5 Limitations of high- and medium-level functions

Note that netcellfun, netarrayfun, and rfeval, but even parcellfun and pararrayfun, execute code in separate Octave processes. This causes some limitations.

For the function handles passed as arguments to parcellfun, pararrayfun, netcellfun, netarrayfun, and rfeval, the following limitations may apply:

• In Octave versions before 5.1, anonymous functions may only be passed if they don’t use varargin and don’t contain nested anonymous functions.
• If a handle to a named function, written in a function file, is passed, this function must exist on the server machine. This is usually no problem for parcellfun and pararrayfun.
• Named functions defined in the command line interface or by scripts are not exported to the working sessions used by parcellfun or pararrayfun, and not automatically to the server for remote parallel execution.
• For handles to subfunctions or to private functions limitations depending on the used Octave version may apply.

Anonymous functions should be usable as usual, since the function specification sent to the server will include the anonymous functions context.

If parcellfun or pararrayfun (and of course the cluster-related functions, too) are called from within a function with subfunctions, the subfunctions won’t be visible in parallel execution.

There is no way to set global variables for the function called by parcellun or pararrayfun. For cluster execution, setting global variables is possible, e.g. with install_vars; however, netcellfun and netarrayfun won’t ’see’ global variables if they use parcellfun internally (which is the default at remote machines with more than one available processor core).