Mircea Namolaru <NAMOLARU@xxxxxxxxxx> wrote:
>> If the bit is going to be widely used and necessary, it makes sense to
>> it to trees. Otherwise, if there are currently no plans to use it outside
>> cha, it's better stored outside (like, keep a pointer_set of the used
>> classes and add a public query like cha_class_used_p). This would make it
>> non-invasive, and also clear that you require cha and whole program mode
>> for the information to be available.
> First no bit have been added to the trees. One of the existents bits
> meaningful only for ARRAY and UNION types but with no meaning for a
> RECORD type has been used.
When I say "add" it, I say that there is a new bit with its macro accessor
and new code to set it, irrespective of whether this enlarges the data
structure, uses a previously unused bit in a bitfield, or overloads a bit
already used for different tree codes.
This definition of "adding" is important because, when we work on
*shrinking* data structures, we need to refactor each and every bit "added"
(by this definition) to the trees, and move their uses in aside data
structures (e.g. external hash tables). So "adding" a new bit means more
work if later this bit is found too be rarely used.
> On the other side, I understand now your (and Daniel's) point. Is this
> information general enough, useful enough, to be placed in a global
> data structure ? No, you (and Daniel) are right, it should be enclosed
> in the cha code.
Fine, we agree then! :)