# [RFC PATCH 0/2] Parallel pull core 3 messages from 2005-08-01 to 2005-08-01. Participants: barkalow@iabervon.org. Thread: https://gitlist.dev/t/1382 ## barkalow@iabervon.org, 2005-08-01 01:11 Subject: [RFC PATCH 0/2] Parallel pull core Message-ID: URL: https://gitlist.dev/e/Pine.LNX.4.62.0507312055030.23721%40iabervon.org ``` This series makes the core of the pull programs parallel. It should not actually make any difference yet. It arranges to call prefetch() for each object as soon as it is determined to be needed, and then call fetch() on each object once there is nothing left to prefetch. By implementing prefetch(), an implementation can make additional requests while waiting for the data from the earlier ones to come in. Additionally, fetch() will be called in the same order that prefetch() was called, so the implementation can just make a series of requests and get responses. If anyone else is also interested in working on this, it could go into -pu; I've tested it reasonably well, and I'm pretty sure that it doesn't have any effect until the implementations are changed to have prefetch() do something. I'm working on support for it in ssh-pull, and haven't started looking at http-pull support. 1: Adds support to the struct object code to produce struct objects when the type is unknown and the content is unavailable; this allocates memory for the union of the supported types, so it is slightly less efficient, but allows the pull code to track objects it doesn't know anything about (such as the targets to tags). 2: Parallelizes the pull algorithm. -Daniel ``` ## barkalow@iabervon.org, 2005-08-01 01:29 Subject: [PATCH 1/2] Support for making struct objects for absent objects of unknown type Message-ID: URL: https://gitlist.dev/e/Pine.LNX.4.62.0507312112360.23721%40iabervon.org In-Reply-To: ``` This adds support for calling lookup_object_type() with NULL for the type, which will cause it to allocate enough memory for the largest type. This allows struct object_lists for objects that need to be fetched to find out what they are. Signed-off-by: Daniel Barkalow --- object.c | 45 ++++++++++++++++++++++++++++++++++++++++++++- object.h | 8 ++++++++ tree.h | 1 + 3 files changed, 53 insertions(+), 1 deletions(-) 6ed7f76e658c02cb2539a52813fe20d3fd9aa250 diff --git a/object.c b/object.c --- a/object.c +++ b/object.c @@ -99,7 +99,9 @@ void mark_reachable(struct object *obj, struct object *lookup_object_type(const unsigned char *sha1, const char *type) { - if (!strcmp(type, blob_type)) { + if (!type) { + return lookup_unknown_object(sha1); + } else if (!strcmp(type, blob_type)) { return &lookup_blob(sha1)->object; } else if (!strcmp(type, tree_type)) { return &lookup_tree(sha1)->object; @@ -113,6 +115,27 @@ struct object *lookup_object_type(const } } +union any_object { + struct object object; + struct commit commit; + struct tree tree; + struct blob blob; + struct tag tag; +}; + +struct object *lookup_unknown_object(const unsigned char *sha1) +{ + struct object *obj = lookup_object(sha1); + if (!obj) { + union any_object *ret = xmalloc(sizeof(*ret)); + memset(ret, 0, sizeof(*ret)); + created_object(sha1, &ret->object); + ret->object.type = NULL; + return &ret->object; + } + return obj; +} + struct object *parse_object(const unsigned char *sha1) { unsigned long size; @@ -150,3 +173,23 @@ struct object *parse_object(const unsign } return NULL; } + +struct object_list *object_list_insert(struct object *item, + struct object_list **list_p) +{ + struct object_list *new_list = xmalloc(sizeof(struct object_list)); + new_list->item = item; + new_list->next = *list_p; + *list_p = new_list; + return new_list; +} + +unsigned object_list_length(struct object_list *list) +{ + unsigned ret = 0; + while (list) { + list = list->next; + ret++; + } + return ret; +} diff --git a/object.h b/object.h --- a/object.h +++ b/object.h @@ -31,8 +31,16 @@ void created_object(const unsigned char /** Returns the object, having parsed it to find out what it is. **/ struct object *parse_object(const unsigned char *sha1); +/** Returns the object, with potentially excess memory allocated. **/ +struct object *lookup_unknown_object(const unsigned char *sha1); + void add_ref(struct object *refer, struct object *target); void mark_reachable(struct object *obj, unsigned int mask); +struct object_list *object_list_insert(struct object *item, + struct object_list **list_p); + +unsigned object_list_length(struct object_list *list); + #endif /* OBJECT_H */ diff --git a/tree.h b/tree.h --- a/tree.h +++ b/tree.h @@ -14,6 +14,7 @@ struct tree_entry_list { unsigned int mode; char *name; union { + struct object *any; struct tree *tree; struct blob *blob; } item; ``` ## barkalow@iabervon.org, 2005-08-01 01:32 Subject: [PATCH 2/2] Parallelize pull algorithm (no change to behavior). Message-ID: URL: https://gitlist.dev/e/Pine.LNX.4.62.0507312129430.23721%40iabervon.org In-Reply-To: ``` This patch makes the core pull algorithm request all of the objects it can (with prefetch()) before actually reading them (with fetch()). Future patches (in a later series) will make use of this behavior to have multiple requests in flight at the same time, reducing the need for round-trips. Signed-off-by: Daniel Barkalow --- http-pull.c | 4 ++ local-pull.c | 4 ++ pull.c | 128 ++++++++++++++++++++++++++++++++++------------------------ pull.h | 7 +++ ssh-pull.c | 4 ++ 5 files changed, 93 insertions(+), 54 deletions(-) 02aae5a570341c5b86b914d59732009f015800d8 diff --git a/http-pull.c b/http-pull.c --- a/http-pull.c +++ b/http-pull.c @@ -68,6 +68,10 @@ static size_t fwrite_sha1_file(void *ptr return size; } +void prefetch(unsigned char *sha1) +{ +} + int fetch(unsigned char *sha1) { char *hex = sha1_to_hex(sha1); diff --git a/local-pull.c b/local-pull.c --- a/local-pull.c +++ b/local-pull.c @@ -11,6 +11,10 @@ static int use_filecopy = 1; static char *path; /* "Remote" git repository */ +void prefetch(unsigned char *sha1) +{ +} + int fetch(unsigned char *sha1) { static int object_name_start = -1; diff --git a/pull.c b/pull.c --- a/pull.c +++ b/pull.c @@ -17,11 +17,8 @@ int get_all = 0; int get_verbosely = 0; static unsigned char current_commit_sha1[20]; -static const char commitS[] = "commit"; -static const char treeS[] = "tree"; -static const char blobS[] = "blob"; - -void pull_say(const char *fmt, const char *hex) { +void pull_say(const char *fmt, const char *hex) +{ if (get_verbosely) fprintf(stderr, fmt, hex); } @@ -48,93 +45,114 @@ static int make_sure_we_have_it(const ch return status; } -static int process_unknown(unsigned char *sha1); +static int process(unsigned char *sha1, const char *type); -static int process_tree(unsigned char *sha1) +static int process_tree(struct tree *tree) { - struct tree *tree = lookup_tree(sha1); struct tree_entry_list *entries; if (parse_tree(tree)) return -1; for (entries = tree->entries; entries; entries = entries->next) { - const char *what = entries->directory ? treeS : blobS; - if (make_sure_we_have_it(what, entries->item.tree->object.sha1)) + if (process(entries->item.any->sha1, + entries->directory ? tree_type : blob_type)) return -1; - if (entries->directory) { - if (process_tree(entries->item.tree->object.sha1)) - return -1; - } } return 0; } -static int process_commit(unsigned char *sha1) +static int process_commit(struct commit *commit) { - struct commit *obj = lookup_commit(sha1); - - if (make_sure_we_have_it(commitS, sha1)) + if (parse_commit(commit)) return -1; - if (parse_commit(obj)) - return -1; + memcpy(current_commit_sha1, commit->object.sha1, 20); if (get_tree) { - if (make_sure_we_have_it(treeS, obj->tree->object.sha1)) - return -1; - if (process_tree(obj->tree->object.sha1)) + if (process(commit->tree->object.sha1, tree_type)) return -1; if (!get_all) get_tree = 0; } if (get_history) { - struct commit_list *parents = obj->parents; + struct commit_list *parents = commit->parents; for (; parents; parents = parents->next) { if (has_sha1_file(parents->item->object.sha1)) continue; - if (make_sure_we_have_it(NULL, - parents->item->object.sha1)) { - /* The server might not have it, and - * we don't mind. - */ - continue; - } - if (process_commit(parents->item->object.sha1)) + if (process(parents->item->object.sha1, + commit_type)) return -1; - memcpy(current_commit_sha1, sha1, 20); } } return 0; } -static int process_tag(unsigned char *sha1) +static int process_tag(struct tag *tag) { - struct tag *obj = lookup_tag(sha1); - - if (parse_tag(obj)) + if (parse_tag(tag)) return -1; - return process_unknown(obj->tagged->sha1); + return process(tag->tagged->sha1, NULL); } -static int process_unknown(unsigned char *sha1) +static struct object_list *process_queue = NULL; +static struct object_list **process_queue_end = &process_queue; + +static int process(unsigned char *sha1, const char *type) { struct object *obj; - if (make_sure_we_have_it("object", sha1)) - return -1; - obj = parse_object(sha1); - if (!obj) - return error("Unable to parse object %s", sha1_to_hex(sha1)); - if (obj->type == commit_type) - return process_commit(sha1); - if (obj->type == tree_type) - return process_tree(sha1); - if (obj->type == blob_type) + if (has_sha1_file(sha1)) return 0; - if (obj->type == tag_type) - return process_tag(sha1); - return error("Unable to determine requirement of type %s for %s", - obj->type, sha1_to_hex(sha1)); + obj = lookup_object_type(sha1, type); + object_list_insert(obj, process_queue_end); + process_queue_end = &(*process_queue_end)->next; + + fprintf(stderr, "prefetch %s\n", sha1_to_hex(sha1)); + prefetch(sha1); + + return 0; +} + +static int loop(void) +{ + while (process_queue) { + struct object *obj = process_queue->item; + fprintf(stderr, "%d objects to pull\n", + object_list_length(process_queue)); + process_queue = process_queue->next; + if (!process_queue) + process_queue_end = &process_queue; + + fprintf(stderr, "fetch %s\n", sha1_to_hex(obj->sha1)); + + if (make_sure_we_have_it(obj->type ?: "object", + obj->sha1)) + return -1; + if (!obj->type) + parse_object(obj->sha1); + if (obj->type == commit_type) { + if (process_commit((struct commit *)obj)) + return -1; + continue; + } + if (obj->type == tree_type) { + if (process_tree((struct tree *)obj)) + return -1; + continue; + } + if (obj->type == blob_type) { + continue; + } + if (obj->type == tag_type) { + if (process_tag((struct tag *)obj)) + return -1; + continue; + } + return error("Unable to determine requirements " + "of type %s for %s", + obj->type, sha1_to_hex(obj->sha1)); + }; + return 0; } static int interpret_target(char *target, unsigned char *sha1) @@ -164,7 +182,9 @@ int pull(char *target) if (interpret_target(target, sha1)) return error("Could not interpret %s as something to pull", target); - if (process_unknown(sha1)) + if (process(sha1, NULL)) + return -1; + if (loop()) return -1; if (write_ref) { diff --git a/pull.h b/pull.h --- a/pull.h +++ b/pull.h @@ -9,6 +9,13 @@ extern int fetch(unsigned char *sha1); /* + * Fetch the specified object and store it locally; fetch() will be + * called later to determine success. To be provided by the particular + * implementation. + */ +extern void prefetch(unsigned char *sha1); + +/* * Fetch ref (relative to $GIT_DIR/refs) from the remote, and store * the 20-byte SHA1 in sha1. Return 0 on success, -1 on failure. To * be provided by the particular implementation. diff --git a/ssh-pull.c b/ssh-pull.c --- a/ssh-pull.c +++ b/ssh-pull.c @@ -10,6 +10,10 @@ static int fd_out; static unsigned char remote_version = 0; static unsigned char local_version = 1; +void prefetch(unsigned char *sha1) +{ +} + int fetch(unsigned char *sha1) { int ret; ```