[erlang-questions] SAX-like JSON Parser for Erlang

[Richard O'Keefe]

On 23/01/2013, at 4:41 PM, Bob Ippolito wrote:

> JSON pointers are much less powerful than XPath, it's not really a query language. No predicates, result must be a single node. It's just '/foo/bar/baz/0' instead of 'obj.foo.bar.baz[0]' (in JS).
> 
> Something in-between (more powerful than JSON pointers, less powerful than XPath) would be something like https://github.com/etrepum/kvc -- It won't generate results from a stream, so you'd need to use it with a standard JSON parser.

The examples in the README.md of https://github.com/etrepum/kvc
do things like

wibble =:= kvc:path(foo.bar.baz, [{foo, [{bar, [{baz, wibble}]}]}]).

This feels wrong to me.  A path should be a _list_ of
step descriptions, [foo,bar,bar].  Reasons:
(1) You can have integer steps (this element of a tuple)
    as well as atom steps (this entry in a dict &c).
    And you can also have atom steps that _look_ like integers.
(2) It is more efficient not to have to split a path into steps
    at run time.
(3) Perhaps the most painfully obvious:
    a named step might need to include a dot (or any other fixed
    character) in its name.
(4) Given a recursive data structure with a small set of labels,
    using dotted atoms you can quickly exhaust the size of
    Erlang's atom table.

In many ways, this is a perfect example of "strings are wrong".
The abstract concept we need here is
   "path"
and
   "path" = sequence of "step"
and
   "step" = receiver-specific position or label

Packing a path up as a dotted atom or any other kind of
string representation means having to recover at run time
and unreliably information that has been _hidden_ inside
the string, when it could have been made directly available
as a simple data structure.

I have a key-value component for my Smalltalk library, and
in that (1,3,4) were not issues, but (2) had my programmer's
instincts screaming 'this is a bad idea'.  In fact one of the
things on my TODO list is to replace
	aPath subStrings: ' .' asClass: Symbol trimmed: true
by
	aPath _steps
in order to let aPath be returned if it's a sequence of steps
already.

Fortunately(?) the README.md is incomplete, and that KVC
implementation _does_ accept a list of steps (which look as
though they have to be binaries).  That module is better than
it sounds.

By the way, things like this remind me irresistibly of
Nicklaus Wirth's "Professor Cleverbyte's Visit to Heaven"
(Software Practice and Experience, Vol 7 pp155-158, 1977).