categorical axes are somewhat peculiar -- the categorical variable always contains all the categories for which it was created and there weren't missing values when the x variable was created; only the associated y value has some zero elements.
To not have those show up in the plot you'll need to create a new categorical variable that matches the wanted locations only--something like
isOK=~ismissing(y); % or y==0 or isfinite(y) or whatever is the actual condition to keep
cats=categories(x); % get the category names from the present x categorical array
xOK=categorical(x(isOK),x(isOK),cats(isOK)); % the new categorical variable to plot with
bar(xOK,y(isOK)) % will plot only the extant/wanted bars with names
Otherwise, you'll get a categorical axes with all categories in the original vector; even though the actual elements may not be shown/plotted, the axes will still be laid out in the original internal numerical position were when created.
categorical variables are kinda' tricky that way, they always carry around the entire definition of all categories in the original or augmented creation whether there are actual values for all categories or not; only the display names are not visible for the missing, but the categories are still defined. It's similar to datetime display formatting -- you can hide the normal display of any field such as the year, but the datetime value is still the same; only the display format has been changed.
