5eeaed09ae
Signed-off-by: Yu-Chen Lin <npes87184@gmail.com>
183 lines
4.2 KiB
C
183 lines
4.2 KiB
C
#include <assert.h>
|
|
#include <string.h>
|
|
#include <SDL2/SDL.h>
|
|
|
|
#include "util/str_util.h"
|
|
|
|
static void test_xstrncpy_simple(void) {
|
|
char s[] = "xxxxxxxxxx";
|
|
size_t w = xstrncpy(s, "abcdef", sizeof(s));
|
|
|
|
// returns strlen of copied string
|
|
assert(w == 6);
|
|
|
|
// is nul-terminated
|
|
assert(s[6] == '\0');
|
|
|
|
// does not write useless bytes
|
|
assert(s[7] == 'x');
|
|
|
|
// copies the content as expected
|
|
assert(!strcmp("abcdef", s));
|
|
}
|
|
|
|
static void test_xstrncpy_just_fit(void) {
|
|
char s[] = "xxxxxx";
|
|
size_t w = xstrncpy(s, "abcdef", sizeof(s));
|
|
|
|
// returns strlen of copied string
|
|
assert(w == 6);
|
|
|
|
// is nul-terminated
|
|
assert(s[6] == '\0');
|
|
|
|
// copies the content as expected
|
|
assert(!strcmp("abcdef", s));
|
|
}
|
|
|
|
static void test_xstrncpy_truncated(void) {
|
|
char s[] = "xxx";
|
|
size_t w = xstrncpy(s, "abcdef", sizeof(s));
|
|
|
|
// returns 'n' (sizeof(s))
|
|
assert(w == 4);
|
|
|
|
// is nul-terminated
|
|
assert(s[3] == '\0');
|
|
|
|
// copies the content as expected
|
|
assert(!strncmp("abcdef", s, 3));
|
|
}
|
|
|
|
static void test_xstrjoin_simple(void) {
|
|
const char *const tokens[] = { "abc", "de", "fghi", NULL };
|
|
char s[] = "xxxxxxxxxxxxxx";
|
|
size_t w = xstrjoin(s, tokens, ' ', sizeof(s));
|
|
|
|
// returns strlen of concatenation
|
|
assert(w == 11);
|
|
|
|
// is nul-terminated
|
|
assert(s[11] == '\0');
|
|
|
|
// does not write useless bytes
|
|
assert(s[12] == 'x');
|
|
|
|
// copies the content as expected
|
|
assert(!strcmp("abc de fghi", s));
|
|
}
|
|
|
|
static void test_xstrjoin_just_fit(void) {
|
|
const char *const tokens[] = { "abc", "de", "fghi", NULL };
|
|
char s[] = "xxxxxxxxxxx";
|
|
size_t w = xstrjoin(s, tokens, ' ', sizeof(s));
|
|
|
|
// returns strlen of concatenation
|
|
assert(w == 11);
|
|
|
|
// is nul-terminated
|
|
assert(s[11] == '\0');
|
|
|
|
// copies the content as expected
|
|
assert(!strcmp("abc de fghi", s));
|
|
}
|
|
|
|
static void test_xstrjoin_truncated_in_token(void) {
|
|
const char *const tokens[] = { "abc", "de", "fghi", NULL };
|
|
char s[] = "xxxxx";
|
|
size_t w = xstrjoin(s, tokens, ' ', sizeof(s));
|
|
|
|
// returns 'n' (sizeof(s))
|
|
assert(w == 6);
|
|
|
|
// is nul-terminated
|
|
assert(s[5] == '\0');
|
|
|
|
// copies the content as expected
|
|
assert(!strcmp("abc d", s));
|
|
}
|
|
|
|
static void test_xstrjoin_truncated_before_sep(void) {
|
|
const char *const tokens[] = { "abc", "de", "fghi", NULL };
|
|
char s[] = "xxxxxx";
|
|
size_t w = xstrjoin(s, tokens, ' ', sizeof(s));
|
|
|
|
// returns 'n' (sizeof(s))
|
|
assert(w == 7);
|
|
|
|
// is nul-terminated
|
|
assert(s[6] == '\0');
|
|
|
|
// copies the content as expected
|
|
assert(!strcmp("abc de", s));
|
|
}
|
|
|
|
static void test_xstrjoin_truncated_after_sep(void) {
|
|
const char *const tokens[] = { "abc", "de", "fghi", NULL };
|
|
char s[] = "xxxxxxx";
|
|
size_t w = xstrjoin(s, tokens, ' ', sizeof(s));
|
|
|
|
// returns 'n' (sizeof(s))
|
|
assert(w == 8);
|
|
|
|
// is nul-terminated
|
|
assert(s[7] == '\0');
|
|
|
|
// copies the content as expected
|
|
assert(!strcmp("abc de ", s));
|
|
}
|
|
|
|
static void test_strquote(void) {
|
|
const char *s = "abcde";
|
|
char *out = strquote(s);
|
|
|
|
// add '"' at the beginning and the end
|
|
assert(!strcmp("\"abcde\"", out));
|
|
|
|
SDL_free(out);
|
|
}
|
|
|
|
static void test_utf8_truncate(void) {
|
|
const char *s = "aÉbÔc";
|
|
assert(strlen(s) == 7); // É and Ô are 2 bytes-wide
|
|
|
|
size_t count;
|
|
|
|
count = utf8_truncation_index(s, 1);
|
|
assert(count == 1);
|
|
|
|
count = utf8_truncation_index(s, 2);
|
|
assert(count == 1); // É is 2 bytes-wide
|
|
|
|
count = utf8_truncation_index(s, 3);
|
|
assert(count == 3);
|
|
|
|
count = utf8_truncation_index(s, 4);
|
|
assert(count == 4);
|
|
|
|
count = utf8_truncation_index(s, 5);
|
|
assert(count == 4); // Ô is 2 bytes-wide
|
|
|
|
count = utf8_truncation_index(s, 6);
|
|
assert(count == 6);
|
|
|
|
count = utf8_truncation_index(s, 7);
|
|
assert(count == 7);
|
|
|
|
count = utf8_truncation_index(s, 8);
|
|
assert(count == 7); // no more chars
|
|
}
|
|
|
|
int main(void) {
|
|
test_xstrncpy_simple();
|
|
test_xstrncpy_just_fit();
|
|
test_xstrncpy_truncated();
|
|
test_xstrjoin_simple();
|
|
test_xstrjoin_just_fit();
|
|
test_xstrjoin_truncated_in_token();
|
|
test_xstrjoin_truncated_before_sep();
|
|
test_xstrjoin_truncated_after_sep();
|
|
test_strquote();
|
|
test_utf8_truncate();
|
|
return 0;
|
|
}
|