I've checked in the json_term_changes branch today. Big thanks to
Chris Anderson (jchris) as this is mostly his work.
The new term format is:
json_string() = binary()
json_number() = integer() | float()
json_array() = [json_term()]
json_object() = { [{json_string(), json_term()}] }
json_term() = json_string() | json_number() | json_array() |
json_object()
The main purpose with these changes are:
1. Less memory. The new json terms use binary strings instead of list
strings, this makes each JSON string is now about 1/8 the size in
memory.
2. To use the "blessed" term format. Among Erlang principals, this
format has been decided upon as standard term format for representing
JSON in Erlang. The means we should benefit from a wider range of
Erlang Json libraries sharing the same format, and much faster parsing
as there is now a proposal to create Erlang built-in functions for
parsing and producing json using this term format.
Everything should continue work identically for clients, there are no
semantic changes here. However, this does break existing databases at
the file level, so you'll have to dump, upgrade and recreate your
existing databases.
These changes individually are mostly minor, but they are numerous and
touch a lot of the code base. I'd appreciate any time people can spend
looking at the changes and testing the code. Comments and criticisms
welcome. Hopefully we can get this merged into trunk quickly.
-Damien
