aboutsummaryrefslogtreecommitdiff
path: root/src/dictzip-input-stream.c
diff options
context:
space:
mode:
authorPřemysl Janouch <p.janouch@gmail.com>2015-02-26 23:26:52 +0100
committerPřemysl Janouch <p.janouch@gmail.com>2015-02-26 23:29:45 +0100
commit10240c716aa7eea6aada39e77a7a3cd290a21663 (patch)
tree88cf815e4d002d0765ad03f97c64faa1acc317b2 /src/dictzip-input-stream.c
parentf9c308765f7aa16d553c72551fd830b7e29c9aec (diff)
downloadtdv-10240c716aa7eea6aada39e77a7a3cd290a21663.tar.gz
tdv-10240c716aa7eea6aada39e77a7a3cd290a21663.tar.xz
tdv-10240c716aa7eea6aada39e77a7a3cd290a21663.zip
Formatting
Diffstat (limited to 'src/dictzip-input-stream.c')
-rw-r--r--src/dictzip-input-stream.c40
1 files changed, 20 insertions, 20 deletions
diff --git a/src/dictzip-input-stream.c b/src/dictzip-input-stream.c
index 8b26096..f17a3c2 100644
--- a/src/dictzip-input-stream.c
+++ b/src/dictzip-input-stream.c
@@ -50,7 +50,7 @@ free_gzip_header (gz_header *gzh)
g_free (gzh->name); gzh->name = NULL;
}
-/* Reading the header in manually due to stupidity of the ZLIB API. */
+// Reading the header in manually due to stupidity of the ZLIB API.
static gboolean
read_gzip_header (GInputStream *is, gz_header *gzh,
goffset *first_block_offset, GError **error)
@@ -279,19 +279,19 @@ static gssize dictzip_input_stream_skip (GInputStream *stream, gsize count,
struct dictzip_input_stream_private
{
- GFileInfo * file_info; //!< File information from gzip header
+ GFileInfo * file_info; ///< File information from gzip header
- goffset first_block_offset; //!< Offset to the first block/chunk
- gsize chunk_length; //!< Uncompressed chunk length
- gsize n_chunks; //!< Number of chunks in file
- guint16 * chunks; //!< Chunk sizes after compression
+ goffset first_block_offset; ///< Offset to the first block/chunk
+ gsize chunk_length; ///< Uncompressed chunk length
+ gsize n_chunks; ///< Number of chunks in file
+ guint16 * chunks; ///< Chunk sizes after compression
- z_stream zs; //!< zlib decompression context
- gpointer input_buffer; //!< Input buffer
+ z_stream zs; ///< zlib decompression context
+ gpointer input_buffer; ///< Input buffer
- goffset offset; //!< Current offset
- gpointer * decompressed; //!< Array of decompressed chunks
- gsize last_chunk_length; //!< Size of the last chunk
+ goffset offset; ///< Current offset
+ gpointer * decompressed; ///< Array of decompressed chunks
+ gsize last_chunk_length; ///< Size of the last chunk
};
G_DEFINE_TYPE_EXTENDED (DictzipInputStream, dictzip_input_stream,
@@ -422,7 +422,7 @@ get_chunk (DictzipInputStream *self, guint chunk_id, GError **error)
gpointer chunk = priv->decompressed[chunk_id];
if (!chunk)
{
- /* Just inflating the file piece by piece as needed. */
+ // Just inflating the file piece by piece as needed.
gsize chunk_size;
chunk = inflate_chunk (self, chunk_id, &chunk_size, error);
if (!chunk)
@@ -452,8 +452,8 @@ dictzip_input_stream_seek (GSeekable *seekable, goffset offset,
if (type == G_SEEK_END)
{
- /* This could be implemented by retrieving the last chunk
- * and deducing the filesize, should the functionality be needed. */
+ // This could be implemented by retrieving the last chunk
+ // and deducing the filesize, should the functionality be needed.
g_set_error (error, G_IO_ERROR, G_IO_ERROR_NOT_SUPPORTED,
"I don't know where the stream ends, cannot seek there");
return FALSE;
@@ -541,7 +541,7 @@ dictzip_input_stream_skip (GInputStream *stream, gsize count,
return count;
}
-/** Create an input stream for the underlying dictzip file. */
+/// Create an input stream for the underlying dictzip file.
DictzipInputStream *
dictzip_input_stream_new (GInputStream *base_stream, GError **error)
{
@@ -560,7 +560,7 @@ dictzip_input_stream_new (GInputStream *base_stream, GError **error)
"base-stream", base_stream, "close-base-stream", FALSE, NULL);
DictzipInputStreamPrivate *priv = self->priv;
- /* Decode the header. */
+ // Decode the header.
gz_header gzh;
if (!read_gzip_header (G_INPUT_STREAM (base_stream),
&gzh, &priv->first_block_offset, &err))
@@ -584,7 +584,7 @@ dictzip_input_stream_new (GInputStream *base_stream, GError **error)
goto error;
}
- /* Store file information. */
+ // Store file information.
priv->file_info = g_file_info_new ();
if (gzh.time != 0)
@@ -596,7 +596,7 @@ dictzip_input_stream_new (GInputStream *base_stream, GError **error)
if (gzh.name && *gzh.name)
g_file_info_set_name (priv->file_info, (gchar *) gzh.name);
- /* Initialise zlib. */
+ // Initialise zlib.
int z_err;
z_err = inflateInit2 (&priv->zs, -15);
if (z_err != Z_OK)
@@ -608,7 +608,7 @@ dictzip_input_stream_new (GInputStream *base_stream, GError **error)
priv->input_buffer = g_malloc (65536);
priv->decompressed = g_new0 (gpointer, priv->n_chunks);
- priv->last_chunk_length = -1; // We don't know yet.
+ priv->last_chunk_length = -1; // We don't know yet.
free_gzip_header (&gzh);
return self;
@@ -619,7 +619,7 @@ error:
return NULL;
}
-/** Return file information for the compressed file. */
+/// Return file information for the compressed file.
GFileInfo *
dictzip_input_stream_get_file_info (DictzipInputStream *self)
{