bug#15097: 24.3.50; json.el can't encode lists of lists

bug#15097: 24.3.50; json.el can't encode lists of lists

[Rolando Pereira]

Hello all,

The function `json-encode' can't encode a list thats composed only of other
lists, i.e. the following doesn't work:

    (json-encode '((1 2 3))) => Error: (json-key-format 1)

If I had an extra "nil" in the topmost list `json-encode' then it works
fine:

    (json-encode '((1 2 3) nil)) => "[[1,2,3],null]"

However if I replace the `nil' with another list then I get the same
error:

    (json-encode '((1 2 3) (4))) => Error: (json-key-format 1)

Here's a small ERT test that should trigger the bug:

    (ert-deftest json-encode-list-of-lists-test ()
      (ert-should (string= (json-encode '(1 2 3)) "[1,2,3]"))              ; works
      (ert-should (string= (json-encode '((1) 2 3)) "[[1],2,3]"))          ; works
      (ert-should (string= (json-encode '((1 2 3))) "[[1,2,3]]"))          ; doesn't work
      (ert-should (string= (json-encode '((1 2 3) nil)) "[[1,2,3],null]")) ; works
      (ert-should (string= (json-encode '((1 2 3) (4))) "[[1,2,3],[4]]"))  ; doesn't work
      )


Best regards,
Rolando Pereira

bug#15097: 24.3.50; json.el can't encode lists of lists

[Glenn Morris]

Rolando Pereira wrote:

> The function `json-encode' can't encode a list thats composed only of other
> lists, i.e. the following doesn't work:
>
>     (json-encode '((1 2 3))) => Error: (json-key-format 1)
[...]
> (ert-should (string= (json-encode '((1 2 3))) "[[1,2,3]]")) ; doesn't work

I have no idea what this stuff is supposed to do, but by inspection it
treats such arguments as alists and requires that the key be encodable
as a string. Eg this works

(json-encode '((a 2 3)))   ->    "{\"a\":[2, 3]}"

Your example fails because 1 is encoded as 1, not "1".

Should the answer be "{\"1\":[2, 3]} or "[[1,2,3]]"?

bug#15097: 24.3.50; json.el can't encode lists of lists

[Rolando Pereira]

Glenn Morris <rgm@gnu.org> writes:

> Rolando Pereira wrote:
>
>> The function `json-encode' can't encode a list thats composed only of other
>> lists, i.e. the following doesn't work:
>>
>>     (json-encode '((1 2 3))) => Error: (json-key-format 1)
> [...]
>> (ert-should (string= (json-encode '((1 2 3))) "[[1,2,3]]")) ; doesn't work
>
> I have no idea what this stuff is supposed to do, but by inspection it
> treats such arguments as alists and requires that the key be encodable
> as a string. Eg this works
>
> (json-encode '((a 2 3)))   ->    "{\"a\":[2, 3]}"
>
> Your example fails because 1 is encoded as 1, not "1".
>
> Should the answer be "{\"1\":[2, 3]} or "[[1,2,3]]"?

0Personally I was expecting the answer to be "[[1,2,3]]" i.e. turn the
'(1 2 3) list into a list and not an object.

Which, for the purposes of comparison, is similar to what the Lisp
library "cl-json" and the Python "json" library do:

* Common Lisp

CL-USER> (ql:quickload 'cl-json)
[...snip...]
CL-USER> (cl-json:encode-json '((1 2 3)))
[[1,2,3]]
NIL
CL-USER>

* Python

Python 2.7.3 (default, Aug  1 2012, 05:16:07) 
[GCC 4.6.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import json
>>> json.dumps([[1,2,3]])
'[[1, 2, 3]]'

bug#15097: 24.3.50; json.el can't encode lists of lists

[Rolando Pereira]

Glenn Morris <rgm@gnu.org> writes:

> Rolando Pereira wrote:
>
>> The function `json-encode' can't encode a list thats composed only of other
>> lists, i.e. the following doesn't work:
>>
>>     (json-encode '((1 2 3))) => Error: (json-key-format 1)
> [...]
>> (ert-should (string= (json-encode '((1 2 3))) "[[1,2,3]]")) ; doesn't work
>
> I have no idea what this stuff is supposed to do, but by inspection it
> treats such arguments as alists and requires that the key be encodable
> as a string. Eg this works
>
> (json-encode '((a 2 3)))   ->    "{\"a\":[2, 3]}"
>
> Your example fails because 1 is encoded as 1, not "1".
>
> Should the answer be "{\"1\":[2, 3]} or "[[1,2,3]]"?

Personally I was expecting the answer to be "[[1,2,3]]" i.e. turn the
'(1 2 3) list into a list and not an object.

Which, for the purposes of comparison, is similar to what the Lisp
library "cl-json" and the Python "json" library do:

* Common Lisp

CL-USER> (ql:quickload 'cl-json)
[...snip...]
CL-USER> (cl-json:encode-json '((1 2 3)))
[[1,2,3]]
NIL
CL-USER>

* Python

Python 2.7.3 (default, Aug  1 2012, 05:16:07) 
[GCC 4.6.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import json
>>> json.dumps([[1,2,3]])
'[[1, 2, 3]]'

bug#15097: 24.3.50; json.el can't encode lists of lists

[Glenn Morris]

Rolando Pereira wrote:

> Python 2.7.3 (default, Aug  1 2012, 05:16:07) 
> [GCC 4.6.3] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
>>>> import json
>>>> json.dumps([[1,2,3]])
> '[[1, 2, 3]]'

Well, maybe that's the actual issue, since json.el and the python
version apparently disagree about what to do in such cases. Eg:

(json-encode '(("a" 2 3)))  ->  "{\"a\":[2, 3]}"
json.dumps([["a",2,3]])     ->  [["a", 2, 3]]


(Context is at http://debbugs.gnu.org/15097 )

bug#15097: 24.3.50; json.el can't encode lists of lists

[Edward O'Connor]

The basic problem is that (("a" 1 2)) is the same thing as (("a" . (1
2))). Is that a list of one three-item list, or is it an alist with
one entry, whose key is "a" and whose value is a two-item list? It
might be reasonable for `json-encode-list' to provide programmatic
control over this, but the current behavior isn't a bug.

On Tue, Aug 20, 2013 at 10:04 AM, Glenn Morris <rgm@gnu.org> wrote:
> Rolando Pereira wrote:
>
>> Python 2.7.3 (default, Aug  1 2012, 05:16:07)
>> [GCC 4.6.3] on linux2
>> Type "help", "copyright", "credits" or "license" for more information.
>>>>> import json
>>>>> json.dumps([[1,2,3]])
>> '[[1, 2, 3]]'
>
> Well, maybe that's the actual issue, since json.el and the python
> version apparently disagree about what to do in such cases. Eg:
>
> (json-encode '(("a" 2 3)))  ->  "{\"a\":[2, 3]}"
> json.dumps([["a",2,3]])     ->  [["a", 2, 3]]
>
>
> (Context is at http://debbugs.gnu.org/15097 )
>



-- 
Edward O'Connor

bug#15097: Wontfix, apparently

[Daniel Colascione]

[-- Attachment #1: Type: text/plain, Size: 35 bytes --]

Wontfix according to discussion


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 901 bytes --]