summaryrefslogtreecommitdiff
path: root/gtk/gtkimcontextsimple.c
diff options
context:
space:
mode:
Diffstat (limited to 'gtk/gtkimcontextsimple.c')
-rw-r--r--gtk/gtkimcontextsimple.c26
1 files changed, 13 insertions, 13 deletions
diff --git a/gtk/gtkimcontextsimple.c b/gtk/gtkimcontextsimple.c
index ab721fe6f9..bd581f8c53 100644
--- a/gtk/gtkimcontextsimple.c
+++ b/gtk/gtkimcontextsimple.c
@@ -68,14 +68,14 @@ struct _GtkComposeTableCompact
gint n_index_stride;
};
-/* This file contains the table of the compose sequences,
+/* This file contains the table of the compose sequences,
* static const guint16 gtk_compose_seqs_compact[] = {}
- * IT is generated from the compose-parse.py script.
+ * It is generated from the compose-parse.py script.
*/
#include "gtkimcontextsimpleseqs.h"
-/* From the values below, the value 23 means the number of different first keysyms
- * that exist in the Compose file (from Xorg). When running compose-parse.py without
+/* From the values below, the value 24 means the number of different first keysyms
+ * that exist in the Compose file (from Xorg). When running compose-parse.py without
* parameters, you get the count that you can put here. Needed when updating the
* gtkimcontextsimpleseqs.h header file (contains the compose sequences).
*/
@@ -253,7 +253,7 @@ check_table (GtkIMContextSimple *context_simple,
guint16 *prev_seq;
/* Back up to the first sequence that matches to make sure
- * we find the exact match if their is one.
+ * we find the exact match if there is one.
*/
while (seq > table->data)
{
@@ -480,7 +480,8 @@ check_normalize_nfc (gunichar* combination_buffer, gint n_compose)
/* Xorg reuses dead_tilde for the perispomeni diacritic mark.
* We check if base character belongs to Greek Unicode block,
- * and if so, we replace tilde with perispomeni. */
+ * and if so, we replace tilde with perispomeni.
+ */
if (combination_buffer[0] >= 0x390 && combination_buffer[0] <= 0x3FF)
{
for (i = 1; i < n_compose; i++ )
@@ -494,7 +495,7 @@ check_normalize_nfc (gunichar* combination_buffer, gint n_compose)
{
g_unicode_canonical_ordering (combination_buffer_temp, n_compose);
combination_utf8_temp = g_ucs4_to_utf8 (combination_buffer_temp, -1, NULL, NULL, NULL);
- nfc_temp = g_utf8_normalize (combination_utf8_temp, -1, G_NORMALIZE_NFC);
+ nfc_temp = g_utf8_normalize (combination_utf8_temp, -1, G_NORMALIZE_NFC);
if (g_utf8_strlen (nfc_temp, -1) == 1)
{
@@ -589,9 +590,8 @@ check_algorithmically (GtkIMContextSimple *context_simple,
i--;
}
- /* If the buffer normalizes to a single character,
- * then modify the order of combination_buffer accordingly, if necessary,
- * and return TRUE.
+ /* If the buffer normalizes to a single character, then modify the order
+ * of combination_buffer accordingly, if necessary, and return TRUE.
*/
if (check_normalize_nfc (combination_buffer, n_compose))
{
@@ -616,8 +616,8 @@ check_algorithmically (GtkIMContextSimple *context_simple,
/* In addition to the table-driven sequences, we allow Unicode hex
* codes to be entered. The method chosen here is similar to the
* one recommended in ISO 14755, but not exactly the same, since we
- * don't want to steal 16 valuable key combinations.
- *
+ * don't want to steal 16 valuable key combinations.
+ *
* A hex Unicode sequence must be started with Ctrl-Shift-U, followed
* by a sequence of hex digits entered with Ctrl-Shift still held.
* Releasing one of the modifiers or pressing space while the modifiers
@@ -668,7 +668,7 @@ check_hex (GtkIMContextSimple *context_simple,
n = strtoul (str->str, &nptr, 16);
- /* if strtoul fails it probably means non-latin digits were used;
+ /* If strtoul fails it probably means non-latin digits were used;
* we should in principle handle that, but we probably don't.
*/
if (nptr - str->str < str->len)