Implement Nested Reverse Conversion Method from CSV to JSON about json-csv

[intelliapps]

Clearly there is an issue with the actual JSON but do you care whether the output is formatted, give that an IDE or other tool can format JSON to make it more readable when that is required?

[ozdemirburak]

No, formatting is not a problem actually, just setting the encoding option in the end like below will also take care of it.

$csv->setConversionKey('options', JSON_PRETTY_PRINT|JSON_UNESCAPED_UNICODE|JSON_UNESCAPED_SLASHES);

[intelliapps]

Is it safe to assume that the JSON key names in the original input will never contain the underscore character? Otherwise I don't think the can be fully resolved by replacing convertAndSave()?

[intelliapps]

Perhaps the library could be modified so that users can specify their own delimiter char/string for joining / splitting the csv column names, based on their knowledge of the json they are passing in? The potential issue with this is what characters it is acceptable to put into CSV column names, although I don't think there is a definitive specification for that.

[ozdemirburak]

This is a great question, and I have not thought about it before actually. I think instead of forcing to join with underscore, developer should decide what operator will be used.

Maybe, this line can be changed like below.

$result = array_merge($result, $this->flatten($value, $prefix . $key . $this->conversion['join']));

[intelliapps]

Great minds think alike ;0)

[ozdemirburak]

This was funny and impressive actually :)

[intelliapps]

I'll have a look at fixing this, if that's ok?

[ozdemirburak]

Sure, it will be great, thanks.

[intelliapps]

As per the comments on my pull request: I've implemented changes to get the output of csv->json to match the input to json->csv. The result is a significant improvement but there are still issues which I don't think can be resolved without changing the conversion from json to csv, or at least without further consideration.

1. Previously, the json output represented all values as strings, wrapped in double quotes. As the example input above contains numeric values, along with most of the other test data, this resulted in a mismatch between the input and output json. I have resolved this by adding a 'numbers' conversion option which defaults to 'strings' but when set to 'numbers' causes all numeric strings to be converted to numeric types. This could still result in mis-matching input / output json where:
   a) input contains numeric f.p. values with 0 fractional value, e.g. 6.0 is output as 6
   b) input contains a mix of numbers and numeric strings

2. With the stats data, the name of each object from the stats.json file is lost when it is converted to csv, e.g. "kemal-kilicdaroglu" and "muharrem-ince". This is consistent with the stored stats.csv test file, which does not contain these values. Since these values are not in the csv at all I can't even consider how to reinstate them into the output json to make it match the input. A related issue is that the outer braces are then {} where they should be [].

3. With the stats data the "avg" values from the input json are rounded to fewer digits during the json to csv conversion, which means that it is not possible for the values in the output json to match.