+#define CHUNK_SIZE (4 * 1024 * 1024)
+
+/*
+ * File layout:
+ * - Fixed size 8MiB data part at offset 0.
+ * - Either one byte per sample for LA8.
+ * - Or two bytes per sample for LA16, in little endian format.
+ * - Five byte "header" at offset 8MiB.
+ * - One "clock divider" byte. The byte value is the divider factor
+ * minus 1. Value 0xff is invalid. Base clock is 100MHz for LA8, or
+ * 200MHz for LA16.
+ * - Four bytes for the trigger position. This 32bit value is the
+ * sample number in little endian format, or 0 when unused.
+ */
+#define CHRONOVU_LA8_DATASIZE (8 * 1024 * 1024)
+#define CHRONOVU_LA8_HDRSIZE (sizeof(uint8_t) + sizeof(uint32_t))
+#define CHRONOVU_LA8_FILESIZE (CHRONOVU_LA8_DATASIZE + CHRONOVU_LA8_HDRSIZE)
+
+/*
+ * Implementation note:
+ *
+ * The .format_match() routine only checks the file size, but none of
+ * the header fields. Only little would be gained (only clock divider
+ * 0xff could get tested), but complexity would increase dramatically.
+ * Also the .format_match() routine is unlikely to receive large enough
+ * a buffer to include the header. Neither is the filename available to
+ * the .format_match() routine.
+ *
+ * There is no way to programmatically tell whether the file was created
+ * by LA8 or LA16 software, i.e. with 8 or 16 logic channels. If the
+ * filename was available, one might guess based on the file extension,
+ * but still would require user specs if neither of the known extensions
+ * were used or the input is fed from a pipe.
+ *
+ * The current input module implementation assumes that users specify
+ * the (channel count and) sample rate. Input data gets processed and
+ * passed along to the session bus, before the file "header" is seen.
+ * A future implementation could move channel creation from init() to
+ * receive() or end() (actually: a common routine called from those two
+ * routines), and could defer sample processing and feeding the session
+ * until the header was seen, including deferred samplerate calculation
+ * after having seen the header. But again this improvement depends on
+ * the availability of either the filename or the device type. Also note
+ * that applications then had to keep sending data to the input module's
+ * receive() routine until sufficient amounts of input data were seen
+ * including the header (see bug #1017).
+ */