Check to make sure children are not nested lists. Fix plotly/dash-renderer#23

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340

[rmarren1]

Proposed fix to plotly/dash-renderer#23

This solution borrows heavily from #273. Basically the same except instead of checking if values in the tree are a member of a valid set, we just check if the value is not a nested list.

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

Examples:

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which is not JSON serializable.
The value in question is located at
[*] list and has string representation
`['nested']`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

dash.exceptions.InvalidCallbackReturnValue: The callback for property `children` of component `output`
returned a tree with one value having type `list`
which can not be serialized by Dash.
The value in question is located at
[*] Div and has string representation
`[['nested']]`
In general, Dash properties can only be
dash components, strings, dictionaries, numbers, None,
or un-nested lists of those.

[chriddyp]

A major difference is that this check if performed before each callback rather than being called only if json.dumps fails. In this case, json.dumps succeeds since nested arrays are valid JSON.

This is an interesting approach. I'm a little worried about:

• Performance issues in crawling the tree, especially with really large apps
• Keeping this in check with our json serializer. I believe that children=pandas.Series(...) is technically valid but it would fail in this case.

---

Architecturally, I originally imagined this working on the component-level. That way, we could verify:

• All properties, not just the children prop
• Not just primative types but also shapes and nested types

In the most general solution, I imagined this as being part of the base Component and doing type-checking with the metadata that we are provided through propTypes. That way, the component author could specify the shapes of their props and the validation would automatically be done. I have some more details about this here: #264

If we ended up implementing #264, then I think that we would get this issue (children as nested children) for free. Of course, there are still some issues:

• The higher-level types. Technically a pandas.Series should be considered valid just like a list. Perhaps for each propType we implement a list of valid Python types or validators

{
'array': [pandas.Series, list, numpy.array]
}

• We might want to make it configurable (similar to suppress_callback_exceptions, something like suppress_component_validation_exceptions), in case the component author didn't fully annotate their proptypes

[rmarren1]

closed in favor of more general #340