Feature request: more options for reporting errors

[bwbecker]

Hi -- Just getting acquainted with jsoniter. Looks like a lovely library; thanks for the efforts.

However (you knew that was coming, right?), I wish there were more options for reporting errors than the message from JsonReaderException:

missing required field "student_take_units", offset: 0x00007255, buf:
+----------+-------------------------------------------------+------------------+
|          |  0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f | 0123456789abcdef |
+----------+-------------------------------------------------+------------------+
| 00007230 | 73 74 75 64 65 6e 74 5f 74 61 6b 65 5f 75 6e 69 | student_take_uni |
| 00007240 | 74 73 58 58 58 22 3a 20 30 2e 31 33 0a 20 20 20 | tsXXX": 0.13.    |
| 00007250 | 20 20 20 20 20 7d 2c 0a 20 20 20 20 20 20 20 20 |      },.         |
| 00007260 | 7b 0a 20 20 20 20 20 20 20 20 20 20 20 20 22 73 | {.            "s |
| 00007270 | 75 62 6a 65 63 74 5f 63 6f 64 65 22 3a 20 22 57 | ubject_code": "W |
+----------+-------------------------------------------------+------------------+

This isn't a very user-friendly error message.

What I'd love is easy ways to get the error message ("missing required field.."), the offset as an integer line number plus offset into the line (base 10), and access to the buffer so I can format it as appropriate for my users.

[plokhotnyuk]

Hi, @bwbecker!

Thanks for trying jsoniter-scala and sending feedback here!

The library developed to be safe, correct and performant so we cannot implement roughly defined or costly in runtime features.

But I will happy to help you solve any concrete use case.

Feel free to reach me here or in the Gitter chat for any question or discussion.

[bwbecker]

Thanks for the quick reply, @plokhotnyk!

My concrete use case is to provide a comprehensible error message to an end user rather than to a developer. The end user isn't going to understand a hex offset and the hex dump will scare them off of even trying to read the error message.

Currently I'm parsing the error message from JsonReaderException to get just the error message and hex offset. I then read the file until the offset, counting lines so I can present something like this to the user:

missing required field "student_take_units" @ line 959
957             "student_take_unitsXXX": 0.13
958         },
959         {
960             "subject_code": "WKRPT",

I understand that the input may not always have distinct lines, like mine tend to, so an offset into the line would also be important.

But the main point is that the information should be accessible in the JsonReaderException object without parsing the error message. The offset is obviously already available. I think all the information is readily accessible when JsonReaderException is thrown at line 773 in JsonReader.scala (the decodeError method) except for the line number. I'd guess that counting the lines would have a very minor impact on performance. As for the buffer, perhaps just provide it as a string with information about the line & offset where it starts.

Thanks for considering this suggestion!

[plokhotnyuk]

@bwbecker You are right that error messages are formatted for developers (not for users who afraid hex dumps ;)

I see you have a workaround by rereading an affected file, counting lines until the error offset and collecting required lines for the final error for your users.

But I don't see how the proposed built-in line counter can be implemented without significant extra penalty in runtime and without severe security risks for cases when malicious counterpart will use weakness of line collecting implementation.

Feel free to propose a PR if you see some safe and efficient implementation.