We need to add "with" as a new row to the table in the (earlier) section "Operator precedence and associativity." And as we have historically organized the sections in this clause in descending precedence order, we might need to move this new section up or down, accordingly.

FYI: In V9,record meansrecord class but we can't say that explicitly. In V10, this is permitted and we'll replace these 2 rules with'record'? 'class'.

Heads up grammarians! As you will see below, I have a generous syntax for a class in which not all combinations are permitted, and those combinations are described in text as constraints. (For example, a record class can have adelimited_parameter_list while a non-record class cannot. And for a record class,class_modifier shall not bestatic.) Yes, we could try to break this up into separate rules for record classes and non-record classes to eliminate some (but maybe not all) of those constraints. However, there will be tweaks to this grammar in future versions; specifically:

• V10 allowsrecord class as well asrecord.
• V10 supportsrecord struct.
• V12 allows a primary constructor (which usesdelimited_parameter_list) on non-record classes and structs.

which means that extra work done now to make the grammar rules more complete likely will be (at least partially) undone in future revs.

Do we really want to expose this type in the spec?