pack-object: slightly more efficient
[gitweb.git] / pack-redundant.c
index 28b82ee65aed94d2a738602679e0f7d55921ddb3..cd81f5a66eae2a4c1fdab4086b896d6a06dcf687 100644 (file)
@@ -8,22 +8,24 @@
 
 #include "cache.h"
 
+#define BLKSIZE 512
+
 static const char pack_redundant_usage[] =
 "git-pack-redundant [ --verbose ] [ --alt-odb ] < --all | <.pack filename> ...>";
 
-int load_all_packs = 0, verbose = 0, alt_odb = 0;
+static int load_all_packs = 0, verbose = 0, alt_odb = 0;
 
 struct llist_item {
        struct llist_item *next;
-       char *sha1;
+       unsigned char *sha1;
 };
-struct llist {
+static struct llist {
        struct llist_item *front;
        struct llist_item *back;
        size_t size;
 } *all_objects; /* all objects which must be present in local packfiles */
 
-struct pack_list {
+static struct pack_list {
        struct pack_list *next;
        struct packed_git *pack;
        struct llist *unique_objects;
@@ -35,23 +37,47 @@ struct pll {
        struct pack_list *pl;
 };
 
-inline void llist_free(struct llist *list)
+static struct llist_item *free_nodes = NULL;
+
+static inline void llist_item_put(struct llist_item *item)
+{
+       item->next = free_nodes;
+       free_nodes = item;
+}
+
+static inline struct llist_item *llist_item_get(void)
+{
+       struct llist_item *new;
+       if ( free_nodes ) {
+               new = free_nodes;
+               free_nodes = free_nodes->next;
+       } else {
+               int i = 1;
+               new = xmalloc(sizeof(struct llist_item) * BLKSIZE);
+               for(;i < BLKSIZE; i++) {
+                       llist_item_put(&new[i]);
+               }
+       }
+       return new;
+}
+
+static void llist_free(struct llist *list)
 {
        while((list->back = list->front)) {
                list->front = list->front->next;
-               free(list->back);
+               llist_item_put(list->back);
        }
        free(list);
 }
 
-inline void llist_init(struct llist **list)
+static inline void llist_init(struct llist **list)
 {
        *list = xmalloc(sizeof(struct llist));
        (*list)->front = (*list)->back = NULL;
        (*list)->size = 0;
 }
 
-struct llist * llist_copy(struct llist *list)
+static struct llist * llist_copy(struct llist *list)
 {
        struct llist *ret;
        struct llist_item *new, *old, *prev;
@@ -61,13 +87,13 @@ struct llist * llist_copy(struct llist *list)
        if ((ret->size = list->size) == 0)
                return ret;
 
-       new = ret->front = xmalloc(sizeof(struct llist_item));
+       new = ret->front = llist_item_get();
        new->sha1 = list->front->sha1;
 
        old = list->front->next;
        while (old) {
                prev = new;
-               new = xmalloc(sizeof(struct llist_item));
+               new = llist_item_get();
                prev->next = new;
                new->sha1 = old->sha1;
                old = old->next;
@@ -78,10 +104,11 @@ struct llist * llist_copy(struct llist *list)
        return ret;
 }
 
-inline struct llist_item * llist_insert(struct llist *list,
-                                       struct llist_item *after, char *sha1)
+static inline struct llist_item * llist_insert(struct llist *list,
+                                              struct llist_item *after,
+                                              unsigned char *sha1)
 {
-       struct llist_item *new = xmalloc(sizeof(struct llist_item));
+       struct llist_item *new = llist_item_get();
        new->sha1 = sha1;
        new->next = NULL;
 
@@ -101,13 +128,12 @@ inline struct llist_item * llist_insert(struct llist *list,
        return new;
 }
 
-inline struct llist_item * llist_insert_back(struct llist *list, char *sha1)
+static inline struct llist_item *llist_insert_back(struct llist *list, unsigned char *sha1)
 {
        return llist_insert(list, list->back, sha1);
 }
 
-inline struct llist_item * llist_insert_sorted_unique(struct llist *list,
-                                       char *sha1, struct llist_item *hint)
+static inline struct llist_item *llist_insert_sorted_unique(struct llist *list, unsigned char *sha1, struct llist_item *hint)
 {
        struct llist_item *prev = NULL, *l;
 
@@ -127,41 +153,8 @@ inline struct llist_item * llist_insert_sorted_unique(struct llist *list,
        return llist_insert_back(list, sha1);
 }
 
-/* computes A\B */
-void llist_sorted_difference_inplace(struct llist *A,
-                                    struct llist *B)
-{
-       struct llist_item *prev, *a, *b, *x;
-
-       prev = a = A->front;
-       b = B->front;
-
-       while (a != NULL && b != NULL) {
-               int cmp = memcmp(a->sha1, b->sha1, 20);
-               if (!cmp) {
-                       x = a;
-                       if (a == A->front)
-                               A->front = a->next;
-                       a = prev->next = a->next;
-
-                       if (a == NULL) /* end of list */
-                               A->back = prev;
-                       A->size--;
-                       free(x);
-                       b = b->next;
-               } else
-                       if (cmp > 0)
-                               b = b->next;
-                       else {
-                               prev = a;
-                               a = a->next;
-                       }
-       }
-}
-
 /* returns a pointer to an item in front of sha1 */
-inline struct llist_item * llist_sorted_remove(struct llist *list, char *sha1,
-                                              struct llist_item *hint)
+static inline struct llist_item * llist_sorted_remove(struct llist *list, const unsigned char *sha1, struct llist_item *hint)
 {
        struct llist_item *prev, *l;
 
@@ -184,7 +177,7 @@ inline struct llist_item * llist_sorted_remove(struct llist *list, char *sha1,
                                prev->next = l->next;
                        if (l == list->back)
                                list->back = prev;
-                       free(l);
+                       llist_item_put(l);
                        list->size--;
                        return prev;
                }
@@ -194,7 +187,22 @@ inline struct llist_item * llist_sorted_remove(struct llist *list, char *sha1,
        return prev;
 }
 
-inline struct pack_list * pack_list_insert(struct pack_list **pl,
+/* computes A\B */
+static void llist_sorted_difference_inplace(struct llist *A,
+                                    struct llist *B)
+{
+       struct llist_item *hint, *b;
+
+       hint = NULL;
+       b = B->front;
+
+       while (b) {
+               hint = llist_sorted_remove(A, b->sha1, hint);
+               b = b->next;
+       }
+}
+
+static inline struct pack_list * pack_list_insert(struct pack_list **pl,
                                           struct pack_list *entry)
 {
        struct pack_list *p = xmalloc(sizeof(struct pack_list));
@@ -204,7 +212,7 @@ inline struct pack_list * pack_list_insert(struct pack_list **pl,
        return p;
 }
 
-inline size_t pack_list_size(struct pack_list *pl)
+static inline size_t pack_list_size(struct pack_list *pl)
 {
        size_t ret = 0;
        while(pl) {
@@ -214,10 +222,11 @@ inline size_t pack_list_size(struct pack_list *pl)
        return ret;
 }
 
-struct pack_list * pack_list_difference(struct pack_list *A,
-                                       struct pack_list *B)
+static struct pack_list * pack_list_difference(const struct pack_list *A,
+                                              const struct pack_list *B)
 {
-       struct pack_list *ret, *pl;
+       struct pack_list *ret;
+       const struct pack_list *pl;
 
        if (A == NULL)
                return NULL;
@@ -234,7 +243,7 @@ struct pack_list * pack_list_difference(struct pack_list *A,
        return ret;
 }
 
-void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
+static void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
 {
        int p1_off, p2_off;
        void *p1_base, *p2_base;
@@ -266,54 +275,68 @@ void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
        }
 }
 
+static void pll_free(struct pll *l)
+{
+       struct pll *old;
+       struct pack_list *opl;
+
+       while (l) {
+               old = l;
+               while (l->pl) {
+                       opl = l->pl;
+                       l->pl = opl->next;
+                       free(opl);
+               }
+               l = l->next;
+               free(old);
+       }
+}
+
 /* all the permutations have to be free()d at the same time,
  * since they refer to each other
  */
-struct pll * get_all_permutations(struct pack_list *list)
+static struct pll * get_permutations(struct pack_list *list, int n)
 {
-       struct pll *subset, *pll, *new_pll = NULL; /*silence warning*/
+       struct pll *subset, *ret = NULL, *new_pll = NULL, *pll;
 
-       if (list == NULL)
+       if (list == NULL || pack_list_size(list) < n || n == 0)
                return NULL;
 
-       if (list->next == NULL) {
-               new_pll = xmalloc(sizeof(struct pll));
-               new_pll->next = NULL;
-               new_pll->pl = list;
-               return new_pll;
+       if (n == 1) {
+               while (list) {
+                       new_pll = xmalloc(sizeof(pll));
+                       new_pll->pl = NULL;
+                       pack_list_insert(&new_pll->pl, list);
+                       new_pll->next = ret;
+                       ret = new_pll;
+                       list = list->next;
+               }
+               return ret;
        }
 
-       pll = subset = get_all_permutations(list->next);
-       while (pll) {
-               new_pll = xmalloc(sizeof(struct pll));
-               new_pll->next = pll->next;
-               pll->next = new_pll;
-
-               new_pll->pl = xmalloc(sizeof(struct pack_list));
-               memcpy(new_pll->pl, list, sizeof(struct pack_list));
-               new_pll->pl->next = pll->pl;
-
-               pll = new_pll->next;
+       while (list->next) {
+               subset = get_permutations(list->next, n - 1);
+               while (subset) {
+                       new_pll = xmalloc(sizeof(pll));
+                       new_pll->pl = subset->pl;
+                       pack_list_insert(&new_pll->pl, list);
+                       new_pll->next = ret;
+                       ret = new_pll;
+                       subset = subset->next;
+               }
+               list = list->next;
        }
-       /* add ourself to the end */
-       new_pll->next = xmalloc(sizeof(struct pll));
-       new_pll->next->pl = xmalloc(sizeof(struct pack_list));
-       new_pll->next->next = NULL;
-       memcpy(new_pll->next->pl, list, sizeof(struct pack_list));
-       new_pll->next->pl->next = NULL;
-
-       return subset;
+       return ret;
 }
 
-int is_superset(struct pack_list *pl, struct llist *list)
+static int is_superset(struct pack_list *pl, struct llist *list)
 {
        struct llist *diff;
 
        diff = llist_copy(list);
 
        while (pl) {
-               llist_sorted_difference_inplace(diff,
-                                               pl->all_objects);
+               llist_sorted_difference_inplace(diff, pl->all_objects);
                if (diff->size == 0) { /* we're done */
                        llist_free(diff);
                        return 1;
@@ -324,7 +347,7 @@ int is_superset(struct pack_list *pl, struct llist *list)
        return 0;
 }
 
-size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
+static size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
 {
        size_t ret = 0;
        int p1_off, p2_off;
@@ -355,14 +378,14 @@ size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
 }
 
 /* another O(n^2) function ... */
-size_t get_pack_redundancy(struct pack_list *pl)
+static size_t get_pack_redundancy(struct pack_list *pl)
 {
        struct pack_list *subset;
+       size_t ret = 0;
 
        if (pl == NULL)
                return 0;
 
-       size_t ret = 0;
        while ((subset = pl->next)) {
                while(subset) {
                        ret += sizeof_union(pl->pack, subset->pack);
@@ -373,7 +396,7 @@ size_t get_pack_redundancy(struct pack_list *pl)
        return ret;
 }
 
-inline size_t pack_set_bytecount(struct pack_list *pl)
+static inline size_t pack_set_bytecount(struct pack_list *pl)
 {
        size_t ret = 0;
        while (pl) {
@@ -384,13 +407,14 @@ inline size_t pack_set_bytecount(struct pack_list *pl)
        return ret;
 }
 
-void minimize(struct pack_list **min)
+static void minimize(struct pack_list **min)
 {
        struct pack_list *pl, *unique = NULL,
                *non_unique = NULL, *min_perm = NULL;
        struct pll *perm, *perm_all, *perm_ok = NULL, *new_perm;
        struct llist *missing;
        size_t min_perm_size = (size_t)-1, perm_size;
+       int n;
 
        pl = local_packs;
        while (pl) {
@@ -404,8 +428,7 @@ void minimize(struct pack_list **min)
        missing = llist_copy(all_objects);
        pl = unique;
        while (pl) {
-               llist_sorted_difference_inplace(missing,
-                                               pl->all_objects);
+               llist_sorted_difference_inplace(missing, pl->all_objects);
                pl = pl->next;
        }
 
@@ -416,17 +439,21 @@ void minimize(struct pack_list **min)
        }
 
        /* find the permutations which contain all missing objects */
-       perm_all = perm = get_all_permutations(non_unique);
-       while (perm) {
-               if (is_superset(perm->pl, missing)) {
-                       new_perm = xmalloc(sizeof(struct pll));
-                       new_perm->pl = perm->pl;
-                       new_perm->next = perm_ok;
-                       perm_ok = new_perm;
+       for (n = 1; n <= pack_list_size(non_unique) && !perm_ok; n++) {
+               perm_all = perm = get_permutations(non_unique, n);
+               while (perm) {
+                       if (is_superset(perm->pl, missing)) {
+                               new_perm = xmalloc(sizeof(struct pll));
+                               memcpy(new_perm, perm, sizeof(struct pll));
+                               new_perm->next = perm_ok;
+                               perm_ok = new_perm;
+                       }
+                       perm = perm->next;
                }
-               perm = perm->next;
+               if (perm_ok)
+                       break;
+               pll_free(perm_all);
        }
-       
        if (perm_ok == NULL)
                die("Internal error: No complete sets found!\n");
 
@@ -449,16 +476,14 @@ void minimize(struct pack_list **min)
        }
 }
 
-void load_all_objects()
+static void load_all_objects(void)
 {
        struct pack_list *pl = local_packs;
        struct llist_item *hint, *l;
-       int i;
 
        llist_init(&all_objects);
 
        while (pl) {
-               i = 0;
                hint = NULL;
                l = pl->all_objects->front;
                while (l) {
@@ -477,7 +502,7 @@ void load_all_objects()
 }
 
 /* this scales like O(n^2) */
-void cmp_packs()
+static void cmp_local_packs(void)
 {
        struct pack_list *subset, *pl = local_packs;
 
@@ -486,25 +511,34 @@ void cmp_packs()
                        cmp_two_packs(pl, subset);
                pl = pl->next;
        }
+}
 
-       pl = altodb_packs;
-       while (pl) {
-               subset = local_packs;
-               while (subset) {
-                       llist_sorted_difference_inplace(subset->unique_objects,
-                                                       pl->all_objects);
-                       subset = subset->next;
+static void scan_alt_odb_packs(void)
+{
+       struct pack_list *local, *alt;
+
+       alt = altodb_packs;
+       while (alt) {
+               local = local_packs;
+               while (local) {
+                       llist_sorted_difference_inplace(local->unique_objects,
+                                                       alt->all_objects);
+                       local = local->next;
                }
-               pl = pl->next;
+               llist_sorted_difference_inplace(all_objects, alt->all_objects);
+               alt = alt->next;
        }
 }
 
-struct pack_list * add_pack(struct packed_git *p)
+static struct pack_list * add_pack(struct packed_git *p)
 {
        struct pack_list l;
        size_t off;
        void *base;
 
+       if (!p->pack_local && !(alt_odb || verbose))
+               return NULL;
+
        l.pack = p;
        llist_init(&l.all_objects);
 
@@ -512,17 +546,17 @@ struct pack_list * add_pack(struct packed_git *p)
        base = (void *)p->index_base;
        while (off <= p->index_size - 3 * 20) {
                llist_insert_back(l.all_objects, base + off);
-               off+=24;
+               off += 24;
        }
        /* this list will be pruned in cmp_two_packs later */
        l.unique_objects = llist_copy(l.all_objects);
        if (p->pack_local)
                return pack_list_insert(&local_packs, &l);
        else
-               return alt_odb ? pack_list_insert(&altodb_packs, &l) : NULL;
+               return pack_list_insert(&altodb_packs, &l);
 }
 
-struct pack_list * add_pack_file(char *filename)
+static struct pack_list * add_pack_file(char *filename)
 {
        struct packed_git *p = packed_git;
 
@@ -537,7 +571,7 @@ struct pack_list * add_pack_file(char *filename)
        die("Filename %s not found in packed_git\n", filename);
 }
 
-void load_all()
+static void load_all(void)
 {
        struct packed_git *p = packed_git;
 
@@ -551,6 +585,11 @@ int main(int argc, char **argv)
 {
        int i;
        struct pack_list *min, *red, *pl;
+       struct llist *ignore;
+       unsigned char *sha1;
+       char buf[42]; /* 40 byte sha1 + \n + \0 */
+
+       setup_git_directory();
 
        for (i = 1; i < argc; i++) {
                const char *arg = argv[i];
@@ -587,11 +626,31 @@ int main(int argc, char **argv)
        if (local_packs == NULL)
                die("Zero packs found!\n");
 
-       cmp_packs();
-
        load_all_objects();
 
+       cmp_local_packs();
+       if (alt_odb)
+               scan_alt_odb_packs();
+
+       /* ignore objects given on stdin */
+       llist_init(&ignore);
+       if (!isatty(0)) {
+               while (fgets(buf, sizeof(buf), stdin)) {
+                       sha1 = xmalloc(20);
+                       if (get_sha1_hex(buf, sha1))
+                               die("Bad sha1 on stdin: %s", buf);
+                       llist_insert_sorted_unique(ignore, sha1, NULL);
+               }
+       }
+       llist_sorted_difference_inplace(all_objects, ignore);
+       pl = local_packs;
+       while (pl) {
+               llist_sorted_difference_inplace(pl->unique_objects, ignore);
+               pl = pl->next;
+       }
+
        minimize(&min);
+
        if (verbose) {
                fprintf(stderr, "There are %lu packs available in alt-odbs.\n",
                        (unsigned long)pack_list_size(altodb_packs));
@@ -616,6 +675,9 @@ int main(int argc, char **argv)
                       pl->pack->pack_name);
                pl = pl->next;
        }
+       if (verbose)
+               fprintf(stderr, "%luMB of redundant packs in total.\n",
+                       (unsigned long)pack_set_bytecount(red)/(1024*1024));
 
        return 0;
 }