The problem is that, by default (without the "hilev" option), legends are turned into regular axes for which the "visible" property is not interpreted the same way by the rederer.

@Rik: The idea of the "hilev" argument was to make use of high level functions to restore listeners. This option was meant to be enabled whenever we want to load dynamic objects, like in hdload, unfortunately it is untested up to now and the code behind it is probably not the right way to go. I would not recommend to enable it until we have a proper mechanism to reload objects which heavily rely on listeners (legends, colorbars, bar groups ...).

PS: I have proposed an improvement in bug #44670 that make struct2hdl more compatible with Matlabs old .fig format. When I get some time, I'll try to rebase those changes and propose some cleanup same time.

Adding Pantxo to the CC list since I believe he wrote these functions originally.

I can confirm that hdl2struct successfully saves a legend object in the struct.  But struct2hdl does not restore it completely.  There is an optional third input to struct2hdl which when set to 'true' helps a bit.  But I'm shooting in the dark here.

Changing title to reflect the root issue.  Although an axes object is partially restored, it is not a true legend object.  It lacks, for example, the tag "legend".  Restoring that tag is enough to correct the display of the object.

In the example

plot([2 1])
legend("a") ;
legend boxoff

hgsave foo.og
hgload foo.og

the loaded graphics has the legend box on.