This code was working some months ago:

import std.json;
JSONValue total;
total.type = JSON_TYPE.OBJECT;
total["results"].type = JSON_TYPE.OBJECT;


Now, last line gives me a segmentation fault. What's wrong with this code?

On Monday, 4 February 2013 at 10:04:10 UTC, Andrea Fontana wrote:
> This code was working some months ago:
>
> import std.json;
> JSONValue total;
> total.type = JSON_TYPE.OBJECT;
> total["results"].type = JSON_TYPE.OBJECT;
>
>
> Now, last line gives me a segmentation fault. What's wrong with this code?

https://github.com/D-Programming-Language/phobos/pull/561/files

Is this code right? Has it ever worked? I don't think returning a ref value of an AA works for adding items to an AA.

That pull request didn't come with any unit tests... shouldn't have made it through review, IMO.

On 2/4/2013 2:04 AM, Andrea Fontana wrote:
> This code was working some months ago:
>
> import std.json;
> JSONValue total;
> total.type = JSON_TYPE.OBJECT;
> total["results"].type = JSON_TYPE.OBJECT;
>
>
> Now, last line gives me a segmentation fault. What's wrong with this code?

1. That code is not valid D source - it's presumably missing the function it is embedded inside. Please post valid D source when a bug is suspected, otherwise people will have to guess at what the rest is, and often the rest is where the real issue was.

2. Regressions (and other bugs) should be posted to bugzilla, along with a reproducible test case.

On Monday, 4 February 2013 at 21:16:51 UTC, Vladimir Panteleev wrote:
> On Monday, 4 February 2013 at 10:04:10 UTC, Andrea Fontana wrote:
>> This code was working some months ago:
>>
>> import std.json;
>> JSONValue total;
>> total.type = JSON_TYPE.OBJECT;
>> total["results"].type = JSON_TYPE.OBJECT;
>>
>>
>> Now, last line gives me a segmentation fault. What's wrong with this code?
>
> https://github.com/D-Programming-Language/phobos/pull/561/files
>
> Is this code right? Has it ever worked?

I think it depends on how std.json is defined. It is up to implementator to define whether such things should be supported or not.

> I don't think returning a ref value of an AA works for adding items to an AA.

He doesn't add item to AA. The code just queries a reference to item from array and since opIndex returns null, any attempt to construct a stack variable from null reference leads to segfault. By no means the code can manipulate AA using returned struct since it is copied, even if opIndex returns by reference.

> That pull request didn't come with any unit tests... shouldn't have made it through review, IMO.

On Monday, 4 February 2013 at 10:04:10 UTC, Andrea Fontana wrote:
> This code was working some months ago:
>
> import std.json;
> JSONValue total;
> total.type = JSON_TYPE.OBJECT;
> total["results"].type = JSON_TYPE.OBJECT;
>
>
> Now, last line gives me a segmentation fault. What's wrong with this code?

Here is a full test case. I don't want to submit it to Bugzilla yet, since I don't know whether it's intended behaviour or a bug. I have a slight feeling that this is NOT a bug. Somehow, in the first case a reference to an invalid struct is returned by json["hello"] and attempted to be overridden by a new JSONValue. In the second case, no reference is generated, instead a simple struct copy (a blit) is invoked, which works. Can someone explain this more correctly?

This program crashes (tested on dmd 2.061)
import std.json;
void main() {
  JSONValue json;
  json.type = JSON_TYPE.OBJECT;
  json["hello"] = JSONValue();
}

This program works fine:
import std.json;
void main() {
  JSONValue json;
  json.type = JSON_TYPE.OBJECT;
  json.object["hello"] = JSONValue();
}

On Wednesday, 6 February 2013 at 09:41:28 UTC, Stephan wrote:
> Here is a full test case. I don't want to submit it to Bugzilla yet, since I don't know whether it's intended behaviour or a bug. I have a slight feeling that this is NOT a bug. Somehow, in the first case a reference to an invalid struct is returned by json["hello"] and attempted to be overridden by a new JSONValue. In the second case, no reference is generated, instead a simple struct copy (a blit) is invoked, which works.

This depends on whether such code is supposed to be supported or not. Currently it is not supported and should no be used. But you can try write enhancement pull.

> Can someone explain this more correctly?
>
> This program crashes (tested on dmd 2.061)
> import std.json;
> void main() {
>   JSONValue json;
>   json.type = JSON_TYPE.OBJECT;
>   json["hello"] = JSONValue();
> }

opIndex returns by ref an element of associative array as an rvalue. This means that if absent, a new element is not added. In your example array is empty. Json library returns null (because it returns by ref) and the first instruction which attempts to copy object from returned pointer to stack segfaults.

> This program works fine:
> import std.json;
> void main() {
>   JSONValue json;
>   json.type = JSON_TYPE.OBJECT;
>   json.object["hello"] = JSONValue();
> }

In this case a new element of AA is created.

Yes it works in that way. I had to change it in a lot of source files.
It worked fine for months as online webservice used by thousands users and never crashed :)

With last changes to code (a db connection preference - not relevant to std.json) it starts failing every time.



On Wednesday, 6 February 2013 at 09:41:28 UTC, Stephan wrote:
> On Monday, 4 February 2013 at 10:04:10 UTC, Andrea Fontana wrote:
>> This code was working some months ago:
>>
>> import std.json;
>> JSONValue total;
>> total.type = JSON_TYPE.OBJECT;
>> total["results"].type = JSON_TYPE.OBJECT;
>>
>>
>> Now, last line gives me a segmentation fault. What's wrong with this code?
>
> Here is a full test case. I don't want to submit it to Bugzilla yet, since I don't know whether it's intended behaviour or a bug. I have a slight feeling that this is NOT a bug. Somehow, in the first case a reference to an invalid struct is returned by json["hello"] and attempted to be overridden by a new JSONValue. In the second case, no reference is generated, instead a simple struct copy (a blit) is invoked, which works. Can someone explain this more correctly?
>
> This program crashes (tested on dmd 2.061)
> import std.json;
> void main() {
>   JSONValue json;
>   json.type = JSON_TYPE.OBJECT;
>   json["hello"] = JSONValue();
> }
>
> This program works fine:
> import std.json;
> void main() {
>   JSONValue json;
>   json.type = JSON_TYPE.OBJECT;
>   json.object["hello"] = JSONValue();
> }

On Wednesday, 6 February 2013 at 12:09:47 UTC, Andrea Fontana wrote:
> Yes it works in that way. I had to change it in a lot of source files.
> It worked fine for months as online webservice used by thousands users and never crashed :)
>
> With last changes to code (a db connection preference - not relevant to std.json) it starts failing every time.
>

If you are interested, you can apply phobos pull request which fixes json's opIndex.