I making a foray into the world of JSON parsing and NewtonSoft and I'm confused, to say the least.
Take the below PowerShell script:
$json = #"
{
"Array1": [
"I am string 1 from array1",
"I am string 2 from array1"
],
"Array2": [
{
"Array2Object1Str1": "Object in list, string 1",
"Array2Object1Str2": "Object in list, string 2"
}
]
}
"#
#The newtonSoft way
$nsObj = [Newtonsoft.Json.JsonConvert]::DeserializeObject($json, [Newtonsoft.Json.Linq.JObject])
$nsObj.GetType().fullname #Type = Newtonsoft.Json.Linq.JObject
$nsObj[0] #Returns nothing. Why?
$nsObj.Array1 #Again nothing. Maybe because it contains no key:value pairs?
$nsObj.Array2 #This does return, maybe because has object with kv pairs
$nsObj.Array2[0].Array2Object1Str1 #Returns nothing. Why? but...
$nsObj.Array2[0].Array2Object1Str1.ToString() #Cool. I get the string this way.
$nsObj.Array2[0] #1st object has a Path property of "Array2[0].Array2Object1Str1" Great!
foreach( $o in $nsObj.Array2[0].GetEnumerator() ){
"Path is: $($o.Path)"
"Parent is: $($o.Parent)"
} #??? Why can't I see the Path property like when just output $nsObj.Array2[0] ???
#How can I find out what the root parent (Array2) is for a property? Is property even the right word?
I'd like to be able to find the name of the root parent for any given position. So above, I'd like to know that the item I'm looking at (Array2Object1Str1) belongs to the Array2 root parent.
I think I'm not understanding some fundamentals here. Is it possible to determine the root parent? Also, any help in understanding my comments in the script would be great. Namely why I can't return things like path or parent, but can see it when I debug in VSCode.
dbc's answer contains helpful background information, and makes it clear that calling the NewtonSoft Json.NET library from PowerShell is cumbersome.
Given PowerShell's built-in support for JSON parsing - via the ConvertFrom-Json and ConvertTo-Json cmdlets - there is usually no reason to resort to third-party libraries (directly[1]), except in the following cases:
When performance is paramount.
When the limitations of PowerShell's JSON parsing must be overcome (lack of support for empty key names and keys that differ in letter case only).
When you need to work with the Json.NET types and their methods rather than with the method-less "property-bag" [pscustomobject] instances ConvertFrom-Json constructs.
While working with NewtonSoft's Json.NET directly in PowerShell is awkward, it is manageable, if you observe a few rules:
Lack of visible output doesn't necessarily mean that there isn't any output at all:
Due to a bug in PowerShell (as of v7.0.0-preview.4), [JValue] instances and [JProperty] instances containing them produce no visible output by default; access their (strongly typed) .Value property instead (e.g., $nsObj.Array1[0].Value or $nsProp.Value.Value (sic))
To output the string representation of a [JObject] / [JArray] / [JProperty] / [JValue] instance, do not rely on output as-is (e.g, $nsObj), use explicit stringification with .ToString() (e.g., $nsObj.ToString()); while string interpolation (e.g., "$nsObj") does generally work, it doesn't with [JValue] instances, due to the above-mentioned bug.
[JObject] and [JArray] objects by default show a list of their elements' instance properties (implied Format-List applied to the enumeration of the objects); you can use the Format-* cmdlets to shape output; e.g., $nsObj | Format-Table Path, Type.
Due to another bug (which may have the same root cause), as of PowerShell Core 7.0.0-preview.4, default output for [JObject] instances is actually broken in cases where the input JSON contains an array (prints error format-default : Target type System.Collections.IEnumerator is not a value type or a non-abstract class. (Parameter 'targetType')).
To numerically index into a [JObject] instance, i.e. to access properties by index rather than by name, use the following idiom: #($nsObj)[<n>], where <n> is the numerical index of interest.
$nsObj[<n>] actually should work, because, unlike C#, PowerShell exposes members implemented via interfaces as directly callable type members, so the numeric indexer that JObject implements via the IList<JToken> interface should be accessible, but isn't, presumably due to this bug (as of PowerShell Core 7.0.0-preview.4).
The workaround based on #(...), PowerShell's array-subexpression operator, forces enumeration of a [JObject] instance to yield an array of its [JProperty] members, which can then be accessed by index; note that this approach is simple, but not efficient, because enumeration and construction of an aux. array occurs; however, given that a single JSON object (as opposed to an array) typically doesn't have large numbers of properties, this is unlikely to matter in practice.
A reflection-based solution that accesses the IList<JToken> interface's numeric indexer is possible, but may even be slower.
Note that additional .Value-based access may again be needed to print the result (or to extract the strongly typed property value).
Generally, do not use the .GetEnumerator() method; [JObject] and [JArray] instances are directly enumerable.
Keep in mind that PowerShell may automatically enumerate such instances in contexts where you don't expect it, notably in the pipeline; notably, when you send a [JObject] to the pipeline, it is its constituent [JProperty]s that are sent instead, individually.
Use something like #($nsObj.Array1).Value to extract the values of an array of primitive JSON values (strings, numbers, ...) - i.e, [JValue] instances - as an array.
The following demonstrates these techniques in context:
$json = #"
{
"Array1": [
"I am string 1 from array1",
"I am string 2 from array1",
],
"Array2": [
{
"Array2Object1Str1": "Object in list, string 1",
"Array2Object1Str2": "Object in list, string 2"
}
]
}
"#
# Deserialize the JSON text into a hierarchy of nested objects.
# Note: You can omit the target type to let Newtonsoft.Json infer a suitable one.
$nsObj = [Newtonsoft.Json.JsonConvert]::DeserializeObject($json)
# Alternatively, you could more simply use:
# $nsObj = [Newtonsoft.Json.Linq.JObject]::Parse($json)
# Access the 1st property *as a whole* by *index* (index 0).
#($nsObj)[0].ToString()
# Ditto, with (the typically used) access by property *name*.
$nsObj.Array1.ToString()
# Access a property *value* by name.
$nsObj.Array1[0].Value
# Get an *array* of the *values* in .Array1.
# Note: This assumes that the array elements are JSON primitives ([JValue] instances.
#($nsObj.Array1).Value
# Access a property value of the object contained in .Array2's first element by name:
$nsObj.Array2[0].Array2Object1Str1.Value
# Enumerate the properties of the object contained in .Array2's first element
# Do NOT use .GetEnumerator() here - enumerate the array *itself*
foreach($o in $nsObj.Array2[0]){
"Path is: $($o.Path)"
"Parent is: $($o.Parent.ToString())"
}
[1] PowerShell Core - but not Windows PowerShell - currently (v7) actually uses NewtonSoft's Json.NET behind the scenes.
You have a few separate questions here:
$nsObj[0] #Returns nothing. Why?
This is because nsObj corresponds to a JSON object, and, as explained in this answer to How to get first key from JObject?, JObject does not directly support accessing properties by integer index (rather than property name).
JObject does, however, implement IList<JToken> explicitly so if you could upcast nsObj to such a list you could access properties by index -- but apparently it's not straightforward in PowerShell to call an explicitly implemented method. As explained in the answers to How can I call explicitly implemented interface method from PowerShell? it's necessary to do this via reflection.
First, define the following function:
Function ChildAt([Newtonsoft.Json.Linq.JContainer]$arg1, [int]$arg2)
{
$property = [System.Collections.Generic.IList[Newtonsoft.Json.Linq.JToken]].GetProperty("Item")
$item = $property.GetValue($nsObj, #([System.Object]$arg2))
return $item
}
And then you can do:
$firstItem = ChildAt $nsObj 0
Try it online here.
#??? Why can't I see the Path property like when just output $nsObj.Array2[0] ???
The problem here is that JObject.GetEnumerator() does not return what you think it does. Your code assumes it returns the JToken children of the object, when in fact it is declared as
public IEnumerator<KeyValuePair<string, JToken>> GetEnumerator()
Since KeyValuePair<string, JToken> doesn't have the properties Path or Parent your output method fails.
JObject does implement interfaces like IList<JToken> and IEnumerable<JToken>, but it does so explicitly, and as mentioned above calling the relevant GetEnumerator() methods would require reflection.
Instead, use the base class method JContainer.Children(). This method works for both JArray and JObject and returns the immediate children in document order:
foreach( $o in $nsObj.Array2[0].Children() ){
"Path is: $($o.Path)"
"Parent is: $($o.Parent)"
}
Try it online here.
$nsObj.Array1 #Again nothing. Maybe because it contains no key:value pairs?
Actually this does return the value of Array1, if I do
$nsObj.Array1.ToString()
the JSON corresponding to the value of Array1 is displayed. The real issue seems to be that PowerShell doesn't know how to automatically print a JArray with JValue contents -- or even a simple, standalone JValue. If I do:
$jvalue = New-Object Newtonsoft.Json.Linq.JValue 'my jvalue value'
'$jvalue' #Nothing output
$jvalue
'$jvalue.ToString()' #my jvalue value
$jvalue.ToString()
Then the output is:
$jvalue
$jvalue.ToString()
my jvalue value
Try it online here and, relatedly, here.
Thus the lesson is: when printing a JToken hierarchy in PowerShell, always use ToString().
As to why printing a JObject produces some output while printing a JArray does not, I can only speculate. JToken implements the interface IDynamicMetaObjectProvider which is also implemented by PSObject; possibly something about the details of how this is implemented for JObject but not JValue or JArray are compatible with PowerShell's information printing code.
Related
I am trying to convert a dict into JSON format and not seeing any easy method using TclLib Json Package. Say, I have defined a dict as follows :
set countryDict [dict create USA {population 300 capital DC} Canada {population 30 capital Ottawa}]
I want to convert this to json format as shown below:
{
"USA": {
"population": 300,
"captial": "DC"
},
"Canada": {
"population": 30,
"captial": "Ottawa"
}
}
("population" is number and capital is string). I am using TclLib json package (https://wiki.tcl-lang.org/page/Tcllib+JSON) . Any help would be much appreciated.
There's two problems with the “go straight there” approach that you appear to be hoping for:
Tcl's type system is extremely different to JSON's; in Tcl, every value is a (subtype of) string, but JSON expects objects, arrays, numbers and strings to wholly different things.
The capital becomes captial. For bonus fun. (Hopefully that's just a typo on your part, but we'll cope.)
I'd advise using rl_json for this; it's a much more capable package that treats JSON as a fundamental type. (It's even better at it when it comes to querying into the JSON structure.)
package require rl_json
set result {{}}; # Literal empty JSON object
dict for {countryID data} $countryDict {
rl_json::json set result $countryID [rl_json::json template {{
"population": "~N:population",
"captial": "~S:capital"
}} $data]
# Yes, that was {{ … }}, the outer ones are for Tcl & the inner ones for a JSON object
}
puts [rl_json::json pretty $result]
That produces almost exactly the output you asked for, except with different indentation. $result is the “production” version of the output that you can work with for further processing, but which has no excess whitespace at all (which is a great choice when you're dealing with documents over 100MB long).
Notes:
The initial JSON object could have been done like this:
set result "{}"
that would have worked just as well (and been the same Tcl bytecode).
json set puts an item into an object or array; that's exactly what we want here (in a dict for to go over the input data).
json template takes an optional dictionary for mapping substitution names in the template to values; that's perfect for your use case. Otherwise we'd have had to do dict with data {} to map the contents of the dictionary into variables, and that's less than perfect when the input data isn't strictly controlled.
The contents of template argument to json template is itself JSON. The ~N: prefix in a leaf string value says “replace this with a number from the substitution called…”, and ~S: says “replace this with a string from the substitution called…”. There are others.
I am currently struggling to convert a JSON-string into an array of objects and to GENERALLY handle the properties/attributes of each object.
Here is a simple demo, that shows that e.g. the attribute "address" seems to be a bit special:
cls
$json = '[{"id":"1","address":"1"},{"id":"2","address":"2"}]'
$list = $json | ConvertFrom-Json
$list.id # OK
$list.address # gives a weired result - is this a bug?
$list.GetEnumerator().address # that works
This is the output:
1
2
OverloadDefinitions
-------------------
System.Object&, mscorlib, Version=4.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089 Address(int )
1
2
As you can see, I need to add ".GetEnumerator()" to get the correct "address"-values.
Is this the expected? Should I ALWAYS use the ".GetEnumerator()" to be safe?
Since $list is an array (collection), using something like $list.id performs member-access enumeration; that is, the .id property is automatically accessed on each element, and the resulting values are returned as an array ([object[]])[1].
However, if the array / collection type itself has a member by that name (a property or method), it takes precedence, which is what happened in your case: .NET arrays have an .Address method (that is added by the runtime - see this answer), and that is what preempted accessing the elements' .Address property.
(What you saw was PowerShell's representation of the method's signature (overload), which is what you get when you access a method by name only, without actually calling it by appending ((...)); try 'foo'.ToUpper, fo instance.)
That PowerShell provides no distinct operator syntax to distinguish between direct member access and member-access enumeration is the subject of [GitHub issue #7445](https://github.com/PowerShell/PowerShell/issues/7445), but the existing behavior is unlikely to change.
The most efficient way to work around that problem is to use the PSv4+ .ForEach() array method, which always targets the collection's elements:
$list = '[{"id":"1","address":"1"},{"id":"2","address":"2"}]' | ConvertFrom-Json
$list.ForEach('address')
Caveat: If you're running Windows PowerShell and $list can situationally just result in a single pscustomobject rather than an array, you must enclose $list in #(...), the array-subexpression operator, to ensure that the .ForEach() method is available. This is a [pscustomobject]-specific bug (given that even single objects should consistently have a .ForEach() method), which has been fixed in PowerShell (Core) 6+.
# #(...) is necessary in Windows PowerShell only.
#($list).ForEach('address')
Note:
Technically, this returns a [System.Collections.ObjectModel.Collection[PSObject]] collection, but in most cases you can use it like a regular [object[]] PowerShell array.
Unlike with member-access enumeration, a collection instance is also returned if there's only one element in the input collection (rather than returning that one element's property value as-is, the way member-access enumeration does).
In PSv3- you can use the ForEach-Object cmdlet, or Select-Object -ExpandProperty:
$list | ForEach-Object address # PSv2: | ForEach-Object { $_.address }
# OR
$list | Select-Object -ExpandProperty address
[1] If there's only one element in the collection, its property value is returned as-is, not wrapped in a (single-element) array, which is the same logic that is applied when collecting pipeline output in a variable. That this logic may be unexpected in the context of member-access enumeration, which is an expression context, is being discussed in GitHub issue #6802.
I have a JSON file with a structure like this:
[
obj1,
obj2,
...
objN
]
All of the sub-objects are entirely self-contained, i.e. there are no cross references between them. The problem is that the file as a whole is huge ( >100k entries in the root array).
Is there any way in Jackson to stream in the contents of the root array via databinding, such that the root array never resides fully in main memory? I would like to avoid the low level JsonGenerator/JsonParser API.
Yes. Check out ObjectReader (constructed using various methods in ObjectMapper, like .readerFor(ElementType.class)), and then its readValues() method, which returns MappingIterator<ElementType> (for whatever type you use). This method will only bind one item at a time.
If the values are in root-level array, this should work as is. If they were somewhere deeper in JSON structure, you would have to construct JsonParser first, then iterate (with nextToken()) to the first value, but after that you could still create MappingIterator for efficient item-by-item binding.
I am returning some xml structure as json using the built-in MarkLogic json module. For the most part it does what I expect. However, when an element marked as an array is empty, it returns an empty string instead of an empty array. Here is an example:
xquery version "1.0-ml";
import module namespace json = "http://marklogic.com/xdmp/json"
at "/MarkLogic/json/json.xqy";
let $config := json:config("custom")
return (
map:put( $config, "array-element-names", ("item") ),
json:transform-to-json(<result>
<item>21</item>
<item>22</item>
<item>23</item>
</result>, $config),
json:transform-to-json(<result></result>, $config))
Result:
{"result":{"item":["21", "22", "23"]}}
{"result":""}
I would expect an empty array if there were no items matching in the array-element-name called "item". i.e.
{"result":{"item":[]}}
Is there some way to configure it so it knows the element is required ?
No - it will not create anything that is not there. In your case, what if the XML was more complex. There is no context of 'where' such an element might live - so it could not create it even if it wanted to.
Solution is to repair the content if needed by adding one element - or transforming it into the json-basic namespace - where those elements live inside of of an element noted as an array (which can be empty) - or third, use an XSD to hint to the processor what to do . But that would still need a containing element for the 'array' - and then the items would be minOccurance=0. But if this is the case, then repair and transform into the json/basic namespace is probably nice and simple for your example.
I'm using JsonCpp v0.6.0 to parse the following JSON string:
{
"3.7":"de305d54-75b4-431b-adb2-eb6b9e546011",
"3.7":"de305d54-75b4-431b-adb2-eb6b9e546012",
"3.8":"de305d54-75b4-431b-adb2-eb6b9e546013"
}
as follows:
Json::Value root;
Json::Reader reader;
// value contains the JSON string
if (!reader.parse(value, root, false))
{
// parse error
}
After the call to parse, root contains two entries in a map:
[0] first = "3.7", second = "de305d54-75b4-431b-adb2-eb6b9e546012",
[1] first = "3.8", second = "de305d54-75b4-431b-adb2-eb6b9e546013",
i.e. the first JSON record has been overwritten by the second. No errors are reported.
Is this behaviour expected? Is it correct?
I thought that an error might have been reported indicating that there is a duplicate key in the JSON string.
Like the JSON RFC sad the object names (keys) should be unique.
The names within an object SHOULD be unique.
Also the RFC defines if they are not, the behavior is unpredictable.
See this quote from the RFC:
An object whose names are all unique is interoperable in the sense
that all software implementations receiving that object will agree on
the name-value mappings. When the names within an object are not
unique, the behavior of software that receives such an object is
unpredictable. Many implementations report the last name/value pair
only. Other implementations report an error or fail to parse the
object, and some implementations report all of the name/value pairs,
including duplicates.
I agree with what you say, but I think JsonCpp tries to be a helpful tool, not something that tries to scrape by with minimal conformance to the RFCs.
It would make more sense if it either maintained the structure of the input stream, and supported duplicate keys, or (and this is what the OP and I'd expect) if it doesn't like it, to flag an error.
Silently changing the structure is unhelpful, as a check for the validity of the JSON input would have to be made with some other JSON tool prior to sending it to JsonCpp.