Re: Problem with string replacement

[Richard Frith-Macdonald]

On 28 Oct 2011, at 12:21, Fred Kiefer wrote:

> Yesterday I noticed that I have a few additional failing tests for base. Most 
> of these come from the test for NSJSONSerialization. After a closer look I 
> was able to restrict the issue down to the following code, which hasn't 
> anything to do with JSON any more.
> In the JSON test code we had a string that contained the character " and the 
> unicode value 1 and the serialisation tries to deal with these value by 
> replacing them.
> 
> 
>  NSMutableString *str1, *str2;
>  unichar chars[3];
> 
>  chars[0] = '\"';
>  chars[1] = 1;
>  str1 = [NSMutableString stringWithCharacters: chars length: 2];
>  [str1 replaceOccurrencesOfString: @"\""
>                       withString: @"\\\""
>                          options: 0
>                            range: NSMakeRange(0, [str1 length])];
> 
>  chars[0] = '\\';
>  chars[1] = '\"';
>  chars[2] = 1;
>  str2 = [NSMutableString stringWithCharacters: chars length: 3];
>  NSLog(@"string 1 %@ string 2 %@", str1, str2);
>  PASS([str1 isEqual: str2],
>    "string occurrences replacement works");
> 
> To me this looks like a bug in the code of replaceOccurrencesOfString:... I 
> tried to understand that code but completely failed inside the GSeq.h 
> functions. At least I noticed a few things, first we end of using the 
> function strRangeCsNs, this is because the searched string is an 
> NSConstantString and all the optimisations don't kick in for that case. The 
> second is that this function returns twice the range {0, 1} although the 
> second time the search range is {2, 1}.
> 
> Could somebody please explain this to me? If this is an actual problem I 
> would like to add the above code into the NSMutableString tests.

Please add it to the tests ... I'll have a look at it.

I couldn't reproduce the problem on my main system, but on my 64bit test system 
where I'm experimenting with clang it shows up ... so it's probably a 64bit 
related issue.