corTree — the Tree, and Nothing Else

The in-memory tree every other part of the coraine stack works on: the node, and the functions that build, edit, search, clone and sort a tree of them. It has no parser and no renderer. JSON in and out is corJson; any other format — the binary one the cor:// protocol and corDB persistence will use, XML, YAML — is a sibling of corJson that produces or consumes the same tree.

• Version: 0.1.0
• Language: C
• License: Apache License 2.0

The only dependencies are corAlloc, corLog and corBase.

Where it comes from

corTree and corJson are the two halves of kjson, which was one library that held both the representation and the JSON parser. That was the design mistake the split corrects: with the tree on its own, every format is just another producer or consumer of it. kjson itself is untouched and keeps serving its own users; this is a copy of its implementation under new names, not a fork.

The one change of substance is the allocator. kjson's builders took the parser's handle, Kjson*, and used nothing of it but the allocator inside — which made the tree depend on the parser. corTree's builders take the allocator itself:

CorNode* corTreeString(CorAlloc* kaP, const char* name, const char* value);

NULL still means malloc, exactly as a NULL Kjson* did.

The node

typedef struct CorNode
{
  char*            name;       // NULL/"" for an array item
  CorValueType     type;       // CorString, CorInt, CorFloat, CorBoolean, CorNull, CorObject, CorArray
  unsigned char    flags;      // for the users of the library - opaque to corTree
  uint16_t         termId;     // for the users of the library (NGSI-LD: which core term, 0 = none) - opaque to corTree
  CorValue         value;      // b, i, f, s, or - for a container - head and tail
  struct CorNode*  next;       // next sibling
} CorNode;

40 bytes on a 64-bit machine - flags and termId sit in the alignment padding between type and value, and CorNode.h asserts both the size and the offset. Every builder zeroes flags and termId (under malloc too), and corTreeClone copies them. Any other code that builds nodes (a parser, a database reader) must do the same.

API

build	corTreeObject, corTreeArray, corTreeString, corTreeInteger, corTreeFloat, corTreeBoolean, corTreeNull
edit	corTreeChildAdd, corTreeChildAddSorted, corTreeChildAddSortedReverse, corTreeChildPrepend, corTreeChildRemove, corTreeChildMove, corTreeChildReplace, corTreeChildAddOrReplace, corTreeNodeDecouple
find	corTreeLookup, corTreeNavigate, corTreeStringSiblingFind, corTreeStringValueLookupInArray, corTreeChildCount
copy, free	corTreeClone, corTreeFree
sort	corTreeSort, corTreeArraySort, corTreeStringArraySort, corTreeStringArraySortedInsert

Build

A sibling repo of the rest of the stack: -I.. and ../corTree/libcorTree.a.

make di