summaryrefslogtreecommitdiff
path: root/libnm-core/tests/test-general.c
diff options
context:
space:
mode:
Diffstat (limited to 'libnm-core/tests/test-general.c')
-rw-r--r--libnm-core/tests/test-general.c174
1 files changed, 87 insertions, 87 deletions
diff --git a/libnm-core/tests/test-general.c b/libnm-core/tests/test-general.c
index 20b3b222eb..7c10f02756 100644
--- a/libnm-core/tests/test-general.c
+++ b/libnm-core/tests/test-general.c
@@ -142,9 +142,9 @@ _test_hash_str(const char *str)
v = nm_hash_complete(&h);
/* assert that hashing a string and a buffer yields the
- * same result.
- *
- * I think that is a desirable property. */
+ * same result.
+ *
+ * I think that is a desirable property. */
nm_hash_init(&h, SEED);
nm_hash_update_mem(&h, str, strlen(str));
v2 = nm_hash_complete(&h);
@@ -167,8 +167,8 @@ _test_hash_str(const char *str)
nm_hash_init(&h3, 10); \
\
/* assert that it doesn't matter, whether we hash the values individually,
- * or all at once, or via the convenience macros nm_hash_update_val()
- * and nm_hash_update_vals(). */ \
+ * or all at once, or via the convenience macros nm_hash_update_val()
+ * and nm_hash_update_vals(). */ \
for (i = 0; i < G_N_ELEMENTS(v); i++) { \
nm_hash_update(&h0, &v[i], sizeof(type)); \
nm_hash_update_val(&h1, v[i]); \
@@ -300,7 +300,7 @@ _do_test_nm_utils_strsplit_set_f_one(NMUtilsStrsplitSetFlags flags,
initial_offset = (f_preserve_empty || !str) ? 0u : strspn(str, DELIMITERS);
/* first compare our expected values with what g_strsplit_set() would
- * do. */
+ * do. */
words_g = str ? g_strsplit_set(str, DELIMITERS, -1) : NULL;
if (str == NULL) {
g_assert_cmpint(words_len, ==, 0);
@@ -365,9 +365,9 @@ _do_test_nm_utils_strsplit_set_f_one(NMUtilsStrsplitSetFlags flags,
}
/* while strsplit removes all delimiters, we can relatively easily find them
- * in the original string. Assert that the original string and the pointer offsets
- * of words correspond. In particular, find idx_delim_after and idx_delim_before
- * to determine which delimiter was after/before a word. */
+ * in the original string. Assert that the original string and the pointer offsets
+ * of words correspond. In particular, find idx_delim_after and idx_delim_before
+ * to determine which delimiter was after/before a word. */
{
gsize idx_word_start;
gsize idx_delim_after_old = G_MAXSIZE;
@@ -379,9 +379,9 @@ _do_test_nm_utils_strsplit_set_f_one(NMUtilsStrsplitSetFlags flags,
gsize idx_delim_before;
/* find the delimiter *after* words[i]. We can do that by looking at the next
- * word and calculating the pointer difference.
- *
- * The delimiter after the very last word is '\0' and requires strlen() to find. */
+ * word and calculating the pointer difference.
+ *
+ * The delimiter after the very last word is '\0' and requires strlen() to find. */
idx_delim_after = initial_offset + ((words[i] - words[0]) + l_i);
if (idx_delim_after != idx_word_start + l_i) {
g_assert(!f_preserve_empty);
@@ -408,7 +408,7 @@ _do_test_nm_utils_strsplit_set_f_one(NMUtilsStrsplitSetFlags flags,
/* find the delimiter *before* words[i]. */
if (i == 0) {
/* there is only a delimiter *before*, with !f_preserve_empty and leading
- * delimiters. */
+ * delimiters. */
idx_delim_before = G_MAXSIZE;
if (initial_offset > 0) {
g_assert(!f_preserve_empty);
@@ -755,8 +755,8 @@ _escaped_tokens_create_random_word_full(const char *const *tokens, gsize n_token
}
/* reallocate the string, so that we don't have any excess memory from
- * the GString buffer. This is so that valgrind may better detect an out
- * or range access. */
+ * the GString buffer. This is so that valgrind may better detect an out
+ * or range access. */
return nm_str_realloc(g_string_free(gstr, FALSE));
}
@@ -967,10 +967,10 @@ test_nm_utils_escaped_tokens(void)
if (i_option % 2u == 1 && nmtst_get_rand_uint32() % 5 == 0
&& strlen(options->pdata[options->len - 1]) > 0u) {
/* For some options, leave the value unset and only generate a key.
- *
- * If key is "", then we cannot do that, because the test below would try
- * to append "" to the combined list, which the parser then would drop.
- * Only test omitting the value, if strlen() of the key is positive. */
+ *
+ * If key is "", then we cannot do that, because the test below would try
+ * to append "" to the combined list, which the parser then would drop.
+ * Only test omitting the value, if strlen() of the key is positive. */
} else
word = _escaped_tokens_create_random_word();
g_ptr_array_add(options, word);
@@ -1032,18 +1032,18 @@ test_nm_utils_escaped_tokens(void)
g_assert_cmpint(NM_PTRARRAY_LEN(strv_split), ==, num_options);
/* Above we show a full round-trip of random option key-value pairs, that they can
- * without loss escape, concatenate, split-list, and split. This proofed that every
- * option key-value pair can be represented as a combined string and parsed back.
- *
- * Now, just check that we can also parse arbitrary random words in nm_utils_escaped_tokens_options_split().
- * split() is a non-injective surjective function. As we check the round-trip above for random words, where
- * options-split() is the last step, we show that every random word can be the output of the function
- * (which shows, the surjective part).
- *
- * But multiple random input arguments, may map to the same output argument (non-injective).
- * Just test whether we can handle random input words without crashing. For that, just use the
- * above generate list of random words.
- */
+ * without loss escape, concatenate, split-list, and split. This proofed that every
+ * option key-value pair can be represented as a combined string and parsed back.
+ *
+ * Now, just check that we can also parse arbitrary random words in nm_utils_escaped_tokens_options_split().
+ * split() is a non-injective surjective function. As we check the round-trip above for random words, where
+ * options-split() is the last step, we show that every random word can be the output of the function
+ * (which shows, the surjective part).
+ *
+ * But multiple random input arguments, may map to the same output argument (non-injective).
+ * Just test whether we can handle random input words without crashing. For that, just use the
+ * above generate list of random words.
+ */
for (i = 0; i < 1u + 2u * i_option; i++) {
gs_free char *str = NULL;
const char * cstr;
@@ -1726,9 +1726,9 @@ del_iter_func(const char *key, const char *value, gpointer user_data)
int i;
/* Record how many times this function gets called; it should get called
- * exactly as many times as there are keys in the hash table, regardless
- * of what keys we delete from the table.
- */
+ * exactly as many times as there are keys in the hash table, regardless
+ * of what keys we delete from the table.
+ */
info->called++;
/* During the iteration, remove a bunch of stuff from the table */
@@ -1774,8 +1774,8 @@ test_setting_vpn_modify_during_foreach(void)
}
/* And make sure the foreach callback was called the same number of times
- * as there were keys in the table at the beginning of the foreach.
- */
+ * as there were keys in the table at the beginning of the foreach.
+ */
g_assert_cmpint(info.called, ==, TO_DEL_NUM * 2);
g_object_unref(s_vpn);
@@ -1812,8 +1812,8 @@ test_setting_ip4_config_labels(void)
g_assert(label == NULL);
/* The 'address-labels' property should be omitted from the serialization if
- * there are no non-NULL labels.
- */
+ * there are no non-NULL labels.
+ */
conn = nmtst_create_minimal_connection("label test", NULL, NM_SETTING_WIRED_SETTING_NAME, NULL);
nm_connection_add_setting(conn, nm_setting_duplicate(NM_SETTING(s_ip4)));
dict = nm_connection_to_dbus(conn, NM_CONNECTION_SERIALIZE_ALL);
@@ -1876,8 +1876,8 @@ test_setting_ip4_config_labels(void)
g_assert(label == NULL);
/* If we serialize as the daemon, the labels should appear in the D-Bus
- * serialization under both 'address-labels' and 'address-data'.
- */
+ * serialization under both 'address-labels' and 'address-data'.
+ */
conn = nmtst_create_minimal_connection("label test", NULL, NM_SETTING_WIRED_SETTING_NAME, NULL);
nm_connection_add_setting(conn, NM_SETTING(s_ip4));
_nm_utils_is_manager_process = TRUE;
@@ -1915,8 +1915,8 @@ test_setting_ip4_config_labels(void)
g_variant_unref(setting_dict);
/* We should be able to deserialize the labels from either 'address-labels'
- * or 'address-data'.
- */
+ * or 'address-data'.
+ */
dict2 = g_variant_ref(dict);
NMTST_VARIANT_EDITOR(
@@ -2018,8 +2018,8 @@ test_setting_ip4_config_address_data(void)
nmtst_assert_setting_verifies(NM_SETTING(s_ip4));
/* The client-side D-Bus serialization should include the attributes in
- * "address-data", and should not have an "addresses" property.
- */
+ * "address-data", and should not have an "addresses" property.
+ */
conn = nmtst_create_minimal_connection("address-data test",
NULL,
NM_SETTING_WIRED_SETTING_NAME,
@@ -2512,8 +2512,8 @@ test_connection_to_dbus_setting_name(void)
dict = nm_connection_to_dbus(connection, NM_CONNECTION_SERIALIZE_ALL);
/* Make sure the keys of the first level dict are setting names, not
- * the GType name of the setting objects.
- */
+ * the GType name of the setting objects.
+ */
g_assert(_variant_contains(dict, NM_SETTING_WIRELESS_SECURITY_SETTING_NAME));
g_variant_unref(dict);
@@ -2727,15 +2727,15 @@ test_setting_new_from_dbus_bad(void)
GError * error = NULL;
/* We want to test:
- * - ordinary scalar properties
- * - string properties
- * - GBytes-valued properties (which are handled specially by set_property_from_dbus())
- * - enum/flags-valued properties
- * - overridden properties
- * - transformed properties
- *
- * No single setting class has examples of all of these, so we need two settings.
- */
+ * - ordinary scalar properties
+ * - string properties
+ * - GBytes-valued properties (which are handled specially by set_property_from_dbus())
+ * - enum/flags-valued properties
+ * - overridden properties
+ * - transformed properties
+ *
+ * No single setting class has examples of all of these, so we need two settings.
+ */
conn = nm_simple_connection_new();
@@ -3927,8 +3927,8 @@ test_connection_bad_base_types(void)
GError * error = NULL;
/* Test various non-base connection types to make sure they are rejected;
- * using a fake 'wired' connection so the rest of it verifies
- */
+ * using a fake 'wired' connection so the rest of it verifies
+ */
/* Connection setting */
connection = nm_simple_connection_new();
@@ -4353,8 +4353,8 @@ test_setting_compare_secrets(gconstpointer test_data)
gs_unref_object NMSetting *new = NULL;
/* Make sure that a connection with transient/unsaved secrets compares
- * successfully to the same connection without those secrets.
- */
+ * successfully to the same connection without those secrets.
+ */
old = nm_setting_wireless_security_new();
g_object_set(old,
@@ -4433,8 +4433,8 @@ test_setting_compare_vpn_secrets(gconstpointer test_data)
gboolean success;
/* Make sure that a connection with transient/unsaved secrets compares
- * successfully to the same connection without those secrets.
- */
+ * successfully to the same connection without those secrets.
+ */
old = nm_setting_vpn_new();
nm_setting_vpn_add_secret(NM_SETTING_VPN(old), "foobarbaz", "really secret password");
@@ -4650,7 +4650,7 @@ _netmask_to_prefix(guint32 netmask)
g_assert_cmpint(prefix, <=, 32);
/* we re-implemented the netmask-to-prefix code differently. Check
- * that they agree. */
+ * that they agree. */
g_assert_cmpint(prefix, ==, nm_utils_ip4_netmask_to_prefix(netmask));
return prefix;
@@ -4706,7 +4706,7 @@ test_ip4_netmask_to_prefix(void)
continue;
/* create an invalid netmask with holes and check that the function
- * returns the longest prefix. */
+ * returns the longest prefix. */
prefix_holey = _netmask_to_prefix(netmask_holey);
g_assert_cmpint(i, ==, prefix_holey);
@@ -5193,7 +5193,7 @@ test_setting_old_uuid(void)
gs_unref_object NMSetting *setting = NULL;
/* NetworkManager-0.9.4.0 generated 40-character UUIDs with no dashes,
- * like this one. Test that we maintain compatibility. */
+ * like this one. Test that we maintain compatibility. */
const char *uuid = "f43bec2cdd60e5da381ebb1eb1fa39f3cc52660c";
setting = nm_setting_connection_new();
@@ -5312,8 +5312,8 @@ test_connection_normalize_virtual_iface_name(void)
g_object_unref(con);
/* But removing connection.interface-name should result in vlan.connection-name
- * being "promoted".
- */
+ * being "promoted".
+ */
NMTST_VARIANT_EDITOR(connection_dict,
NMTST_VARIANT_DROP_PROPERTY(NM_SETTING_CONNECTION_SETTING_NAME,
NM_SETTING_CONNECTION_INTERFACE_NAME););
@@ -5838,8 +5838,8 @@ test_connection_normalize_gateway_never_default(void)
g_assert_cmpstr("1.1.1.254", ==, nm_setting_ip_config_get_gateway(s_ip4));
/* Now set never-default to TRUE and check that the gateway is
- * removed during normalization
- * */
+ * removed during normalization
+ * */
g_object_set(s_ip4, NM_SETTING_IP_CONFIG_NEVER_DEFAULT, TRUE, NULL);
nmtst_assert_connection_verifies_after_normalization(con,
@@ -5880,8 +5880,8 @@ test_connection_normalize_may_fail(void)
nmtst_assert_connection_verifies_and_normalizable(con);
/* Now set method=disabled/ignore and check that may-fail becomes TRUE
- * after normalization
- * */
+ * after normalization
+ * */
g_object_set(s_ip4, NM_SETTING_IP_CONFIG_METHOD, NM_SETTING_IP4_CONFIG_METHOD_DISABLED, NULL);
g_object_set(s_ip6, NM_SETTING_IP_CONFIG_METHOD, NM_SETTING_IP6_CONFIG_METHOD_IGNORE, NULL);
@@ -5922,8 +5922,8 @@ test_connection_normalize_shared_addresses(void)
nmtst_assert_connection_verifies_and_normalizable(con);
/* Now we add other addresses and check that they are
- * removed during normalization
- * */
+ * removed during normalization
+ * */
addr = nm_ip_address_new(AF_INET, "2.2.2.2", 24, &error);
g_assert_no_error(error);
nm_setting_ip_config_add_address(s_ip4, addr);
@@ -6345,8 +6345,8 @@ test_setting_ip4_gateway(void)
nmtst_assert_ip4_address(addr_vals_0[0], "192.168.1.10");
/* When serializing on the daemon side, ipv4.gateway is copied to the first
- * entry of ipv4.addresses
- */
+ * entry of ipv4.addresses
+ */
conn = nmtst_create_minimal_connection("test_setting_ip4_gateway",
NULL,
NM_SETTING_WIRED_SETTING_NAME,
@@ -6398,8 +6398,8 @@ test_setting_ip4_gateway(void)
g_variant_unref(ip4_dict);
/* When deserializing an old-style connection, the first non-0 gateway in
- * ipv4.addresses is copied to :gateway.
- */
+ * ipv4.addresses is copied to :gateway.
+ */
NMTST_VARIANT_EDITOR(
conn_dict,
NMTST_VARIANT_DROP_PROPERTY(NM_SETTING_IP4_CONFIG_SETTING_NAME,
@@ -6500,8 +6500,8 @@ test_setting_ip6_gateway(void)
GError * error = NULL;
/* When serializing on the daemon side, ipv6.gateway is copied to the first
- * entry of ipv6.addresses
- */
+ * entry of ipv6.addresses
+ */
conn = nmtst_create_minimal_connection("test_setting_ip6_gateway",
NULL,
NM_SETTING_WIRED_SETTING_NAME,
@@ -6555,8 +6555,8 @@ test_setting_ip6_gateway(void)
g_variant_unref(ip6_dict);
/* When deserializing an old-style connection, the first non-0 gateway in
- * ipv6.addresses is copied to :gateway.
- */
+ * ipv6.addresses is copied to :gateway.
+ */
NMTST_VARIANT_EDITOR(
conn_dict,
NMTST_VARIANT_DROP_PROPERTY(NM_SETTING_IP6_CONFIG_SETTING_NAME,
@@ -8280,7 +8280,7 @@ test_nm_utils_is_power_of_two(void)
if (xyes != 0) {
again:
/* Find another @xno, that is not a power of two. Do that,
- * by randomly setting bits. */
+ * by randomly setting bits. */
numbits = g_rand_int_range(rand, 1, 65);
while (xno != ~((guint64) 0) && numbits > 0) {
guint64 v = (((guint64) 1) << g_rand_int_range(rand, 0, 64));
@@ -8512,7 +8512,7 @@ test_nm_utils_ptrarray_find_binary_search_with_duplicates(void)
for (i_test = 0; i_test < N_TEST; i_test++) {
for (i_len = 0; i_len < BIN_SEARCH_W_DUPS_LEN; i_len++) {
/* fill with random numbers... surely there are some duplicates
- * there... or maybe even there are none... */
+ * there... or maybe even there are none... */
for (i = 0; i < i_len; i++)
arr[i] =
GINT_TO_POINTER(nmtst_get_rand_uint32() % (i_len + BIN_SEARCH_W_DUPS_JITTER));
@@ -8884,8 +8884,8 @@ _do_test_utils_str_utf8safe(const char * str,
g_assert_cmpstr(buf_safe, ==, str_safe);
/* nm_utils_buf_utf8safe_escape() can only return a pointer equal to the input string,
- * if and only if str_len is negative. Otherwise, the input str won't be NUL terminated
- * and cannot be returned. */
+ * if and only if str_len is negative. Otherwise, the input str won't be NUL terminated
+ * and cannot be returned. */
g_assert(buf_safe != str);
g_assert(buf_safe == str_free_1);
} else
@@ -9425,7 +9425,7 @@ test_nm_set_out(void)
int call_count;
/* NM_SET_OUT() has an unexpected non-function like behavior
- * wrt. side-effects of the value argument. Test it */
+ * wrt. side-effects of the value argument. Test it */
p_val = &val;
call_count = 0;
@@ -10121,14 +10121,14 @@ test_strsplit_quoted(void)
gs_free char * str = NULL;
/* create random strv array and join them carefully so that splitting
- * them will yield the original value. */
+ * them will yield the original value. */
strv = _strsplit_quoted_create_strv_rand();
str = _strsplit_quoted_join_strv_rand((const char *const *) strv);
_strsplit_quoted_test(str, (const char *const *) strv);
}
/* Create random words and assert that systemd and our implementation can
- * both split them (and in the exact same way). */
+ * both split them (and in the exact same way). */
for (i_run = 0; i_run < 1000; i_run++) {
gs_free char *s = _strsplit_quoted_create_str_rand(nmtst_get_rand_uint32() % 150);