ans = 3.1416	Errors	MATLAB resolves `pi` as a variable in the function workspace, so `myfun` now errors when `pi` is not defined.
function myfun disp(pi) eval('pi = 1'); end >> myfun	3.1416 pi = 1	Same output but warns on assignment.	MATLAB resolves `pi` as a variable in the function workspace. In a future release, `myfun` errors. This new behavior is consistent with processing the following code: function myfun disp(pi) pi = 1; end
% assignLocal.m script: local = 1; % myfun.m file with local function: function myfun local() assignLocal eval("local") end function local disp("local fx") end % function call: >> myfun	local fx local = 1	Same output but warns on assignment.	In a future release, MATLAB will give precedence as described in Function Precedence Order. In this example, precedence goes to the local function. This new behavior is consistent with processing the following code: function myfun local() assignLocal local() end

function myfun(pi) 
   eval('pi') 
end 

>> myfun

ans =  

3.1416

Errors

MATLAB resolves pi as a variable in the function workspace, so myfun now errors when pi is not defined.

function myfun 
   disp(pi)  
   eval('pi = 1'); 
end 

>> myfun

3.1416 

pi = 

1

Same output but warns on assignment.

MATLAB resolves pi as a variable in the function workspace.

In a future release, myfun errors. This new behavior is consistent with processing the following code:

function myfun 
   disp(pi)  
   pi = 1; 
end

% assignLocal.m script: 
local = 1;  

% myfun.m file with local function: 
function myfun 
   local() 
   assignLocal  
   eval("local") 
end 
function local 
   disp("local fx") 
end 

% function call: 
>> myfun

local fx  

local = 
1

Same output but warns on assignment.

In a future release, MATLAB will give precedence as described in Function Precedence Order. In this example, precedence goes to the local function. This new behavior is consistent with processing the following code:

function myfun 
   local() 
   assignLocal 
   local() 
end

Functionality being removed or changed

format with no arguments is not recommended

Still runs

The format command, by itself, resets the output display format to the default, which is the short, fixed-decimal format for floating-point notation and loose line spacing for all output lines.

format

For clearer code, explicitly specify the default style.

format default

Defining classes and packages: Using schema.m will not be supported in a future release

Still runs

Support for classes and packages defined using schema.m files will be removed in a future release. Replace existing schema-based classes with classes defined using the classdef keyword.

compose does not accept an invalid hexadecimal value, octal value, or trailing backslash

Errors

Previously, when the formatSpec input to compose contained an invalid hexadecimal value, octal value, or trailing backslash it would issue a warning and truncate the output at the point of the invalid value. Starting in R2021a, MATLAB will issue an error instead. With this change, compose will error for all invalid formatSpec inputs.

Using get and set to access or change display format is not recommended

Still runs

Using get and set to programmatically access or change the numeric display format and the display line spacing is not recommended. Use settings instead. For example:

s = settings;
myformat = s.matlab.commandwindow.NumericFormat.ActiveValue

myformat =
    'short'

s.matlab.commandwindow.DisplayLineSpacing.TemporaryValue = 'compact';
myspacing = s.matlab.commandwindow.DisplayLineSpacing.ActiveValue

myspacing =
    'compact'

For more information, see matlab.commandwindow Settings (R2021a).

Using feature('EightyColumns') to access and change Command Window display width is not recommended

Still runs

Using the command feature('EightyColumns') or feature('EightyColumns',value) to programmatically determine or change whether the Command Window display width limit is enabled is not recommended. Use settings instead. For example:

s = settings;
s.matlab.commandwindow.UseEightyColumnDisplayWidth.TemporaryValue = 1;
limitwidth = s.matlab.commandwindow.UseEightyColumnDisplayWidth.ActiveValue

limitwidth =
  logical
   0

For more information, see matlab.commandwindow Settings (R2021a).

Data Analysis

Data Preprocessing Live Editor Tasks: Operate on multiple table variables and specify output format for table input

When you are working with data in a table or timetable, these Live Editor tasks now allow you to operate on multiple table variables at the same time:

You can also choose which variable to display when visualizing the results.

In addition, tasks that modify variables provide new output options. You can return a table with all of the variables, or with only the variables that were modified. For tasks that return logical arrays, you can specify the size of the output. The size can match the size of the input table or a table containing only the variables that were used in the calculation.

Clean Outlier Data Live Editor Task: Visualize results with a histogram

The Clean Outlier Data (R2021a) Live Editor task now offers histogram plots for most detection methods. The histogram can summarize the input data, the outliers, the cleaned data with the outliers filled, and the outlier detection thresholds and center value.

fillmissing Function: Specify custom fill method

You can now specify a custom method for filling missing values when using the fillmissing (R2021a) function. Specify the custom method as a function handle.

normalize Function: Normalize multiple data sets with same parameters

normalize (R2021a) can now return the centering and scaling parameter values used to perform the normalization. You can reuse these parameters to normalize subsequent data sets in the same manner. For example, you can normalize an array of data A and then normalize a second array B with the same parameters:

[Anorm,C,S] = normalize(A);
Bnorm = normalize(B,'center',C,'scale',S);

The new outputs C and S contain the centering and scaling parameter values, respectively. So that you can easily reuse them in a later normalization step, you can also now specify the 'center' and 'scale' normalization methods at the same time. These are the only two normalization methods that you can specify together.

To further support these changes, when method is 'center' or 'scale', the possible values of methodtype now include arrays and tables. While these methodtype values are intended to work with the new outputs C and S, you also can compute your own normalization parameters to specify.

groupcounts Function: Display percentages of group counts

groupcounts (R2021a) now displays information about the percentage each group count represents.

For table and timetable inputs, groupcounts automatically displays the percentages represented by each group count in the output table.
For array inputs, groupcounts has a new third output argument to return the percentages represented by each group count.

Compatibility Considerations

When groupcounts operates on data in a table or timetable, the output contains an additional table variable for the percentages. The percentages are in the range [0 100] and are included in the table variable Percent.

Any code that references specific table variables is unaffected. However, you might need to update code that depends on the number of variables in the output table.

ts2timetable Function: Convert timeseries objects to timetables

To convert timeseries objects to timetables, use the ts2timetable (R2021a) function.

table and timetable Functions: Specify dimension names using the 'DimensionNames' name-value argument

When you create tables and timetables, you can specify their dimension names by using the 'DimensionNames' name-value argument with these functions:

array2table (R2021a)
array2timetable (R2021a)
cell2table (R2021a)
struct2table (R2021a)
table (R2021a)
timetable (R2021a)

In previous releases, you could change the dimension names only by creating a table or timetable and then changing its DimensionNames property.

Functionality being removed or changed

Table dimension names cannot match reserved names

Behavior change

MATLAB raises an error if you assign a dimension name that matches one of these reserved names: 'Properties', 'RowNames', 'VariableNames', or ':'. In previous releases, MATLAB raised a warning and modified the dimension names so that they were different from the reserved names.

For example, if you create a table and then assign 'Properties' as a dimension name, the result is an error.

T = array2table(magic(3));
T.Properties.DimensionNames = {'Rows','Properties'}

'SamplingRate' will be removed

Warns

The 'SamplingRate' name-value argument will be removed in a future release. Use 'SampleRate' instead. The corresponding timetable property is also named SampleRate.

For backward compatibility, you still can specify 'SamplingRate' as the name of the name-value argument. However, the value is assigned to the SampleRate property.

This change in behavior affects the timetable functions shown in the list:

array2timetable (R2021a)
retime (R2021a)
synchronize (R2021a)
table2timetable (R2021a)
timetable (R2021a)

Data Import and Export

XML Files: Read, write, and import XML files using readtable, readtimetable, and other functions

The readtable, writetable, readtimetable, writetimetable, and detectImportOptions functions now support reading and writing XML files. This list outlines the added capabilities of each function.

readtable (R2021a) and readtimetable (R2021a) — Read XML data into MATLAB as a table or timetable. You can specify optional name-value arguments to control how readtable and readtimetable treat XML data. For example, specify ‘ImportAttributes’,false to ignore attribute nodes.
writetable (R2021a) and writetimetable (R2021a) — Write a table or timetable in MATLAB to an XML file. Specify optional name-value arguments to control how writetable and writetimetable treat XML data. For example, specify ‘AttributeSuffix','_att' to specify that all table or timetable variables with the suffix '_att' should be written as attributes in the output XML file.
detectImportOptions (R2021a) — The detectImportOptions function now returns an XMLImportOptions object when you call it on an XML file. Its behavior when you call it on other file types has not changed. Use the XMLImportOptions object with readtable to customize import options. For instance:
- Import only a subset of data using the SelectedVariableNames property.
- Specify the names and data types of the variables in the input file using the VariableNames and the VariableTypes properties.
- Manage the import of specific nodes in the XML file using name-value arguments such as 'TableSelector', 'RowSelector', or 'VariableSelectors'.

For more information, see XMLImportOptions.

MATLAB API for Advanced XML Processing: Create, read, write, transform, and query XML

Use the MATLAB API for XML Processing (MAXP) to develop advanced applications that create, read, write, transform, and query XML documents. MAXP consists of these packages:

matlab.io.xml.dom (R2021a) — Classes for creating, reading, and writing XML files and strings following the W3C DOM standard.
matlab.io.xml.transform (R2021a) — Classes for transforming XML documents from one type to another following the XSLT 1.0 standard.
matlab.io.xml.xpath (R2021a) — Classes for querying XML documents using XPath 1.0 expressions.

XML Files: Register XML namespace prefixes for evaluating XPath expressions using readtable,readstruct, and other functions

Use the RegisteredNamespaces name-value argument to specify namespace prefixes that readtable (R2021a), readtimetable (R2021a),readstruct (R2021a), XMLImportOptions (R2021a), and detectImportOptions (R2021a) use when evaluating XPath expressions in an XML file. RegisteredNamespaces can be used when you also evaluate an XPath expression specified by a selector name-value argument, such as StructSelector for readstruct, or VariableSelectors for readtable and readtimetable.

The readstruct function automatically detects namespace prefixes to register for use in XPath evaluation, but you can also register new namespace prefixes using the RegisteredNamespaces name-value argument. You might register a new namespace prefix when an XML node has a namespace URL, but no declared namespace prefix in the XML file. In that case, you can specify RegisteredNamespaces as a string array containing a namespace prefix and the associated URL.

For example, evaluate an XPath expression on an XML file named example.xml which does not contain a namespace prefix declaration. Specify 'RegisteredNamespaces' as [“myprefix”, “https://www.mathworks.com”] to assign the prefix myprefix to the URL https://www.mathworks.com.

data = readstruct("example.xml", "StructSelector", "/myprefix:Data",...
 "RegisteredNamespaces", [“myprefix”, “https://www.mathworks.com”])

In the resulting structure, the namespace prefix and URL will appear as attributes belonging to the element specified in the StructSelector name-value argument.

Low-level file I/O functions and remote data: Perform read and write operations on remotely stored files

You can now use low-level file I/O functions, such as fopen (R2021a), fread (R2021a), and fwrite (R2021a), to work with files stored in remote locations. Some supported remote locations include Amazon S3™ and Windows Azure^® Blob Storage.

When reading from or writing to a remote location, you must specify the full path using a uniform resource locator (URL). For example, open a binary file from the Amazon S3 cloud.

fid = fopen(“s3://bucketname/path_to_file/example.bin”);

For more information on setting up MATLAB to access your online storage service, see Work with Remote Data (R2021a).

save and load functions and remote data: Save, load, and append data to remotely stored v7.3 MAT-files

You can now access v7.3 MAT-files stored in remote locations, such as Amazon S3 and Windows Azure Blob Storage, using the save (R2021a) and load (R2021a) functions.

When saving, loading, or appending data to a remote location, you must specify the full path using a uniform resource locator (URL). For example, load a MAT-file from the Amazon S3 cloud.

load(“s3:://bucketname/path_to_file/example.mat”);

For more information on setting up MATLAB to access your online storage service, see Work with Remote Data (R2021a).

Reading Online Data: Read files over HTTP and HTTPS using readtable, audioread, and other reading functions

Read files from an internet URL by specifying filename as a string that contains the protocol type 'http://' or 'https://'. This lets you read data from their primary online sources.

This functionality is supported by these functions: audioread, audioinfo, parquetread, parquetinfo, readtable, readtimetable, readvars, readstruct, readmatrix, readcell, readlines, and detectImportOptions.

Parquet Data Format: Use categorical data in parquet data format

Read, write, and analyze parquet data that contain the categorical data type.

This functionality is supported by these functions: parquetread, parquetwrite, parquetinfo, and parquetDatastore.

Datastores: Read all data from a datastore using parallel processing

You can use parallel processing when reading all data from a datastore (requires Parallel Computing Toolbox™). Parallel processing results in improved performance when reading data, especially with remote data.

Data Compression Functions: Improved functionality in zip/unzip and tar/untar

On Windows, macOS, and Linux^® systems:

zip can compress an individual file of any size.
zip can compress any number of files in a single function call.
tar can compress a group of files of any cumulative size.

Additionally, on Windows systems, unzip and untar replace invalid characters with underscores if they occur in entry path names of the original file.

imfinfo function: Get information about all Adobe Digital Negative (DNG) file tags

The imfinfo (R2021a) function returns information on all DNG file tags as individual named fields in the output structure. For a complete list of DNG file tags, see Chapter 4 of the Adobe^® Digital Negative (DNG) Specification.

jsonencode: Add indentation to JSON text

Use the jsonencode (R2021a) 'PrettyPrint' option to display JSON text with an indentation of two spaces.

s.Width = 800;
s.Height = 600;
s.Title = 'View from the 15th Floor';
s.Animated = false;
s.IDs = [116, 943, 234, 38793];
jsonencode(s,'PrettyPrint',true)

ans = 
    '{
       "Width": 800,
       "Height": 600,
       "Title": "View from the 15th Floor",
       "Animated": false,
       "IDs": [
         116,
         943,
         234,
         38793
       ]
     }'

Functionality being removed or changed

The H5I.get_name function only accepts named HDF5 datatypes as input arguments.

Behavior change

Starting in R2020a, the H5I.get_name function only accepts committed (previously called named) HDF5 datatypes as input arguments, and will error if you pass other datatypes as input. In releases R2019b and earlier, H5I.get_name does not error if you pass other datatypes as input.

To verify that the input is a committed HDF5 datatype, call the H5T.committed function on it. The H5T.committed function returns a value of 1 if the input is a committed HDF5 datatype, and a value of 0 if it is not.

imfinfo now returns Adobe DNG tags belonging to versions 1.2 through 1.5 in individual named fields in the output structure

Behavior change

When you call the imfinfo function on an Adobe DNG file, it now returns tags belonging to versions 1.2 through 1.5 as individual named fields in the output structure. Previously, tags belonging to these versions were stored in the 'UnknownTags' field of the output structure. For a complete list of DNG file tags, see Chapter 4 of the Adobe Digital Negative (DNG) Specification.

imread reads the first frame in a GIF file by default

Behavior change

Starting in R2021a, when you read a GIF file without specifying additional arguments, the imread function reads the first frame by default. Previously, imread read all the frames in the file by default.

To read all the frames in the order that they appear in the GIF file, specify the value of the 'Frames' name-value argument as 'all'.

serial function will be removed

Still runs

serial and its object properties will be removed. Previously, serial and its object properties were not recommended. Use serialport (R2021a) and its properties instead.

This example shows how to connect to a serial port device using the recommended functionality.

Functionality	Use This Instead
s = serial("COM1"); s.BaudRate = 115200; fopen(s)	s = serialport("COM1",115200);

See Transition Your Code to serialport Interface (R2021a) for more information about using the recommended functionality.

Mathematics

Graph Algorithms: Compute all paths, all cycles, and cycle basis

graph (R2021a) and digraph (R2021a) objects have new functions to compute paths and cycles:

allpaths (R2021a) — Compute all paths between two nodes in a graph or digraph object.
allcycles (R2021a) — Compute all cycles in a graph or digraph object.
cyclebasis (R2021a) — Compute the fundamental cycle basis of a graph object.
hascycles (R2021a) — Determine whether a graph or digraph object contains cycles.

griddedInterpolant Object: Use multivalued interpolation to interpolate multiple data sets simultaneously

griddedInterpolant (R2021a) can now interpolate multiple data sets on the same grid at the same query points. For example, if you specify a 2-D grid, a 3-D array of values at the grid points, and a 2-D collection of query points, then griddedInterpolant returns the interpolated values at the query points for each 2-D page in the 3-D array of values.

Previously, this functionality was available in interp1 for 1-D interpolation, but this improvement to griddedInterpolant adds support for N-D multivalued interpolation.

eig Function: Improved algorithm for skew-Hermitian matrices

eig (R2021a) now has an improved algorithm for input matrices that are skew-Hermitian. With the function call [V,D] = eig(A), where A is skew-Hermitian, eig now guarantees that the matrix of eigenvectors V is unitary and the diagonal matrix of eigenvalues D is purely imaginary.

cdf2rdf Function: Improved algorithm for all inputs

cdf2rdf (R2021a) has an improved algorithm for all input matrices that reduces floating-point round-off errors in the calculation.

Functionality being removed or changed

Line Continuation: Ellipsis following operator treated as a space

Behavior change

Previously, if an ellipsis followed an operator inside of a matrix or cell array, it caused the operator to be treated as unary. Ellipses will now be treated as a space in all cases.

Old Behavior	New Behavior
x = [1 -... 2] x = 1 -2 Previously, this code was equivalent to the expression, `x = [1 -2]`.	x = [1 -... 2] x = -1 Now, the ellipsis will be treated as a space so this code is equivalent to the expression, `x = [1 - 2]`.

Graphics

Create Plot Live Editor Task: Create plots interactively and generate code

The Create Plot (R2021a) Live Editor Task makes generating and exploring visualizations for data simple and interactive. With this task you can select the data you wish to visualize and choose the plot type that best represents that data. Alternatively, you can explore the visualizations available in MATLAB to find the desired plot type and add your data. This task creates labels for the visualization based on the data and can be used to add or adjust the optional parameters of the visualization.

bubblecloud Function: Visualize part-to-whole relationships

Use the bubblecloud (R2021a) function to illustrate the relationship between elements in your data set and the set as a whole. For example, you can visualize data collected from different cities, and represent each city as a bubble whose size is proportional to the value for that city.

A bubble cloud chart.

tiledlayout Function: Control the tile indexing scheme

Control whether the tile indices increase across the rows or down the columns of a layout by setting the TileIndexing (R2021a) property of a TiledChartLayout object. Select one of the following options:

'rowmajor' — Increment the tile indices across the first row before moving to the next row. This is the default behavior.
'columnmajor' — Increment the indices down the first column before moving to the next column. This indexing scheme is the same as linear indexing for arrays.

The nexttile function populates tiles according to the indexing scheme. If you change the tile indexing of a populated layout, the tile positions change to match the new scheme.

PolarAxes Objects: Use the CurrentPoint property or call ginput to get the cursor location within polar axes

Query the CurrentPoint (R2021a) property of a PolarAxes object to get the location of the last click within the axes. The ginput (R2021a) function also supports querying coordinates of polar axes.

Scatter Plots and Constant Lines: Create multiple scatter plots or constant lines at once

The scatter (R2021a), polarscatter (R2021a), and swarmchart (R2021a) functions now accept the same combinations of matrices and vectors as the plot (R2021a) function does. As a result, you can visualize multiple data sets at once rather using the hold function between plotting commands.
The xline (R2021a) and yline (R2021a) functions now accept vectors of values for creating multiple vertical or horizontal reference lines. You can also specify separate labels for each line using a string array or a cell array.

Scatter plot of multiple data sets.

Axis Limits: Define LimitsChangedFcn callback that executes when the limits of an axis change

Define the LimitsChangedFcn (R2021a) callback function on any type of ruler object such as a numeric ruler. The callback function executes when the limits of the corresponding axis change. For example, you can define the callback in an app to update another aspect of the UI when the user pans within the axes.

Axis Limits: Control axis limits

Control the axis limits for your plots by setting the XLimitMethod, YLimitMethod, or ZLimitMethod on the axes. Select one of the following property values:

'tickaligned' — Align the edges of the axes box with the tick marks that are closest to your data without excluding any data. This is the default option.
'tight' — Fit the axes box tightly around the data by setting the axis limits equal to the range of the data.
'padded' — Fit the axes box around the data with a thin margin of padding on each side. The width of the margin is approximately 7% of your data range.

Plots that use the 'tickaligned', 'tight', and 'padded' limit methods.

exportgraphics and copygraphics Functions: Specify RGB, CMYK, or grayscale output

Choose a color space when exporting or copying graphics. Specify the Colorspace name-value argument when you call the exportgraphics (R2021a) or copygraphics (R2021a) functions. Select one of the following options:

'rgb' — Capture truecolor RGB content. This is the default color space.
'gray' — Convert the content to grayscale.
'cmyk' (exportgraphics only) — Convert the content to cyan, magenta, yellow, and black (CMYK) before exporting the content as an EPS file.

colororder Function: Control colors in stacked plots

The colororder (R2021a) function now supports charts created with the stackedplot (R2021a) function.

Tick Labels: Automatically rotate tick labels

When you manually specify the ticks or the tick labels for a chart, the tick labels automatically rotate to give the best possible presentation given the size of the figure and the number of the tick labels.

A bar chart that shown at two different sizes with different tick angles.

patch and errorbar Functions: Expanded data type support

The patch (R2021a) and errorbar (R2021a) functions now support more data types:

The patch function accepts numeric, datetime, duration, and categorical values for the x-, y-, and z-coordinates.
The errorbar function accepts numeric, datetime, duration, and categorical values for the x- and y-coordinates. It also accepts numeric and duration values for the error bar lengths above, below, and on either side of the data points.

Geographic Plots: Access basemaps using additional proxy server authentication types

You can now access basemaps for geographic axes and charts using additional proxy server authentication types.

On Windows, you can use Basic, Digest, NTLM, Negotiate (SPNEGO), and Kerberos authentication.
On Linux and macOS, you can use Basic, Digest, and NTLM authentication.

Prior to R2021a, geographic axes and charts supported only types without authentication or with Basic authentication. For more information about specifying proxy server settings, see Use MATLAB Web Preferences For Proxy Server Settings (R2021a).

Functionality being removed or changed

Tile spacing and padding options for tiled chart layouts have changed

Behavior change

When you create a tiled chart layout, some of the TileSpacing and Padding properties provide a different result or have new names.

The new TileSpacing options are 'loose', 'compact', 'tight', and 'none'. The new Padding options are 'loose', 'compact', and 'tight'. The following tables describe how the previous options relate to the new options.

TileSpacing Changes

Previous TileSpacing Option R2021a TileSpacing Option How to Update Your Code

Previous `TileSpacing` Option	R2021a `TileSpacing` Option	How to Update Your Code
`'normal'`	`'loose'`	Consider changing instances of `'normal'` to `'loose'`. The `'normal'` option is no longer recommended, but it continues to work, and there are no plans to remove it.
`'compact'`	`'compact'`	No changes needed.
Not Applicable	`'tight'`	`'tight'` is a new option. It provides the same result as `'none'` does in previous releases.
`'none'`	`'none'`	The `'none'` option removes all the space between adjacent plot boxes, and the tick labels overlap with neighboring plot boxes. To preserve the spacing between the plot boxes, change instances of `'none'` to `'tight'`.

'normal'

Tiled chart layout with 'normal' tile spacing in R2020b and earlier releases.

'loose'

Tiled chart layout with 'loose' tile spacing in R2021a.

Consider changing instances of 'normal' to 'loose'.

The 'normal' option is no longer recommended, but it continues to work, and there are no plans to remove it.

'compact'

Tiled chart layout with 'compact' tile spacing.

'compact'

Tiled chart layout with 'compact' tile spacing.

No changes needed.

Not Applicable

'tight'

Tiled chart layout with 'tight' tile spacing in R2021a.

'tight' is a new option. It provides the same result as 'none' does in previous releases.

'none'

Tiled chart layout with 'none' tile spacing in R2020b.

'none'

Tiled chart layout with 'none' tile spacing in R2021a.

The 'none' option removes all the space between adjacent plot boxes, and the tick labels overlap with neighboring plot boxes.

To preserve the spacing between the plot boxes, change instances of 'none' to 'tight'.

Padding Changes

Previous Padding Option R2021a Padding Option How to Update Your Code

Previous `Padding` Option	R2021a `Padding` Option	How to Update Your Code
`'normal'`	`'loose'`	Consider changing instances of `'normal'` to `'loose'`. The `'normal'` option is no longer recommended, but it continues to work, and there are no plans to remove it.
`'compact'`	`'compact'`	No changes needed.
`'none'`	`'tight'`	Consider changing instances of `'none'` to `'tight'`. The `'none'` option is no longer recommended, but it continues to work, and there are no plans to remove it.

'normal'

Tiled chart layout with 'normal' padding in R2020b.

'loose'

Tiled chart layout with 'loose' padding in R2021a.

Consider changing instances of 'normal' to 'loose'.

The 'normal' option is no longer recommended, but it continues to work, and there are no plans to remove it.

'compact'

Tiled chart layout with 'compact' padding.

'compact'

Tiled chart layout with 'compact' padding.

No changes needed.

'none'

Tiled chart layout with 'none' padding in R2020b.

'tight'

Tiled chart layout with 'tight' padding in R2021a.

Consider changing instances of 'none' to 'tight'.

The 'none' option is no longer recommended, but it continues to work, and there are no plans to remove it.

Passing an empty label to the legend function omits the entry from the legend

Behavior change

When you call the legend (R2021a) function and specify a label as an empty character vector, an empty string, or an empty element in a cell array or string array, the corresponding entry is omitted from the legend. In R2020b and earlier releases, the entry appears in the legend without a label.

For example, this code plots two sine waves and a reference line at y=0. Then it creates a legend with three labels, where the last label is empty. In R2020b, the third line appears in the legend without a label. In R2021a, the third line is omitted from the legend.

x = 0:0.2:10;
plot(x,sin(x),x,sin(x+1));
hold on
yline(0,'--')
legend('sin(x)','sin(x+1)','')

Comparison of empty legend labels in R2020b and R2021a.

To keep an entry in the legend without a label, include a space character in the label. For example, to update the preceding code, specify the last label as a character vector containing a space (' ').

legend('sin(x)','sin(x+1)',' ')

Alternatively, if you do not want to display a space character, you can pass the individual line objects to the legend function with an array of labels. To get the individual line objects, call each plotting function with an output argument.

x = 0:0.2:10;
p = plot(x,sin(x),x,sin(x+1));
hold on
line0 = yline(0,'--');
legend([p(1) p(2) line0], {'sin(x)','sin(x+1)',''});

The XData, YData, and ZData properties on Patch objects created with the fill and fill3 functions return values of the original data type

Behavior change

The XData, YData, and ZData properties on a Patch object created by the fill (R2021a) or fill3 (R2021a) functions return the coordinates using the original input data type, rather than returning them as double values.

In previous releases, datetime, duration, and categorical coordinates are converted to double values when they are stored in the XData, YData, and ZData properties.

For example, this code creates a filled polygon with datetime x-coordinates. Then it calculates x2 using the values stored in the XData property. In R2020b, h.XData and x2 are double arrays. In R2021a, h.XData and x2 are datetime arrays.

x = datetime('01-Jan-2018') + days([0 1 1 0]);
y = [0 0 1 1];
h = fill(x,y,'red');
x2 = h.XData + 1;

To preserve the double values in your code, get the double values from the Vertices property of the Patch object. The x-, y-, and z-coordinates are stored as double values in the first, second, and third columns of the Vertices array.

x2 = h.Vertices(:,1) + 1;

Alternatively, use the ruler2num (R2021a) function. Pass the coordinate values and the corresponding axis ruler to the ruler2num function.

ax = gca;
x2 = ruler2num(h.XData,ax.XAxis) + 1;

Graphics objects and UI components sized using 'points', 'inches', and 'centimeters' units will increase in size on macOS platforms

Behavior change in future release

In a future release, graphics and UI objects that have Units or FontUnits properties set to 'points', 'inches', or 'centimeters' will use a conversion value of 1 pixel = 1/96th inch on macOS platforms. The current conversion value is 1 pixel = 1/72nd inch. As a result, these objects and text elements will display 1.33 times larger than their previous size. This change will provide a more readable default font size and will ensure a consistent object size across Windows and macOS platforms.

The following objects use a unit value of 'pixels' by default and will not be affected by this change:

UI components in App Designer or in apps created with the uifigure (R2021a) function
UI components in apps created in GUIDE and migrated to App Designer using the GUIDE to App Designer Migration Tool for MATLAB
Axes objects created using the uiaxes (R2021a) function

For the most control when sizing and laying out your graphics objects and UI components, use a value of 'pixels' for Units and FontUnits properties. To maintain object sizes in current and future releases, make these updates in your code:

Objects with Units or FontUnits set to 'points' — Update the value of the property from 'points' to 'pixels'.
Objects with Units or FontUnits set to 'inches' — Update the value of the property from 'inches' to 'pixels' and multiply all Position values by 72.
Objects with Units or FontUnits set to 'centimeters' — Update the value of the property from 'centimeters' to 'pixels' and multiply all Position values by 72/2.54.

For example, this code creates a push button in a figure window, with its position specified in inches:

uicontrol('Units','inches','Position',[0.6 0.1 1.75 0.5]);

In a future release, the push button created by this code will display 1.33 times larger on macOS platforms. To maintain the size and position of the push button in current and future releases, update the code to:

uicontrol('Units','pixels','Position',[50 10 126 36]);

Align Distribute Tool will be removed in a future release

Warns

The Align Distribute Tool will be removed in a future release.

To control the arrangement of multiple plots in a figure, create a tiled chart layout using the tiledlayout function instead.

To align or distribute graphics objects within a figure, select Tools > Align or Tools > Distribute from the figure toolbar instead.

App Building

uihyperlink Function: Add and configure clickable links in apps and on the App Designer canvas

To create a link, call the uihyperlink (R2021a) function or, in App Designer, drag a hyperlink UI component from the Component Library onto the canvas.

Hyperlink UI components are supported only in App Designer apps and in figures created with the uifigure (R2021a) function.

uitree Function: Add and configure check box trees in apps and on the App Designer canvas

A check box tree is a styled with check boxes to the left of every item. You can now create check box trees in apps and on the App Designer canvas. Check box trees allow for easier selection of multiple tree nodes.

In apps created programmatically with the uifigure (R2021a) function, create a check box tree using the uitree (R2021a) function by specifying the style 'checkbox'.

In App Designer, create a check box tree by dragging it from the Component Library onto the canvas.

Interpreter Property: Style text and display equations in labels with HTML and LaTeX markup

Use the Interpreter property on Label objects (created with the uilabel (R2021a) function) to enable markup in the label text. To add HTML markup, set the Interpreter property to 'html'. To add LaTeX markup, set the Interpreter property to 'latex'.

For more information, see Label Properties (R2021a).

WindowStyle Property: Create UI figures that remain in the foreground

To keep a specific UI figure window in front of other windows, set the WindowStyle property to 'alwaysontop'. Unlike modal figures, UI figure windows with this property setting do not restrict keyboard and mouse interactions.

The 'alwaysontop' property value is available only in figures created using the uifigure (R2021a) function.

For more information, see UI Figure Properties (R2021a).

scroll Function: Scroll to a location within a table UI component programmatically

Use the scroll (R2021a) function to scroll within a table UI component programmatically. Specify the scroll location as the top, bottom, left, or right side of the table, or as a specific row, column, or table cell.

UI Component Accessibility: Select ListBox items, Table cells, ColorPicker colors, and DatePicker menus using the keyboard

You can now use keyboard shortcuts to change the focus and make selections in various UI components. These shortcuts are supported for components in figures created using the uifigure (R2021a) function.

Component	Action	Keyboard Shortcut
`ListBox` with `Multiselect` property set to `'off'`	Toggle list box selection.	Space
`ListBox` with `Multiselect` property set to `'on'`	Select all items.	Ctrl+A
	Move focus to previous item and toggle selection.	Shift+Up
	Move focus to next item and toggle selection.	Shift+Down
	Move focus to previous item without toggling selection.	Ctrl+Up
	Move focus to next item without toggling selection.	Ctrl+Down
	Toggle selection of item currently in focus.	Ctrl+Space
`Table`	Select table cell in top-left corner.	Ctrl+Home
`Table`	Select table cell in bottom-right corner.	Ctrl+End
`ColorPicker` gradient selector	Move gradient selector.	Arrow keys
`ColorPicker` hue slider	Move hue slider.	Up and Down
`DatePicker`	Cycle between drop down menus, buttons, and calendar.	Tab

App Designer: Use custom UI components in App Designer

You can now configure custom UI component classes to appear in the App Designer Component Library and to be used interactively in Design View.

To configure a custom UI component class for use in App Designer, follow these steps:

Define your own UI component class by creating a subclass of the matlab.ui.componentcontainer.ComponentContainer (R2021a) base class.
Call the appdesigner.customcomponent.configureMetadata (R2021a) function and specify the path to the component class .m file.
Use the resulting dialog box to configure the metadata associated with the component, including the component name, icon, author, and category.

This creates a resources folder with the App Designer metadata. To view the component in the App Designer Component Library, add the folder containing the component class file and the generated resources folder to the MATLAB path.

To share your custom UI component for others to use in App Designer, share both the component class file and the resources folder.

For more information, see Configure Custom UI Components for App Designer (R2021a).

App Designer: Zoom and pan in the canvas, and zoom in the Code View editor

To zoom in or out in the App Designer canvas and in the Code View editor, hold Ctrl and move the scroll wheel, or press Ctrl+Plus and Ctrl+Minus. To return to the default scale, press Ctrl+Alt+0. Zooming does not affect the Component Library, Component Browser, or Code Browser.

To pan in the App Designer canvas, use one of the following:

Click and drag with the middle mouse button.
Hold Space while clicking and dragging with the left mouse button.

Compatibility Considerations

In previous releases, the Ctrl+Plus and Ctrl+Minus keyboard shortcuts zoomed the entire App Designer desktop.

App Designer: Control color and tab settings in Code View using MATLAB preferences

Color and tab preferences applied to the MATLAB Editor are now also applied to the App Designer Editor.

Additionally, you can now change the background color of the App Designer read-only code. Access this setting in the App Designer tab of MATLAB preferences.

For more information, see Personalize Code View Appearance (R2021a).

App Designer: Customize split-screen layouts in the App Designer editor

To view your document in horizontal or vertical split-screen mode in the App Designer Code View editor, select a layout in the App Designer toolstrip.

App Testing Framework: Perform gestures on panels and tables

The app testing framework supports gestures on more UI components:

Perform press (R2021a), hover (R2021a), and chooseContextMenu (R2021a) gestures on panels.
Perform choose (R2021a), type (R2021a), and chooseContextMenu (R2021a) gestures on table UI components.

App Testing Framework: Close alert dialog box in front of figure window

You can now use the dismissAlertDialog (R2021a) method as part of a test case to programmatically close an alert dialog box in front of a figure window. For example, create a modal alert dialog box and close it by calling the method.

fig = uifigure;
uialert(fig,'File not found','Invalid File')

tc = matlab.uitest.TestCase.forInteractiveUse;
tc.dismissAlertDialog(fig)

Web Apps and Standalone Applications: Datatips supported in graphics

Graphics created in web apps and standalone applications support datatips. Use datatips in these applications just as you would in MATLAB figures.

Functionality Being Removed or Changed

GUIDE will be removed in a future release

Warns

The GUIDE environment and the guide (R2021a) function will be removed in a future release.

After GUIDE is removed, existing GUIDE apps will continue to run in MATLAB but will not be editable using the drag-and-drop GUIDE environment. To continue editing an existing GUIDE app and help maintain its compatibility with future MATLAB releases, use one of the suggested migration strategies listed in the table.

App Development	Migration Strategy	How to Migrate
Frequent or ongoing development	Migrate your app to App Designer.	Use the GUIDE to App Designer Migration Tool for MATLAB on `mathworks.com`.
Minimal or occasional editing	Export your app to a single MATLAB file to manage your app layout and code using MATLAB functions.	Open the app in GUIDE and select File > Export to MATLAB-file.

To create new apps, use App Designer and the appdesigner (R2021a) function instead. App Designer is the recommended app development environment in MATLAB.

To learn more about migrating apps, see GUIDE Migration Strategies (R2021a).

For more information about App Designer, go to Comparing GUIDE and App Designer on mathworks.com.

GUIDE templates have been removed

All GUIDE templates other than the blank GUI have been removed. To create new apps interactively, use App Designer and the appdesigner (R2021a) function instead.

Graphics objects and UI components sized using 'points', 'inches', and 'centimeters' units will increase in size on macOS platforms

Behavior change in future release

The following objects use a unit value of 'pixels' by default and will not be affected by this change:

UI components in App Designer or in apps created with the uifigure (R2021a) function
UI components in apps created in GUIDE and migrated to App Designer using the GUIDE to App Designer Migration Tool for MATLAB
Axes objects created using the uiaxes (R2021a) function

Objects with Units or FontUnits set to 'points' — Update the value of the property from 'points' to 'pixels'.
Objects with Units or FontUnits set to 'inches' — Update the value of the property from 'inches' to 'pixels' and multiply all Position values by 72.
Objects with Units or FontUnits set to 'centimeters' — Update the value of the property from 'centimeters' to 'pixels' and multiply all Position values by 72/2.54.

For example, this code creates a push button in a figure window, with its position specified in inches:

uicontrol('Units','inches','Position',[0.6 0.1 1.75 0.5]);

uicontrol('Units','pixels','Position',[50 10 126 36]);

Performance

Sparse Matrix Multiplication: Improved performance multiplying large sparse matrices

Matrix multiplication performance has improved when multiplying sparse matrices. The performance improvement arises from added support for multithreading in the operation, and therefore the speedup gets better as the matrix size and number of nonzeros increase.

For example, if you multiply two 1e5-by-1e5 random sparse matrices with approximately two million nonzeros, performance in R2021a is about 4.4x faster than in R2020b on a machine with 6 physical cores.

function timingSparseMult
rng default
A = sprand(1e5,1e5,0.0002);
B = sprand(1e5,1e5,0.0002);
tic
C = A*B;
toc
end

The approximate execution times are:

R2020b: 2.2 s

R2021a: 0.5 s

The code was timed on a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system by calling the function timingSparseMult.

Sparse Linear Systems: Improved performance solving sparse linear systems A*X = B with multicolumn B

Solving a linear system of the form A*X = B by executing X = A\B shows improved performance when A is a sparse square matrix and B is a matrix with two or more columns. The speedup applies to the solving step of the calculation but not the factorization step. The performance improvement arises from added support for multithreading, and therefore the speedup gets better as the number of columns in B increases.

For example, if you solve A*X = B using a 1e4-by-1e4 sparse coefficient matrix with approximately 40,000 nonzeros and a B matrix with 100 columns, performance in R2021a is about 5x faster than in R2020b on a machine with 6 physical cores. This code uses decomposition to factor the coefficient matrix, so only the solving process is timed. If you use X = A\B instead, you still see a speedup, but the time required to factor the matrix is included and has not changed.

function timingSparseBackslashMultRHS
rng default
A = sprand(1e4,1e4,0.0003) + speye(1e4);
B = sprand(1e4,100,0.002);
dA = decomposition(A);
tic
x = dA\B;
toc
end

The approximate execution times are:

R2020b: 1.5 s

R2021a: 0.3 s

The code was timed on a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system by calling the function timingSparseBackslashMultRHS.

vecnorm Function: Improved performance operating on data with multiple columns

The performance of the vecnorm function has improved for all norm types when the data has 16 or more columns and at least 2¹⁷ elements.

The improvement also applies to N-D array data that can be permuted into a matrix with the requisite number of columns. The speedup varies depending on the type of norm being calculated.

For example, if you calculate the 2-norm of a 1000-by-1000-by-3 array along the third dimension, performance in R2021a is about 7.3x faster than in R2020b.

function timingVecnorm
rng default
N = 1000;
A = rand(N,N,3);
for k = 1:200
  D = vecnorm(A,2,3);
end
end

The approximate execution times are:

R2020b: 8.8 s

R2021a: 1.2 s

The code was timed on a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system using the timeit function:

timeit(@timingVecnorm)

ismember Function: Improved performance for cell inputs

The ismember function shows improved performance operating on cell inputs. The speedup depends on the size and layout of the data, with the largest speedup when the input has many cells that contain few elements in each cell.

For example, if you use ismember to compare two 1000-by-1 cell arrays with 10 elements in each cell, performance in R2021a is about 4.7x faster than in R2020b.

function timingIsmember
a = num2cell(char(randi(127,[1000 10])),2);
b = num2cell(char(randi(127,[1000 10])),2);
tic
for ii = 1:1e4
  ismember(a,b); 
end
toc
end

The approximate execution times are:

R2020b: 6.6 s

R2021a: 1.4 s

The code was timed on a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system by calling the function timingIsmember.

unique Function: Improved performance for numeric, logical, char, and cell inputs

The unique function shows improved performance operating on numeric, logical, char, and cell inputs. The speedup generally gets better as the size of the inputs increases, and the improvement applies when using any optional flags except the 'legacy' flag.

For example, if you operate on a 100-by-1 cell array with 10 elements in each cell, performance in R2021a is about 3.3x faster than in R2020b.

function timingUniqueCell
a = num2cell(char(randi(127,[100 10])),2);
tic
for ii = 1:1e5
  b = unique(a); 
end
toc
end

The approximate execution times are:

R2020b: 3.9 s

R2021a: 1.2 s

Also, if you use unique on a numeric input with 10,000 elements and specify three outputs with the 'stable' option, performance in R2021a is about 3.5x faster than in R2020b.

function timingUniqueNumeric
a = rand(10000,1);
tic
for ii=1:1e4
  [C,ia,ic] = unique(a,'stable');
end
toc
end

The approximate execution times are:

R2020b: 9.4 s

R2021a: 2.7 s

In both cases, the code was timed on a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system by calling the functions timingUniqueCell and timingUniqueNumeric.

Graph Functions: Improved performance modifying node and edge lists

graph and digraph functions that modify the node and edge lists of the graph show improved performance. This applies to the functions addedge, rmedge, addnode, rmnode, subgraph, and reordernodes. The improvement applies to graphs that have no node properties (or only node names), and graphs with no edge properties (or only edge weights). The improvement is most noticeable when one of these functions is called many times in a loop, and the largest improvement applies to graphs that have both node names and edge weights.

For example, if you use addedge in a loop to add new edges with node names and edge weights to an empty graph, performance in R2021a is about 13x faster than in R2020b.

function timingAddedge
names = string(('A':'Z')') + (1:10);
names = names(1:100);
rng default
g = graph;
for ii = 1:1e3
    g = addedge(g, names(randi(100)), names(randi(100)), randn);
end
end

The approximate execution times are:

R2020b: 2.6 s

R2021a: 0.2 s

The code was timed on a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system by using the timeit function:

timeit(@timingAddedge)

Axes Toolbar: Appears without delay when axes are ready

Prior to R2021a, when hovering the cursor over figure axes, there was delay before the axes toolbar appeared. Now the toolbar will appear as soon as the axes are ready.

Rearranging UI Components: Improved performance when rearranging UI components in a UI figure

Programmatically rearranging existing UI components in a figure created with the uifigure (R2021a) function shows improved performance.

For example, if you sort 50 panels within a grid layout, performance in R2021a is approximately 1.9x faster than in R2020b.

function timingSortComp
    % Create components
    panels = {};
    fig = uifigure;
    g = uigridlayout(fig,[1,1],'RowHeight',40);
    g.Scrollable = true;
    num = 50;
    for i = 1:num
        p = uipanel(g);
        uilabel(p,'Text',['Panel ', num2str(i)],'Position',[10 10 70 22]);
        g.RowHeight{end} = 40;
        panels{end+1} = p;   
    end
    drawnow;

    % Rearrange components
    tic
    order = length(panels):-1:1;
    for i = 1:length(order)
        panels{i}.Layout.Row = order(i);
    end
    drawnow;
    toc
end

The approximate execution times are:

R2020b: 0.70 s

R2021a: 0.36 s

The code was timed on a Windows 10, Intel Xeon E5-1650 CPU @ 3.60 GHz test system by calling the function timingSortComp.

UI Figure Interactions: Faster responses to scroll, pointer movement, and resize interactions in UI figures

In figures created with the uifigure function, the following interactions have improved performance:

Scrolling in a figure with a WindowScrollWheelFcn callback or an object with a predefined scroll behavior
Resizing a visible figure with a SizeChangedFcn callback or an object with a predefined resize behavior
Moving the mouse pointer in a figure with a WindowButtonMotionFcn callback when the figure contains any UI components except axes components

This performance increase is more noticeable when using a trackpad to interact with the figure.

For example, scrolling to zoom in on the plot created by the code below is smoother and more responsive in R2021a than in R2020b.

t1 = datetime(2019,1,1);
t2 = datetime(2020,1,1);
dates = linspace(t1,t2,10000);
data = rand(10000,10);
fig = uifigure;
stackedplot(fig,dates,data);

On a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system, the responses to the scroll action are:

R2020b: When you zoom in on the plot by scrolling for approximately two seconds, the plot has about a five second delay in completing the zoom animation.

R2021a: The zoom animation completes immediately after you finish the scroll action.

Plots in Apps: Improved performance for polar plots, volume visualizations, plots with more than 16 axes, and older systems

Displaying polar plots, volume visualizations, or more than 16 axes in an app have improved performance. The improvement affects plots displayed in apps:

Plots that are displayed in an app created with App Designer
Plots displayed in a figure created with the uifigure function

Systems with older graphics drivers might experience the improvement for all types of plots that are created within the apps and figures listed above. For example, Intel drivers earlier than version 10.0.0.0 for Windows systems will experience additional improvements.

This code creates a polar plot and executes a for loop that changes the theta values at every iteration. The for loop executes about 2x faster than in R2020b.

function timingPolar
f = uifigure;
pax = polaraxes('parent',f);
theta = 0:0.01:2*pi;
rho = sin(2*theta).*cos(2*theta);
pp = polarplot(pax, theta,rho);
pax.FontSize = 12;
drawnow

tic;
for i=1:100
    pp.ThetaData = pp.ThetaData + .02*pi;
    drawnow
end
toc
end

The approximate execution times are:

R2020b: 10.15 s

R2021a: 5.30 s

The code was timed on a Windows 10, Intel Xeon W-2133 CPU @ 3.60 GHz test system by calling the function timingPolar.

Plots in Apps: Improved performance for plots with large numbers of markers

Performance is improved for modifying certain types of plots in apps. You can observe the improvement when the following conditions are true:

The plots are displayed in an App Designer app, or they are displayed in a figure created with the uifigure function.
Your system is running a locally installed version of MATLAB on a modern Windows or macOS system.
You run the code either from the MATLAB command window or within a program file (.m file).

Note that plots created in live scripts do not show this improvement.

The plots typically contain large numbers of markers, and your code updates an aspect of those markers, such as their positions. The improvement is more significant as you increase the number of markers. For example, if you create a scatter plot with 10 million markers, and change the marker positions 10 times, the performance in R2021a is about 1.3x faster than in R2020b.

function timingScatter
f = uifigure;
a = axes(f);
x = rand(1e7, 2);
s = scatter(a,x(:,1),x(:,2),'Marker','*');
drawnow;

tic;
for i=1:10
    x = rand(1e7,2);
    s.XData = x(:,1);
    s.YData = x(:,2);
    drawnow
end
toc
end

The approximate execution times are:

R2020b: 19.43 s

R2021a: 14.88 s

The code was timed on a macOS 10.14.6, Intel Core^® i9 CPU @ 3.60 GHz test system by calling the function timingScatter.

Live Editor: Improved performance when saving large live scripts or functions

For example, on a Windows 10, Intel Xeon E5-1650 CPU @ 3.60 GHz test system, if you save a live script containing 35,000 lines of code and then click the title bar of another document open in the Live Editor, MATLAB immediately switches to the other open document. In R2020b, there is a noticeable delay before MATLAB switches to the other open document.

Software Development Tools

Projects: List all referenced projects of the current project

You can now use listAllProjectReferences (R2021a) to programmatically list all projects in the reference hierarchy of a specified project.

Projects: List impacted project files

You can now use listImpactedFiles (R2021a) to programmatically list all project files impacted by changes to a specified file.

Dependency Analyzer: Find required add-ons

Starting in R2021a, the Dependency Analyzer detects and lists required add-ons, including apps and toolboxes, for the whole project or for selected files. The Dependency Analyzer might not detect required support packages. For more details, see Find Required Products and Add-Ons (R2021a).

Unit Testing Framework: Create test runners using alternative syntax

You can now use the testrunner (R2021a) function to create a runner for tests authored using the MATLAB unit testing framework or Simulink^® Test™. In previous releases, you can explicitly create a runner only by calling one of the static methods of the matlab.unittest.TestRunner (R2021a) class.

Use the testrunner function to create a default runner, a minimal runner with no plugins installed, or a runner configured for text output. For example, create a default runner to run the tests in a test class.

suite = testsuite('MyTestClass');
runner = testrunner;
results = run(runner,suite);

Unit Testing Framework: Initialize parameterization properties at suite creation time

Starting in R2021a, you can specify parameterization properties that do not have a default value. This feature is useful when parameters cannot be determined at the time MATLAB loads the test class definition. To initialize a parameterization property at test suite creation time, use a static method with the TestParameterDefinition attribute. For more information, see Define Parameters at Suite Creation Time (R2021a).

Unit Testing Framework: Run tests in parallel on thread-based pool

You can now run your tests on a thread-based parallel pool (requires Parallel Computing Toolbox). To run tests using thread workers, start a thread-based pool and then call the runInParallel method or the runtests function with the UseParallel name-value pair argument.

Thread-based parallel pools support only a subset of MATLAB and testing framework functionality. For more information, see runInParallel (R2021a) or runtests (R2021a).

Unit Testing Framework: Run tests in MATLAB Online interactively

Starting in R2021a, you can run tests in MATLAB Online interactively. When you open a .m file defining a function-based or class-based test in MATLAB Online, the toolstrip lets you run all tests in the file or run the current test in the file. Also, you can customize the test run with options, such as running tests in parallel (requires Parallel Computing Toolbox) or running tests with a specified level of output detail.

The Run Tests section in the Editor tab of the toolstrip provides an alternative to programmatically running tests with the runtests function. For more information, see Run Tests in Editor (R2021a).

App Testing Framework: Perform gestures on panels and tables

The app testing framework supports gestures on more UI components:

Perform press (R2021a), hover (R2021a), and chooseContextMenu (R2021a) gestures on panels.
Perform choose (R2021a), type (R2021a), and chooseContextMenu (R2021a) gestures on table UI components.

App Testing Framework: Close alert dialog box in front of figure window

fig = uifigure;
uialert(fig,'File not found','Invalid File')

tc = matlab.uitest.TestCase.forInteractiveUse;
tc.dismissAlertDialog(fig)

Functionality being removed or changed

Character data is not supported in custom examples demos.xml file

Behavior change

Starting in R2021a, when creating custom examples, character data is not supported in the description of the demos.xml file. If you have a demos.xml file that contains character data such as <, >, ', ", and &, the description does not appear correctly in the Help browser.

To patch an existing demos.xml that contains character data, use the patchdemoxmlfile (R2021a) function. patchdemoxmlfile patches the specified demos.xml file, replacing character data with non-character data.

For example, patch the demos.xml file in the folder D:\Work\mytoolbox\help:

patchdemoxmlfile D:\Work\mytoolbox\help

External Language Interfaces

C++ Interface: Support for C++ language features

The C++ interface supports these additional C++ language features.

Support for std::vector values containing std::string values and C++ arrays containing C-style strings. For more information, see String and Character Types (R2021a) in the C++ to MATLAB Data Type Mapping topic.
Support for void* values as input and output arguments. For more information, see Use void* Arguments (R2021a), void* Argument Types (R2021a), and addOpaqueType (R2021a).
Pass C++ functions to function pointers. For more information, see Use Function Type Arguments (R2021a) and addFunctionType (R2021a).
Support for function and member function template (R2021a) instantiations. Publishers can modify function names. For more information, see Customize Function Template Names (R2021a).

C++ Interface: Publisher options and analysis

The C++ interface supports these build configuration features.

Generate an interface from header and source (.cpp) files. Pass a .cpp or .hpp file in the clibgen.generateLibraryDefinition (R2021a) SourceFiles argument.
Generate an interface from a header and a .dll file for Microsoft^® Visual Studio^® compilers. Pass a .dll file in the clibgen.generateLibraryDefinition (R2021a) LibraryFiles argument.
Improved troubleshooting messages.

Java Packages to be removed

Java^® packages and subpackages that currently ship with MATLAB will not be available in MATLAB in a future release.

Compatibility Considerations

To continue using a Java package, install its JAR file and add the JAR file to the static path in MATLAB using the instructions in Static Path (R2021a).

Java Engine: MATLAB value object support

To work with MATLAB value objects in a Java engine application, use the com.mathworks.matlab.types.ValueObject (R2021a) class in the Java Engine API Summary (R2021a). You can create a value object in MATLAB, return it to Java, and call its methods. For information about mapping Java data types to MATLAB data types, see Java Data Type Conversions (R2021a).

Python Interface and Engine: Version 3.6 support discontinued

Support for Python^® version 3.6 is discontinued.

Compatibility Considerations

To ensure continued support for your applications, upgrade to a supported version of Python, either version 3.7 or 3.8. For more information, see Versions of Python Compatible with MATLAB Products by Release.

Perl 5.32.0: MATLAB support on Windows

As of R2021a, MATLAB on Windows ships with an updated version of Perl, version 5.32.0.

See www.perl.org for a standard distribution of Perl, Perl source, and information about using Perl.
See https://metacpan.org/pod/HTML::Parser for a standard distribution of HTML::Parser, source code, and information about using HTML::Parser.
See https://metacpan.org/pod/HTML::Tagset for a standard distribution of HTML:Tagset, source code, and information about using HTML:Tagset.

Compatibility Considerations

If you use the perl (R2021a) command on Windows platforms, see www.perl.org for information about using this version of the Perl programming language.

Hardware Support

Support added for IMU sensors

The MATLAB Support Package for Raspberry Pi^® Hardware now provides code generation and connected IO support to Raspberry Pi functions for these IMU sensors:

HTS221
LPS22HB
LSM303C
LSM6DSL
LSM9DS1
MPU-6050
MPU-9250

New functionalities added to Raspberry Pi Resource Monitor app

The Raspberry Pi Resource Monitor App from the MATLAB Support Package for Raspberry Pi Hardware has been improved to:

Display peripherals used in a MATLAB or Simulink application deployed on the Raspberry Pi hardware
Enable or disable peripherals
Check for missing libraries and packages
Display all processes currently running on the Raspberry Pi hardware