Sometimes I write parsing code next to each other, for example:

auto height = doc["data"]["height"].get_double().value();
auto weight = doc["data"]["weight"].get_double().value();
...
auto first_name = doc["data"]["first_name"].get_string().value();
auto last_name = doc["data"]["last_name"].get_string().value();

There could be 10+ fieilds to read. However if I get a INCORRECT_TYPE exception on production, I couldn't figure out which field it's talking about so I have to either log the json or setup debugging. Do you have better way?

I definitely can try to solve the issue during dev testing, but as you know, client sides can wire whatever json on production so it's still better to have a message indicating which field/value is wrong.

I read the code and see the error messages are pretty static, but there could be a way to put field/value infomation into the exception so what() method can show it. Exception is already expensive so building a string isn't expected to add much.

Thanks.

I would discourage you from writting the code in this manner. It is probably quite inefficient (computationally). You are constantly looking up the field "data". Don't do that!

You should do...

ondemand::object dataobject;
error = doc["data"].get(dataobject);
if(error) {...}

Exceptions make for clean code, but they also make it harder to have precise error control. Please consider using the approach without exceptions... it is more verbose, but it will give you tight control...

error = dataobject["height"].get(val);
if(error) {...}
error = val.get_double().get(height);
if(error) {...}

and so forth. At each step you can include the precise error handling you need, instead of relying on a crude try/catch.

Thanks for your advice on performance.
I do see the way to handle error after each read without expecting an expcetion. It's just a little trivial to me given there are 10+ feilds there to handle, and all the code can do is still throwing an exception with field/value information. Maybe I can use an inline function to save some effort.

You definitely can use your own functions to reduce the programming burden and still have the exact error handling you seek.

Note that I take very seriously feedback such as yours.

Wow.

Your project (simdjson_peval) looks very interesting.

cc @jkeiser