[Question]: Can't deserialize in HashMap<String, String> if there is a subelement

[Pastequee]

Problem Description

Hello, I want to deserialize a list of unknown elements, that contains either a String literal or a sub element, but in case of a sub element I just want a default empty String. And I want to store this in an HashMap<String, String>

What I've found so far

Let say I have this XML

<Data>
    <A>value1</A>
    <B>value2</B>
    <C>value3</C>
    <D>
        <Nested>Breaks !</Nested>
    </D>
</Data>

Where A, B, C have unknown name and their values are strings. I have this code struct that deserialize just well if I have only A, B and C.

#[derive(Debug, Serialize, Deserialize)]
#[serde(rename_all = "PascalCase")]
struct Event {
    #serde[(default)]
    data: HashMap<String, String>,
};

Here, if all my sub elements A, B and C contains only strings it works just well, but if I have a nested element in one of them like in D I have a UnexpextedStart([78,101,115,116,101,100]) error (the int array represent the ascii codes for 'Nested').
So my question is, is there a way to either skip D if there is a nested element that is not a String, or define a default value like "" in those case to not lose all the rest of the data.
Thank you for your time 😄

[Mingun]

You should use a custom type for the value of map that would be able to be deserialized using either Visitor::visit_str or Visitor::visit_map. The String::deserialize only expect Visitor::visit_str. It seems that untagged enum

#[derive(Deserialize)]
#[serde(untagged)]
enum MyString {
  Str(String),
  Nested,
}

should do the work, but probably manually implemented Deserialize will be more ergonomic.

Because this issue is duplicate of #383, I close it.

[treywelsh]

Hi,

It seems I have a similar problem, I want to capture keys and values in a HashMap (the difference is that I'm using HashMap temporarily, for the time of the deserialization) and I'm interested in the solution for this issue.
I already asked my question on stack overflow https://stackoverflow.com/questions/75337112/deserializing-dynamic-xml-in-rust

I didn't find a way to make this work via untagged enum (I wasn't able to capture key and values) so I took the "Implement Deserialize for a custom map type" serde example to update it a bit:

#![allow(dead_code)]

use serde::Deserialize;
use std::{collections::HashMap, fmt};

use serde::de::{self, Deserializer, MapAccess, Visitor};

const XML: &str = r#"
<vm>
    <id>15</id>
    <template>
        <tag1>content1</tag1>
        <tag2>content2</tag2>
        <vec1>
            <tag3>content3</tag3>
            <tag4>content4</tag4>
        </vec1>
        <vec1>
            <tag3>content5</tag3>
        </vec1>
        <tag2>content6</tag2>
        <vec2>
            <tag3>content7</tag3>
        </vec2>
    </template>
</vm>
"#;

#[derive(Debug, Clone)]
struct Pair(String, String);

#[derive(Debug, Clone)]
struct Vector(String, Vec<Pair>);

#[derive(Debug, Clone)]
struct Template {
    pairs: Vec<Pair>,
    vectors: Vec<Vector>,
}

impl Template {
    fn new() -> Self {
        Template {
            pairs: Vec::new(),
            vectors: Vec::new(),
        }
    }
}

#[derive(Debug, Deserialize)]
pub struct VM {
    id: i64,
    template: Template,
}

struct TemplateVisitor;

impl<'de> Visitor<'de> for TemplateVisitor {
    type Value = Template;

    fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
        formatter.write_str("a very special map")
    }

    fn visit_map<M>(self, mut access: M) -> Result<Self::Value, M::Error>
    where
        M: MapAccess<'de>,
    {
        let mut map = Template::new();

        while let Some(key) = access.next_key::<String>()? {
            let map_value = access.next_value::<HashMap<String, String>>().unwrap();

            if map_value.contains_key("$text") {
                map.pairs
                    .push(Pair(key, map_value.get("$text").unwrap().clone()));
            } else {

                let mut vector = Vec::new();
                for (k, v) in map_value {
                    vector.push(Pair(k, v))
                }
                map.vectors.push(Vector(key, vector));
            }
        }

        Ok(map)
    }
}

impl<'de> Deserialize<'de> for Template {
    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
    where
        D: Deserializer<'de>,
    {
        deserializer.deserialize_map(TemplateVisitor {})
    }
}

fn main() {
    let obj: VM = quick_xml::de::from_str(XML).unwrap();
    println!("{:#?}", obj);
}

Did you find an other interesting solution @arthur-pi ?
I'm not happy with the test map_value.contains_key("$text"), I'm feeling like I'm doing something dirty. Any tip to share @Mingun ?

Sorry if this comment was not appropriate in this issue, I wasn't sure that #383 was really a duplicate of this one ?

[Mingun]

I do not see abilities to avoid map_value.contains_key("$text") hack. What's happens there:

// `key` deserialized from
// - <tag1>
// - <tag2>
// - <vec1>
// - <vec1>
// - <tag2>
// - <vec2>
while let Some(key) = access.next_key::<String>()? {
    // map_values deserialized from ... to
    // - <tag1> content1 </tag1>
    //          ^^^^^^^^              -- map: { $text: content1 }
    // - <tag2> content2 </tag2>
    //          ^^^^^^^^              -- map: { $text: content2 }
    // - <vec1> <tag3>content3</tag3><tag4>content4</tag4> </vec1>
    //          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -- map: { tag3: content3, tag4: content4 }
    // - <vec1> <tag3>content5</tag3> </vec1>
    //          ^^^^^^^^^^^^^^^^^^^^^ -- map: { tag: content5 }
    // - <tag2> content6 </tag2>
    //          ^^^^^^^^              -- map: { $text: content6 }
    // - <vec2> <tag3>content7</tag3> </vec2>
    //          ^^^^^^^^^^^^^^^^^^^^^ -- map: { tag3: content7 }
    let map_value = access.next_value::<HashMap<String, String>>().unwrap();

    if map_value.contains_key("$text") {
        map.pairs
            .push(Pair(key, map_value.get("$text").unwrap().clone()));
    } else {

        let mut vector = Vec::new();
        for (k, v) in map_value {
            vector.push(Pair(k, v))
        }
        map.vectors.push(Vector(key, vector));
    }
}

PS. I would better to use HashMap::remove instead of contains_key + get.

https://serde.rs/string-or-struct.html example does not work here, because both deserialize_str and deserialize_map can handle content1 and <tag3>content5</tag3> and return

• String(content1) and String(content5)
• HashMap{ $text: content1 } and HashMap{ tag: content5 }

so enum couldn't distinguish between them (it relies on deserialization failure). Moreover, because deserializer is consumed during deserialization, (Deserialize::deserialize takes ownership) the deserialization is performed into an internal serde::de::__private::Content type, which uses deserialize_any.

Probably, if we'll change our deserialize_any implementation, we could solve your case by using untagged enum

#[derive(Deserialize)]
#[serde(untagged)]
enum {
  String(String),
  Map(HashMap<String, String>),
}

quick-xml/src/de/mod.rs

Lines 2525 to 2536 in 64292c7

	fn deserialize_any<V>(self, visitor: V) -> Result<V::Value, DeError>
	where
	V: Visitor<'de>,
	{
	match self.peek()? {
	DeEvent::Start(_) => self.deserialize_map(visitor),
	// Redirect to deserialize_unit in order to consume an event and return an appropriate error
	DeEvent::End(_) | DeEvent::Eof => self.deserialize_unit(visitor),
	_ => self.deserialize_string(visitor),
	}
	}
	}

The self.peek() here peeks a <tag1>, <tag2> and so on.

Feel free to investigate this! That could open a rabbit hole!

[treywelsh]

Thanks for taking time to answer !

[codingbaobao]

Hi, I have a similar question to transform into hashmap

<imgdir name="info">
    <int name="version" value="10" />
    <float name="speed" value="1.6" />
    <string name="path" value="a/b/c" />
</imgdir>

I have an Enum to handle different types

enum Value {
    Int(i32),
    Float(f32),
    String(String),
}

And I want to transform this xml into HashMap<@name,@value>

#[derive(Debug, Clone)]
struct MyMap(HashMap<String, Value>);

struct Imgdir {
    #[serde(rename = "@name")]
    name: String,
    #[serde(rename = "$value")]
    map: MyMap,
}

I implement Visitor trait like @treywelsh did, however seems like it gets error on access.next_value::<HashMap<String, String>>().
Could you give some tips to get it right? Or it just can't :( @Mingun