Hi,

I just stubled acros this JSON test suite. There graph shows plenty of different JSON parsers and how they compare with this test suite. The Free Pascal JSON parser was not in that list.

Has anybody tried the FPC JSON parser with these tests?

https://github.com/nst/JSONTestSuite

This should catch a lot of corner cases.

Regards,
Graeme

I have not tested it, but patches pertaining to this will surely be accepted.

Michael.

Has anybody tried the FPC JSON parser with these tests?

I just did with the following program:

{$mode objfpc}
uses
  Classes,SysUtils,fpjson,jsonparser;
var
  Input: TStringList;
  JSON: TJSONData;
begin
  try
    try
      Input := TStringList.Create;
      Input.LoadFromFile(ParamStr(1));
      JSON := GetJSON('[""],');
      ExitCode := 0;
    finally
      JSON.Free;
      Input.Free;
    end;
  except
    on e: EParserError do begin
      ExitCode := 1;
    end;
    on e: Exception do begin
      ExitCode := 2;
    end;
  end;
end.

Result is not very convincing:
parsing.html
<http://free-pascal-general.1045716.n5.nabble.com/file/n5726704/parsing.html>

I test some of the bad cases, indeed fpjson parser is too permissive.

Whoops, it looks like a CSS is missing: style.css <http://free-pascal-general.1045716.n5.nabble.com/file/n5726705/style.css> Plus, I compare it with Go 1.7's encoding/json package.

leledumbo wrote

I just did with the following program:

Awesome, thanks.

Result is not very convincing:

In comparison to Go, no it isn't. I guess we have some work to do.

Regards,
Graeme

Ehm. This is not parsing anything parametrizable?

I had a look at some of the constructs.
Some of them are not meant to be supported.

I did some random test, for instance

n_array_1_true_without_comma.json : [1 true]

FPC reports an error, as expected.
In your HTML, it is in color for 'parsing should have failed but succeeded'

So I have some doubts as to your results...

Michael.

This is the program I used:

{$mode objfpc}
uses
   Classes,SysUtils,fpjson,jsonparser;
var
   Input: TStream;
   JSON: TJSONData;
begin
   JSON:=Nil;
   INPUT:=Nil;
   ExitCode:=0;
   try
     try
       Input := TFileStream.Create(ParamStr(1),fmOpenRead or fmShareDenyWrite);
       JSON := GetJSON(Input);
     finally
       JSON.Free;
       Input.Free;
     end;
   except
     on e: EParserError do begin
       ExitCode := 1;
     end;
     on e: Exception do begin
       ExitCode := 2;
     end;
   end;
end.

Well, things may not be so bad...

Michael.

In comparison to Go, no it isn't. I guess we have some work to do.

I mean, if fcl-json wants to be on top. In comparison with others, we're among the second top lines (good cases mostly correct, bad cases a bit behind). At least fcl-json is way above 3 of Perl's parsers and Haskell which crashes even on good cases. Here's the complete comparison: parsing.html
<http://free-pascal-general.1045716.n5.nabble.com/file/n5726718/parsing.html>

Ehm. This is not parsing anything parametrizable?

My bad, I copy-paste-d the modified program for ensuring that it's the parser that really accepts the broken JSON. I think you could easily understand that the hardcoded parameter should be Input.Text :p

I mean, if fcl-json wants to be on top. In comparison with others, we're

among the second top lines (good cases mostly correct, bad cases a bit behind).

To make us proud, for the good cases, there are only 2 parsers that fully accept the test suite: java's nanojson & fcl-json

No problem. What about the false positive I demonstrated ?

Michael.

No problem. What about the false positive I demonstrated ?

I guess we'll need to talk (open issue?) to the author for that. I have no idea if it's really should be supported or just a misunderstanding from the author.

Anyway, FPC fcl-json is now merged into the repository: https://github.com/nst/JSONTestSuite/commit/c9f84db0f130c9eedb87218e34a2a4a1298097cd

Nono: the test is correct.
fcl-json gives an error, but in your test result it shows up as accepting it ?

Michael.

Nono: the test is correct. fcl-json gives an error, but in your test result it shows up as accepting it ?

Eh? Will need to recheck. The requirement is clear:

- 0 for accept
- 1 for reject
- >1 for crash

Could it be one of the two except block is wrongly executed?

I understood that. I got 1 for the indicated test.

Which is correct, it must be rejected.

So I don't think except blocks are to blame.

Michael.

In our previous episode, leledumbo said:

Too late, it already is in the news :-)

http://www.theregister.co.uk/2016/11/01/json_parsers_tested/

Hi,

it is better to use TJSONParser than GetJSON.

GetJSON is defined as:

procedure DefJSONParserHandler(AStream: TStream; const AUseUTF8:
Boolean; out
   Data: TJSONData);
Var
   P : TJSONParser;
begin
   Data:=Nil;
   P:=TJSONParser.Create(AStream,[joUTF8]);
   try
     Data:=P.Parse;
   finally
     P.Free;
   end;
end;

with

procedure DefJSONParserHandler(AStream: TStream; const AUseUTF8:
Boolean; out
   Data: TJSONData);
Var
   P : TJSONParser;
begin
   Data:=Nil;
   P:=TJSONParser.Create(AStream,[*joStrict,*joUTF8]);
   try
     Data:=P.Parse;
   finally
     P.Free;
   end;
end;

it should pass more

Best,

Benito

Hi,
huh, why does the build suddenly fail?

Well, FPC 3.0.0 does not have the joStrict option. It is only there in 3.1.1.

Perhaps there should be two fpc parser entries.

Best,
Benito