xpcom strings

[nskpm]

Hi all,

I work on xpcom component for Pale Moon on macOS, and I'm stuck on xpcom strings.

Here is code which demonstrates error I have in component:

Code: Select all

#include "mozilla-config.h"
#include "nsStringAPI.h"
#include <iostream>

using namespace std;

int main() {
        NS_NAMED_LITERAL_STRING(str, "test");
        cout << str.Length() << endl;
        cout << str.CharAt(0) << endl;
}

Line with Length() outputs 0, and segmentation fault happens on line with CharAt(). Instead of literal I tried nsString, nsEmbedString, nsAutoString, maybe, something else. The result is always the same.

I've seen similar code in Pale Moon sources, so, I can guess that code itself is correct, but something special should be done during compilation. But what exactly? Does anybody know?

[athenian200]

I think you need a check to make sure Length isn't 0 before passing off to CharAt? Actually, I think you want to make sure that CharAt is always a lower value than the length of the string in general. If the string's length is 0, then even being at position 0 is too high, because you'd need a string length of at least 1 (the size of a char) to have a character to display. In other words, you can't get any valid characters out of a zero-length string. There is no valid character position within it, because it has no characters. If you try to move to any position in a zero-length string, you're doing something invalid.

I don't know a lot about xpcom strings specifically, but that's how C/C++ works in my experience.

[vannilla]

I'm not an expert of XPCOM either, but "NS_NAMED_LITERAL_STRING" currently expands to

Code: Select all

const nsDependentString n(reinterpret_cast<const nsAString::char_type*>(s), \
                          uint32_t((sizeof(s) / 2) - 1))

in which the second argument is supposed to be the length of the passed data, in your case "test".
Because this is C++ what is going on is definitely out of my knoweledge, but the fact that the string length is calculated with sizeof is still strange.

[nskpm]

I think you need a check to make sure Length isn't 0 before passing off to CharAt?

Of course, it's clear that accessing char in string of length 0 will crash a program.

The problem is not that program crashes (this is just a consequence), the problem is that string is not initialized properly.

And the question is how to properly initialize it.