summaryrefslogtreecommitdiff
path: root/tests/smart-converter.c
diff options
context:
space:
mode:
authorPerberos <[email protected]>2011-11-07 16:46:58 -0300
committerPerberos <[email protected]>2011-11-07 16:46:58 -0300
commit528c1e5ff51e213936e800fc5a9a25da99c0bdf2 (patch)
tree77f8aa456b09367ba81f04d4562fc935f898a951 /tests/smart-converter.c
downloadpluma-528c1e5ff51e213936e800fc5a9a25da99c0bdf2.tar.bz2
pluma-528c1e5ff51e213936e800fc5a9a25da99c0bdf2.tar.xz
initial
Diffstat (limited to 'tests/smart-converter.c')
-rwxr-xr-xtests/smart-converter.c353
1 files changed, 353 insertions, 0 deletions
diff --git a/tests/smart-converter.c b/tests/smart-converter.c
new file mode 100755
index 00000000..b48240a4
--- /dev/null
+++ b/tests/smart-converter.c
@@ -0,0 +1,353 @@
+/*
+ * smart-converter.c
+ * This file is part of gedit
+ *
+ * Copyright (C) 2009 - Ignacio Casal Quinteiro
+ *
+ * gedit is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * gedit is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with gedit; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor,
+ * Boston, MA 02110-1301 USA
+ */
+
+
+#include "gedit-smart-charset-converter.h"
+#include "gedit-encodings.h"
+#include <gio/gio.h>
+#include <glib.h>
+#include <string.h>
+
+#define TEXT_TO_CONVERT "this is some text to make the tests"
+#define TEXT_TO_GUESS "hello \xe6\x96\x87 world"
+
+static void
+print_hex (gchar *ptr, gint len)
+{
+ gint i;
+
+ for (i = 0; i < len; ++i)
+ {
+ g_printf ("\\x%02x", (unsigned char)ptr[i]);
+ }
+
+ g_printf ("\n");
+}
+
+static gchar *
+get_encoded_text (const gchar *text,
+ gsize nread,
+ const GeditEncoding *to,
+ const GeditEncoding *from,
+ gsize *bytes_written_aux,
+ gboolean care_about_error)
+{
+ GCharsetConverter *converter;
+ gchar *out, *out_aux;
+ gsize bytes_read, bytes_read_aux;
+ gsize bytes_written;
+ GConverterResult res;
+ GError *err;
+
+ converter = g_charset_converter_new (gedit_encoding_get_charset (to),
+ gedit_encoding_get_charset (from),
+ NULL);
+
+ out = g_malloc (200);
+ out_aux = g_malloc (200);
+ err = NULL;
+ bytes_read_aux = 0;
+ *bytes_written_aux = 0;
+
+ if (nread == -1)
+ {
+ nread = strlen (text);
+ }
+
+ do
+ {
+ res = g_converter_convert (G_CONVERTER (converter),
+ text + bytes_read_aux,
+ nread,
+ out_aux,
+ 200,
+ G_CONVERTER_INPUT_AT_END,
+ &bytes_read,
+ &bytes_written,
+ &err);
+ memcpy (out + *bytes_written_aux, out_aux, bytes_written);
+ bytes_read_aux += bytes_read;
+ *bytes_written_aux += bytes_written;
+ nread -= bytes_read;
+ } while (res != G_CONVERTER_FINISHED && res != G_CONVERTER_ERROR);
+
+ if (care_about_error)
+ {
+ g_assert_no_error (err);
+ }
+ else if (err)
+ {
+ g_printf ("** You don't care, but there was an error: %s", err->message);
+ return NULL;
+ }
+
+ out[*bytes_written_aux] = '\0';
+
+ if (!g_utf8_validate (out, *bytes_written_aux, NULL) && !care_about_error)
+ {
+ if (!care_about_error)
+ {
+ return NULL;
+ }
+ else
+ {
+ g_assert_not_reached ();
+ }
+ }
+
+ return out;
+}
+
+static GSList *
+get_all_encodings ()
+{
+ GSList *encs = NULL;
+ gint i = 0;
+
+ while (TRUE)
+ {
+ const GeditEncoding *enc;
+
+ enc = gedit_encoding_get_from_index (i);
+
+ if (enc == NULL)
+ break;
+
+ encs = g_slist_prepend (encs, (gpointer)enc);
+ i++;
+ }
+
+ return encs;
+}
+
+static gchar *
+do_test (const gchar *test_in,
+ const gchar *enc,
+ GSList *encodings,
+ gsize nread,
+ const GeditEncoding **guessed)
+{
+ GeditSmartCharsetConverter *converter;
+ gchar *out, *out_aux;
+ gsize bytes_read, bytes_read_aux;
+ gsize bytes_written, bytes_written_aux;
+ GConverterResult res;
+ GError *err;
+
+ if (enc != NULL)
+ {
+ encodings = NULL;
+ encodings = g_slist_prepend (encodings, (gpointer)gedit_encoding_get_from_charset (enc));
+ }
+
+ converter = gedit_smart_charset_converter_new (encodings);
+
+ out = g_malloc (200);
+ out_aux = g_malloc (200);
+ err = NULL;
+ bytes_read_aux = 0;
+ bytes_written_aux = 0;
+
+ do
+ {
+ res = g_converter_convert (G_CONVERTER (converter),
+ test_in + bytes_read_aux,
+ nread,
+ out_aux,
+ 200,
+ G_CONVERTER_INPUT_AT_END,
+ &bytes_read,
+ &bytes_written,
+ &err);
+ memcpy (out + bytes_written_aux, out_aux, bytes_written);
+ bytes_read_aux += bytes_read;
+ bytes_written_aux += bytes_written;
+ nread -= bytes_read;
+ } while (res != G_CONVERTER_FINISHED && res != G_CONVERTER_ERROR);
+
+ g_assert_no_error (err);
+ out[bytes_written_aux] = '\0';
+
+ if (guessed != NULL)
+ *guessed = gedit_smart_charset_converter_get_guessed (converter);
+
+ return out;
+}
+
+static void
+do_test_roundtrip (const char *str, const char *charset)
+{
+ gsize len;
+ gchar *buf, *p;
+ GInputStream *in, *tmp;
+ GCharsetConverter *c1;
+ GeditSmartCharsetConverter *c2;
+ gsize n, tot;
+ GError *err;
+ GSList *enc = NULL;
+
+ len = strlen(str);
+ buf = g_new0 (char, len);
+
+ in = g_memory_input_stream_new_from_data (str, -1, NULL);
+
+ c1 = g_charset_converter_new (charset, "UTF-8", NULL);
+
+ tmp = in;
+ in = g_converter_input_stream_new (in, G_CONVERTER (c1));
+ g_object_unref (tmp);
+ g_object_unref (c1);
+
+ enc = g_slist_prepend (enc, (gpointer)gedit_encoding_get_from_charset (charset));
+ c2 = gedit_smart_charset_converter_new (enc);
+ g_slist_free (enc);
+
+ tmp = in;
+ in = g_converter_input_stream_new (in, G_CONVERTER (c2));
+ g_object_unref (tmp);
+ g_object_unref (c2);
+
+ tot = 0;
+ p = buf;
+ n = len;
+ while (TRUE)
+ {
+ gssize res;
+
+ err = NULL;
+ res = g_input_stream_read (in, p, n, NULL, &err);
+ g_assert_no_error (err);
+ if (res == 0)
+ break;
+
+ p += res;
+ n -= res;
+ tot += res;
+ }
+
+ g_assert_cmpint (tot, ==, len);
+ g_assert_cmpstr (str, ==, buf);
+
+ g_free (buf);
+ g_object_unref (in);
+}
+
+static void
+test_utf8_utf8 ()
+{
+ gchar *aux;
+
+ aux = do_test (TEXT_TO_CONVERT, "UTF-8", NULL, strlen (TEXT_TO_CONVERT), NULL);
+ g_assert_cmpstr (aux, ==, TEXT_TO_CONVERT);
+
+ aux = do_test ("foobar\xc3\xa8\xc3\xa8\xc3\xa8zzzzzz", "UTF-8", NULL, 18, NULL);
+ g_assert_cmpstr (aux, ==, "foobar\xc3\xa8\xc3\xa8\xc3\xa8zzzzzz");
+
+ aux = do_test ("foobar\xc3\xa8\xc3\xa8\xc3\xa8zzzzzz", "UTF-8", NULL, 9, NULL);
+ g_assert_cmpstr (aux, ==, "foobar\xc3\xa8\xc3");
+
+ /* FIXME: Use the utf8 stream for a fallback? */
+ //do_test_with_error ("\xef\xbf\xbezzzzzz", encs, G_IO_ERROR_FAILED);
+}
+
+static void
+test_xxx_xxx ()
+{
+ GSList *encs, *l;
+
+ encs = get_all_encodings ();
+
+ /* Here we just test all encodings it is just to know that the conversions
+ are done ok */
+ for (l = encs; l != NULL; l = g_slist_next (l))
+ {
+ do_test_roundtrip (TEXT_TO_CONVERT, gedit_encoding_get_charset ((const GeditEncoding *)l->data));
+ }
+
+ g_slist_free (encs);
+}
+
+static void
+test_empty ()
+{
+ const GeditEncoding *guessed;
+ gchar *out;
+ GSList *encodings = NULL;
+
+ /* testing the case of an empty file and list of encodings with no
+ utf-8. In this case, the smart converter cannot determine the right
+ encoding (because there is no input), but should still default to
+ utf-8 for the detection */
+ encodings = g_slist_prepend (encodings, (gpointer)gedit_encoding_get_from_charset ("UTF-16"));
+ encodings = g_slist_prepend (encodings, (gpointer)gedit_encoding_get_from_charset ("ISO-8859-15"));
+
+ out = do_test ("", NULL, encodings, 0, &guessed);
+
+ g_assert_cmpstr (out, ==, "");
+
+ g_assert (guessed == gedit_encoding_get_utf8 ());
+}
+
+static void
+test_guessed ()
+{
+ GSList *encs = NULL;
+ gchar *aux, *aux2, *fail;
+ gsize aux_len, fail_len;
+ const GeditEncoding *guessed;
+
+ aux = get_encoded_text (TEXT_TO_GUESS, -1,
+ gedit_encoding_get_from_charset ("UTF-16"),
+ gedit_encoding_get_from_charset ("UTF-8"),
+ &aux_len,
+ TRUE);
+
+ fail = get_encoded_text (aux, aux_len,
+ gedit_encoding_get_from_charset ("UTF-8"),
+ gedit_encoding_get_from_charset ("ISO-8859-15"),
+ &fail_len,
+ FALSE);
+
+ g_assert (fail == NULL);
+
+ /* ISO-8859-15 should fail */
+ encs = g_slist_append (encs, (gpointer)gedit_encoding_get_from_charset ("ISO-8859-15"));
+ encs = g_slist_append (encs, (gpointer)gedit_encoding_get_from_charset ("UTF-16"));
+
+ aux2 = do_test (aux, NULL, encs, aux_len, &guessed);
+
+ g_assert (guessed == gedit_encoding_get_from_charset ("UTF-16"));
+}
+
+int main (int argc,
+ char *argv[])
+{
+ g_type_init ();
+ g_test_init (&argc, &argv, NULL);
+
+ g_test_add_func ("/smart-converter/utf8-utf8", test_utf8_utf8);
+ //g_test_add_func ("/smart-converter/xxx-xxx", test_xxx_xxx);
+ g_test_add_func ("/smart-converter/guessed", test_guessed);
+ g_test_add_func ("/smart-converter/empty", test_empty);
+
+ return g_test_run ();
+}