...
| expr | value | notes |
|---|---|---|
| <[ ]> | "" | list is empty, yields empty string |
| <[ ]; null="x"> | "" | list is empty; no null element => no x |
| <[[ ], [ ]]:{it | <if(it)>x<endif>}; separator=","> | "" | [[],[]] collapses to [ ] |
| <[ ]:t()> | "" | nothing to apply; template not evaluated |
| <[ ]:{it | <if(it)>x<endif>}> | "" | nothing to apply; template not evaluated |
We recently modified [ ] to mean null, but that is inconsistent with passing in "new ArrayList()", which is non-null but has no elements.
We also added [ ] to mean "missing" or null in dictionaries. We should either use a literally missing value or more likely just don't add that value to the map. Any missing map entry evaluates to null.
For dictionary d ::= [x:"x"], <d.(x):{it | <it>}> is x and <d.(y):{it | <it>}> is "".
So, to summarize, I 'm not sure why we need [ ] or missing values in [ ] but we can leave that stuff. We have to backout recent changes where [ ] is the default value of nullthink we're ok as-is. I propose that we change null values to count in <i> index expressions if there is a null option in a future version like 4.1 but let's not change it on a point release like 4.0.7.
...