Fix error pointer behaviour of cJSON_ParseWithOpts()

[rmallins]

What is
cJSON_ParseWithOpts() doesn't set the error pointer on parse error. This differs from cJSON_Parse() and also from original behaviour.
What should be
cJSON_ParseWithOpts() should set the error pointer on parse error.

Detail
I started using cJSON at work around a year and a half ago, and wrote unit tests around that version of the code. When I updated my copy of the source with the latest code from this repo last week I noticed that my tests failed due to the change in error handling behaviour for cJSON_ParseWithOpts():

original behaviour on error

• parse_end pointer == NULL
• error pointer == last parsed byte + 1

current behaviour on error

• parse_end pointer == last parsed byte + 1
• error pointer == NULL

proposed behaviour on error

• parse_end pointer == last parsed byte + 1
• error pointer == last parsed byte + 1

It makes good sense that the parse_end pointer is now updated. However it looks like a bug to fail to update the error pointer in the error case. It violates the principle of least surprise, and is also an interface change for no good reason that I can see.

My change updates the unit tests to check for the new behaviour, and updates the code to match.

I developed against master before discovering a minute ago that you need changes to go on the "develop" branch, so I submitted the merge request to develop. As will be obvious it's only the last 3 commits that are mine. My apologies for the clutter!

regards,
Robin Mallinson

[FSMaxB]

You are right, it is inconsistent not to set the global error pointer if a return_parse_end was given.

I regard this as a bug fix, not an intrusive change, so please redo the pull request on top of master (sorry for the confusion).

Also please ensure that every single commit passes the unit tests --> combine the RED and GREEN commits into one commit.

Thanks for your fixes.

If you have more unit tests lying around, please let me know. More pull requests welcome.

[rmallins]

OK, have merged the 2nd and 3rd changes into one commit and reinstated the change over master. I doubt I can deliver the unit tests I've written at work but if I get some spare time I'll see what I can add on my own time.

[FSMaxB]

Oh, I forgot that the appveyor.yml is only on the develop branch. So ignore the failed AppVeyor build.