Add incremental read interface suitable for reading from infinite streams.

git-svn-id: http://svn.drobilla.net/serd/trunk@350 490d8e77-9747-427b-9fa3-0b8f29cee8a0

6 files changed, 135 insertions, 17 deletions

diff --git a/NEWS b/NEWS
index a97bfabb..7b147287 100644
--- a/NEWS
+++ b/NEWS
@@ -1,3 +1,10 @@
+serd (9999) unstable;
+
+  * Add incremental read interface suitable for reading from infinite streams.
+  * Add -e option to serdi to use incremental reading.
+
+ -- David Robillard <d@drobilla.net>
+
 serd (0.14.0) stable;
 
   * Use path variables in pkgconfig files
diff --git a/doc/serdi.1 b/doc/serdi.1
index 23d77ac3..45001c90 100644
--- a/doc/serdi.1
+++ b/doc/serdi.1
@@ -1,4 +1,4 @@
-.TH SERDI 1 "17 Jan 2012"
+.TH SERDI 1 "08 May 2012"
 
 .SH NAME
 .B serdi \- Read and write RDF syntax
@@ -17,6 +17,14 @@ Fast bulk output for large serialisations.
 Chop PREFIX from matching blank node IDs.
 
 .TP
+\fB\-e\fR
+Eat input one character at a time, rather than a page at a time which is the
+default.  This is useful when reading from a pipe since output will be
+generated immediately as input arrives, rather than waiting until an entire
+page of input has arrived.  With this option serdi uses one page less memory,
+but will likely be significantly slower.
+
+.TP
 \fB\-f\fR
 Keep full URIs in input (don't qualify).
 
diff --git a/serd/serd.h b/serd/serd.h
index 807664dd..cc2365f4 100644
--- a/serd/serd.h
+++ b/serd/serd.h
@@ -705,6 +705,40 @@ serd_reader_read_file(SerdReader*    reader,
                       const uint8_t* uri);
 
 /**
+   Start an incremental read from a file handle.
+
+   Iff @p bulk is true, @p file will be read a page at a time.  This is more
+   efficient, but uses a page of memory and means that an entire page of input
+   must be ready before any callbacks will fire.  To react as soon as input
+   arrives, set @p bulk to false.
+*/
+SERD_API
+SerdStatus
+serd_reader_start_stream(SerdReader*    me,
+                         FILE*          file,
+                         const uint8_t* name,
+                         bool           bulk);
+
+/**
+   Read a single "chunk" of data during an incremental read.
+
+   This function will read a single top level description, and return.  This
+   may be a directive, statement, or several statements; essentially it reads
+   until a '.' is encountered.  This is particularly useful for reading
+   directly from a pipe or socket.
+*/
+SERD_API
+SerdStatus
+serd_reader_read_chunk(SerdReader* me);
+
+/**
+   Finish an incremental read from a file handle.
+*/
+SERD_API
+SerdStatus
+serd_reader_end_stream(SerdReader* me);
+
+/**
    Read @c file.
 */
 SERD_API
diff --git a/src/reader.c b/src/reader.c
index 70dc7fe5..5c623039 100644
--- a/src/reader.c
+++ b/src/reader.c
@@ -78,7 +78,9 @@ struct SerdReaderImpl {
 	unsigned          next_id;
 	uint8_t*          read_buf;
 	int32_t           read_head;  ///< Offset into read_buf
+	uint8_t           read_byte;  ///< 1-byte 'buffer' used when not paging
 	bool              from_file;  ///< True iff reading from @ref fd
+	bool              paging;  ///< True iff reading a page at a time
 	bool              eof;
 	bool              seen_genid;
 #ifdef SERD_STACK_CHECK
@@ -124,14 +126,19 @@ static inline uint8_t
 eat_byte_safe(SerdReader* reader, const uint8_t byte)
 {
 	assert(peek_byte(reader) == byte);
-	++reader->read_head;
 	switch (byte) {
 	case '\0': reader->eof = true; break;
 	case '\n': ++reader->cur.line; reader->cur.col = 0; break;
 	default:   ++reader->cur.col;
 	}
 
-	if (reader->from_file && (reader->read_head == SERD_PAGE_SIZE)) {
+	if (reader->from_file && !reader->paging) {
+		const int c = fgetc(reader->fd);
+		reader->read_byte = (c == EOF) ? 0 : (uint8_t)c;
+		if (c == EOF) {
+			reader->eof = true;
+		}
+	} else if (++reader->read_head == SERD_PAGE_SIZE && reader->paging) {
 		page(reader);
 	}
 	return byte;
@@ -548,7 +555,7 @@ read_comment(SerdReader* reader)
 {
 	eat_byte_safe(reader, '#');
 	uint8_t c;
-	while (((c = peek_byte(reader)) != 0xA) && (c != 0xD)) {
+	while (((c = peek_byte(reader)) != 0xA) && (c != 0xD) && c) {
 		eat_byte_safe(reader, c);
 	}
 }
@@ -1458,14 +1465,17 @@ static void
 skip_bom(SerdReader* me)
 {
 	const uint8_t* const b = me->read_buf;
-	if (b[0] == 0xEF && b[1] == 0xBB && b[2] == 0xBF) {
+	if (me->paging && b[0] == 0xEF && b[1] == 0xBB && b[2] == 0xBF) {
 		me->read_head += 3;
 	}
 }
 
 SERD_API
 SerdStatus
-serd_reader_read_file_handle(SerdReader* me, FILE* file, const uint8_t* name)
+serd_reader_start_stream(SerdReader*    me,
+                         FILE*          file,
+                         const uint8_t* name,
+                         bool           bulk)
 {
 	const Cursor cur = { name, 1, 1 };
 	me->fd        = file;
@@ -1473,19 +1483,62 @@ serd_reader_read_file_handle(SerdReader* me, FILE* file, const uint8_t* name)
 	me->cur       = cur;
 	me->from_file = true;
 	me->eof       = false;
-	me->read_buf  = (uint8_t*)serd_bufalloc(SERD_PAGE_SIZE);
+	me->paging    = bulk;
+
+	if (bulk) {
+		me->read_buf = (uint8_t*)serd_bufalloc(SERD_PAGE_SIZE);
+		memset(me->read_buf, '\0', SERD_PAGE_SIZE);
+		SerdStatus st = page(me);
+		if (st) {
+			serd_reader_end_stream(me);
+			return st;
+		}
+		skip_bom(me);
+	} else {
+		me->read_buf  = &me->read_byte;
+		me->read_byte = 0;  // Don't read to avoid potentially blocking
+	}
 
-	memset(me->read_buf, '\0', SERD_PAGE_SIZE);
+	return SERD_SUCCESS;
+}
 
-	SerdStatus st = page(me);
-	if (!st) {
-		skip_bom(me);
-		st = read_turtleDoc(me) ? SERD_SUCCESS : SERD_ERR_UNKNOWN;
+SERD_API
+SerdStatus
+serd_reader_read_chunk(SerdReader* me)
+{
+	if (!me->read_byte) {
+		// Read initial byte
+		const int c = fgetc(me->fd);
+		me->read_byte = (c == EOF) ? 0 : (uint8_t)c;
+		if (c == EOF) {
+			me->eof = true;
+			return SERD_FAILURE;
+		}
 	}
+	return read_statement(me) ? SERD_SUCCESS : SERD_FAILURE;
+}
 
-	free(me->read_buf);
+SERD_API
+SerdStatus
+serd_reader_end_stream(SerdReader* me)
+{
+	if (me->paging) {
+		free(me->read_buf);
+	}
 	me->fd       = 0;
 	me->read_buf = NULL;
+	return SERD_SUCCESS;
+}
+
+SERD_API
+SerdStatus
+serd_reader_read_file_handle(SerdReader* me, FILE* file, const uint8_t* name)
+{
+	SerdStatus st = serd_reader_start_stream(me, file, name, true);
+	if (!st) {
+		st = read_turtleDoc(me) ? SERD_SUCCESS : SERD_ERR_UNKNOWN;
+		serd_reader_end_stream(me);
+	}
 	return st;
 }
 
@@ -1499,6 +1552,7 @@ serd_reader_read_string(SerdReader* me, const uint8_t* utf8)
 	me->read_head = 0;
 	me->cur       = cur;
 	me->from_file = false;
+	me->paging    = false;
 	me->eof       = false;
 
 	skip_bom(me);
diff --git a/src/serdi.c b/src/serdi.c
index 74a84992..4c482884 100644
--- a/src/serdi.c
+++ b/src/serdi.c
@@ -46,6 +46,7 @@ print_usage(const char* name, bool error)
 	fprintf(os, "Use - for INPUT to read from standard input.\n\n");
 	fprintf(os, "  -b           Fast bulk output for large serialisations.\n");
 	fprintf(os, "  -c PREFIX    Chop PREFIX from matching blank node IDs.\n");
+	fprintf(os, "  -e           Eat input one character at a time.\n");
 	fprintf(os, "  -f           Keep full URIs in input (don't qualify).\n");
 	fprintf(os, "  -h           Display this help and exit.\n");
 	fprintf(os, "  -i SYNTAX    Input syntax (`turtle' or `ntriples').\n");
@@ -89,6 +90,7 @@ main(int argc, char** argv)
 	SerdSyntax     input_syntax  = SERD_TURTLE;
 	SerdSyntax     output_syntax = SERD_NTRIPLES;
 	bool           from_file     = true;
+	bool           bulk_read     = true;
 	bool           bulk_write    = false;
 	bool           full_uris     = false;
 	const uint8_t* in_name       = NULL;
@@ -103,6 +105,8 @@ main(int argc, char** argv)
 			break;
 		} else if (argv[a][1] == 'b') {
 			bulk_write = true;
+		} else if (argv[a][1] == 'e') {
+			bulk_read = false;
 		} else if (argv[a][1] == 'f') {
 			full_uris = true;
 		} else if (argv[a][1] == 'h') {
@@ -206,9 +210,18 @@ main(int argc, char** argv)
 	serd_writer_chop_blank_prefix(writer, chop_prefix);
 	serd_reader_add_blank_prefix(reader, add_prefix);
 
-	const SerdStatus status = (from_file)
-		? serd_reader_read_file_handle(reader, in_fd, in_name)
-		: serd_reader_read_string(reader, input);
+	SerdStatus status = SERD_SUCCESS;
+	if (!from_file) {
+		status = serd_reader_read_string(reader, input);
+	} else if (bulk_read) {
+		status = serd_reader_read_file_handle(reader, in_fd, in_name);
+	} else {
+		status = serd_reader_start_stream(reader, in_fd, in_name, false);
+		while (!status) {
+			status = serd_reader_read_chunk(reader);
+		}
+		serd_reader_end_stream(reader);
+	}
 
 	serd_reader_free(reader);
 
diff --git a/wscript b/wscript
index d09612f7..c3db493f 100644
--- a/wscript
+++ b/wscript
@@ -9,7 +9,7 @@ from waflib.extras import autowaf as autowaf
 import waflib.Logs as Logs, waflib.Options as Options
 
 # Version of this package (even if built as a child)
-SERD_VERSION       = '0.14.0'
+SERD_VERSION       = '0.15.0'
 SERD_MAJOR_VERSION = '0'
 
 # Library version (UNIX style major, minor, micro)
@@ -389,6 +389,8 @@ def test(ctx):
             flags += ' -f'
         if (num % 3 == 0):
             flags += ' -r http://www.w3.org/'
+        if (num % 7 == 0):
+            flags += ' -e'
         base_uri = 'http://www.w3.org/2001/sw/DataAccess/df1/' + test.replace('\\', '/')
         out_filename = test + '.thru'
         commands += [