Combine json_alloc_token and json_fill_token

[alexhenrie]

JSMN has been initializing tokens with default values, then initializing them again with the real values. The first step can be skipped to make JSMN smaller and faster.

[pt300]

Interesting proposal. Have you checked that JSMN never really depends on those default values in cases such as missing JSON?

[alexhenrie]

Two of the three calls to json_fill_token are immediately after json_alloc_token. The only place where json_fill_token is called later, and therefore behavior could change, is when JSMN_STRICT is defined and an object or array is used as a key. In that case, token->type was previously uninitialized and token->start was initialized to -1 before returning JSMN_ERROR_INVAL. With this patch, token->type and token->start are set before erroring out.

JSMN itself doesn't do anything with those values before returning, so really we're just talking about what the caller should see if the caller looks at the tokens despite jsmn_parse returning JSMN_ERROR_INVAL. In my opinion, initializing the last token (the bad key) with accurate values before returning an error is more useful than leaving the values uninitialized or invalid. However, if it's important, I could easily reset token->type to JSMN_UNDEFINED and token->start to -1 on the error path. Just let me know and I'll modify the pull request.