Commit f56f8585c095b1589b5e3dca0ea64ef907756fc9

Carlos Martín Nieto 2012-11-19T22:23:16

indexer: use the packfile streaming API The new API allows us to read the object bit by bit from the packfile, instead of needing it all at once in the packfile. This also allows us to hash the object as it comes in from the network instead of having to try to read it all and failing repeatedly for larger objects. This is only the first step, but it already shows huge improvements when dealing with objects over a few megabytes in size. It reduces the memory needs in some cases, but delta objects still need to be completely in memory and the old inefficent method is still used for that.

diff --git a/src/indexer.c b/src/indexer.c
index a51d903..c331a44 100644
--- a/src/indexer.c
+++ b/src/indexer.c
@@ -38,15 +38,19 @@ struct git_indexer {
 
 struct git_indexer_stream {
 	unsigned int parsed_header :1,
-		opened_pack;
+		opened_pack :1,
+		have_stream :1;
 	struct git_pack_file *pack;
 	git_filebuf pack_file;
 	git_filebuf index_file;
 	git_off_t off;
+	git_off_t entry_start;
+	git_packfile_stream stream;
 	size_t nr_objects;
 	git_vector objects;
 	git_vector deltas;
 	unsigned int fanout[256];
+	git_hash_ctx hash_ctx;
 	git_oid hash;
 	git_transfer_progress_callback progress_cb;
 	void *progress_payload;
@@ -216,6 +220,94 @@ static int store_delta(git_indexer_stream *idx, git_off_t entry_start, size_t en
 	return 0;
 }
 
+static void hash_header(git_hash_ctx *ctx, git_off_t len, git_otype type)
+{
+	char buffer[64];
+	size_t hdrlen;
+
+	hdrlen = git_odb__format_object_header(buffer, sizeof(buffer), len, type);
+	git_hash_update(ctx, buffer, hdrlen);
+}
+
+static int hash_object_stream(git_hash_ctx *ctx, git_packfile_stream *stream)
+{
+	char buffer[8*1024];
+	ssize_t read;
+
+	assert(ctx && stream);
+
+	do {
+		if ((read = git_packfile_stream_read(stream, buffer, sizeof(buffer))) < 0)
+			break;
+
+		git_hash_update(ctx, buffer, read);
+	} while (read > 0);
+
+	if (read < 0)
+		return (int)read;
+
+	return 0;
+}
+
+static int store_cache(git_indexer_stream *idx, git_hash_ctx *ctx, git_off_t entry_start)
+{
+	int i;
+	git_oid oid;
+	void *packed;
+	unsigned int left;
+	struct entry *entry;
+	git_off_t entry_size;
+	git_mwindow *w = NULL;
+	git_mwindow_file *mwf = &idx->pack->mwf;
+	struct git_pack_entry *pentry;
+
+	entry = git__calloc(1, sizeof(*entry));
+	GITERR_CHECK_ALLOC(entry);
+
+	pentry = git__malloc(sizeof(struct git_pack_entry));
+	GITERR_CHECK_ALLOC(pentry);
+
+	git_hash_final(&oid, ctx);
+	entry_size = idx->off - entry_start;
+	if (entry_start > UINT31_MAX) {
+		entry->offset = UINT32_MAX;
+		entry->offset_long = entry_start;
+	} else {
+		entry->offset = (uint32_t)entry_start;
+	}
+
+	git_oid_cpy(&pentry->sha1, &oid);
+	pentry->offset = entry_start;
+	if (git_vector_insert(&idx->pack->cache, pentry) < 0)
+		goto on_error;
+
+	git_oid_cpy(&entry->oid, &oid);
+	entry->crc = crc32(0L, Z_NULL, 0);
+
+	packed = git_mwindow_open(mwf, &w, entry_start, entry_size, &left);
+	if (packed == NULL)
+		goto on_error;
+
+	entry->crc = htonl(crc32(entry->crc, packed, (uInt)entry_size));
+	git_mwindow_close(&w);
+
+	/* Add the object to the list */
+	if (git_vector_insert(&idx->objects, entry) < 0)
+		goto on_error;
+
+	for (i = oid.id[0]; i < 256; ++i) {
+		idx->fanout[i]++;
+	}
+
+	return 0;
+
+on_error:
+	git__free(pentry);
+	git__free(entry);
+
+	return -1;
+}
+
 static int hash_and_save(git_indexer_stream *idx, git_rawobj *obj, git_off_t entry_start)
 {
 	int i;
@@ -349,7 +441,7 @@ int git_indexer_stream_add(git_indexer_stream *idx, const void *data, size_t siz
 	/* As the file grows any windows we try to use will be out of date */
 	git_mwindow_free_all(mwf);
 	while (processed < idx->nr_objects) {
-		git_rawobj obj;
+		git_packfile_stream *stream = &idx->stream;
 		git_off_t entry_start = idx->off;
 		size_t entry_size;
 		git_otype type;
@@ -358,52 +450,63 @@ int git_indexer_stream_add(git_indexer_stream *idx, const void *data, size_t siz
 		if (idx->pack->mwf.size <= idx->off + 20)
 			return 0;
 
-		error = git_packfile_unpack_header(&entry_size, &type, mwf, &w, &idx->off);
-		if (error == GIT_EBUFS) {
-			idx->off = entry_start;
-			return 0;
-		}
-		if (error < 0)
-			return -1;
-
-		git_mwindow_close(&w);
-
-		if (type == GIT_OBJ_REF_DELTA || type == GIT_OBJ_OFS_DELTA) {
-			error = store_delta(idx, entry_start, entry_size, type);
+		if (!idx->have_stream) {
+			error = git_packfile_unpack_header(&entry_size, &type, mwf, &w, &idx->off);
 			if (error == GIT_EBUFS) {
 				idx->off = entry_start;
 				return 0;
 			}
 			if (error < 0)
-				return error;
+				return -1;
+
+			git_mwindow_close(&w);
+			idx->entry_start = entry_start;
+
+			if (type == GIT_OBJ_REF_DELTA || type == GIT_OBJ_OFS_DELTA) {
+				error = store_delta(idx, entry_start, entry_size, type);
+				if (error == GIT_EBUFS) {
+					idx->off = entry_start;
+					return 0;
+				}
+				if (error < 0)
+					return error;
+
+				stats->received_objects++;
+				do_progress_callback(idx, stats);
+				continue;
+			}
 
-			stats->received_objects++;
-			do_progress_callback(idx, stats);
-			continue;
+			/* If we got this far, we create the stream for our object */
+			idx->have_stream = 1;
+			git_hash_ctx_init(&idx->hash_ctx);
+			hash_header(&idx->hash_ctx, entry_size, type);
+			idx->entry_start = entry_start;
+			if (git_packfile_stream_open(stream, idx->pack, idx->off) < 0)
+				goto on_error;
 		}
 
-		idx->off = entry_start;
-		error = git_packfile_unpack(&obj, idx->pack, &idx->off);
-		if (error == GIT_EBUFS) {
-			idx->off = entry_start;
+		error = hash_object_stream(&idx->hash_ctx, stream);
+		idx->off = idx->stream.curpos;
+		if (error == GIT_EBUFS)
 			return 0;
-		}
 		if (error < 0)
-			return -1;
-
-		if (hash_and_save(idx, &obj, entry_start) < 0)
 			goto on_error;
 
-		git__free(obj.data);
+		git_packfile_stream_free(&idx->stream);
+		if (store_cache(idx, &idx->hash_ctx,  idx->entry_start) < 0)
+			goto on_error;
 
 		stats->indexed_objects = (unsigned int)++processed;
 		stats->received_objects++;
+		idx->have_stream = 0;
+
 		do_progress_callback(idx, stats);
 	}
 
 	return 0;
 
 on_error:
+	git_packfile_stream_free(&idx->stream);
 	git_mwindow_free_all(mwf);
 	return -1;
 }
diff --git a/src/odb.c b/src/odb.c
index 63b6828..23b3de9 100644
--- a/src/odb.c
+++ b/src/odb.c
@@ -34,7 +34,7 @@ typedef struct
 
 static int load_alternates(git_odb *odb, const char *objects_dir, int alternate_depth);
 
-static int format_object_header(char *hdr, size_t n, size_t obj_len, git_otype obj_type)
+int git_odb__format_object_header(char *hdr, size_t n, size_t obj_len, git_otype obj_type)
 {
 	const char *type_str = git_object_type2string(obj_type);
 	int len = p_snprintf(hdr, n, "%s %"PRIuZ, type_str, obj_len);
@@ -55,7 +55,7 @@ int git_odb__hashobj(git_oid *id, git_rawobj *obj)
 	if (!obj->data && obj->len != 0)
 		return -1;
 
-	hdrlen = format_object_header(header, sizeof(header), obj->len, obj->type);
+	hdrlen = git_odb__format_object_header(header, sizeof(header), obj->len, obj->type);
 
 	vec[0].data = header;
 	vec[0].len = hdrlen;
@@ -133,7 +133,7 @@ int git_odb__hashfd(git_oid *out, git_file fd, size_t size, git_otype type)
 	if ((error = git_hash_ctx_init(&ctx)) < 0)
 		return -1;
 
-	hdr_len = format_object_header(hdr, sizeof(hdr), size, type);
+	hdr_len = git_odb__format_object_header(hdr, sizeof(hdr), size, type);
 
 	if ((error = git_hash_update(&ctx, hdr, hdr_len)) < 0)
 		goto done;
diff --git a/src/odb.h b/src/odb.h
index e9e33dd..ed4ee7e 100644
--- a/src/odb.h
+++ b/src/odb.h
@@ -46,6 +46,10 @@ struct git_odb {
 int git_odb__hashobj(git_oid *id, git_rawobj *obj);
 
 /*
+ * Format the object header such as it would appear in the on-disk object
+ */
+int git_odb__format_object_header(char *hdr, size_t n, size_t obj_len, git_otype obj_type);
+/*
  * Hash an open file descriptor.
  * This is a performance call when the contents of a fd need to be hashed,
  * but the fd is already open and we have the size of the contents.