Avoid using '#' in float format specifier for float attributes for valid JSON

[dougvj]

Recently we encountered a data set that produced output values on variable attribute floats with trailing decimals, for example:

"resolution":1000000.,

This not valid JSON and broke parsing.

I was able to recreate the issue with the test data set in the repo, under tests:

../ncdump-json -j -h -p 1 ./sresa1b_ncar_ccsm3-example.nc

Produces this in the output:

"missing_value":1.e+20

~This PR fixes the problem by detecting the case where a . is followed by a non digit and adds a single 0.~

Now fixed by using the standard variable format specifier which does not have '#'

[jllodra]

Thank you for your contribution, nice executed.

Indeed, that is not valid in JSON (just for the sake of curiosity, let me say that it would be valid as a JavaScript object).

Let me ask you something. I see that you are adding a trailing 0, for example 1.e+20 => 1.0e+20, if I have undestood correctly.

10000. => 10000.0

Wouldn't be an option to remove the . when it is not followed by a number?

[jllodra]

After reading your code comment about the # flag being the culprit. (https://en.wikipedia.org/wiki/Printf_format_string)

In https://github.com/jllodra/ncdump-json/blob/master/src/dumplib.c#L109 you could define also the variables "json_float_att_fmt", "json_float_attx_fmt", and "json_double_att_fmt", with the same format specification but removing the #.

Then, in https://github.com/jllodra/ncdump-json/blob/master/src/ncdump.c#L447, you could write if (is_json) then do the snprintf using the json-specific format, else do what it was doing before.

What do you think about this approach?

[dougvj]

I thought it might be desirable to maintain the float type rather than produce a JSON int type in the case that scientific notation is not produced. However, for our needs this doesn't matter and I would be happy to switch this to dropping the decimal instead if you prefer. 

[dougvj]

Yes this approach would be fine too. 

[jllodra]

If you are OK with it (and also if it solves your problem), I think the best approach would be the latter (dropping the # for json). For sake of code simplicity, and would avoid manipulation of text

[dougvj]

OK thanks for the feedback. I pushed this change and edited the title to reflect.

[jllodra]

Excellent, thank you for your contribution

https://github.com/jllodra/ncdump-json/releases/tag/r17