Bug 37745

Can you please explain why
  x = (unsigned char)*p1 - (unsigned char)*p2;
is correct if neither p1 nor p2 are unsigned chars?

(In reply to comment #1)
> Can you please explain why
>   x = (unsigned char)*p1 - (unsigned char)*p2;
> is correct if neither p1 nor p2 are unsigned chars?

I will provide a code example:

#include <stdio.h>
#include <string.h>

int cmp(const char*s, const char *sA) {
    int n1, i, x;
    const char *p1, *p2;

    n1 = strlen(s);
    for (i = 0, p1 = s, p2 = sA; i < n1 && *p2; ++i, ++p1, ++p2) {
        x = *p1 - *p2;
        if (x != 0) {
            return x;
        }
    }
    if (i < n1) {
        return 1;
    }
    if (*p2) {
        return -1;
    }
    return 0;
}

int unsigned_cmp(const char*s, const char *sA) {
    int n1, i, x;
    const char *p1, *p2;

    n1 = strlen(s);
    for (i = 0, p1 = s, p2 = sA; i < n1 && *p2; ++i, ++p1, ++p2) {
        x = (unsigned char)*p1 - (unsigned char)*p2;
        if (x != 0) {
            return x;
        }
    }
    if (i < n1) {
        return 1;
    }
    if (*p2) {
        return -1;
    }
    return 0;
}

int main()
{
    const char* str1 = "17_scuole";
    const char* str2 = "þÿ";
    
    int ret = strcmp(str1, str2);
    printf("strcmp ret: %d\n", ret);

    ret = cmp(str1, str2);
    printf("cmp ret: %d\n", ret);

    ret = unsigned_cmp(str1, str2);
    printf("unsigned_cmp ret: %d\n", ret);
}





The output is:

strcmp ret: -1
cmp ret: 110
unsigned_cmp ret: -146

A correct returned value is either -1 or -146, a positive value is instead incorrect. This fools the bsearch (binary tree sarch), pointing the search for the entry "17_scuole" to a incorrect branch, resulting in a NULL even if the entry "17_scuole" is present. I hope this example explains clearly the problem. 
Regards.

Ok, i see the issue here, entries is "sorted" by the PDF file and the sorting in GooString::cmp does not agree with that sorting so the binary search fails. I understand you have such a file, could you please attach it to the bug?

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/poppler/poppler/issues/186.