t/Makefile cleanup
[git.git] / read-cache.c
index 53da500..da2adf4 100644 (file)
  *
  * Copyright (C) Linus Torvalds, 2005
  */
+#include <stdarg.h>
 #include "cache.h"
 
-const char *sha1_file_directory = NULL;
 struct cache_entry **active_cache = NULL;
-unsigned int active_nr = 0, active_alloc = 0;
-
-void usage(const char *err, ...)
-{
-       va_list args;
-
-       va_start(args, err);
-       vfprintf(stderr, err, args);
-       va_end(args);
-       exit(1);
-}
-
-static unsigned hexval(char c)
-{
-       if (c >= '0' && c <= '9')
-               return c - '0';
-       if (c >= 'a' && c <= 'f')
-               return c - 'a' + 10;
-       if (c >= 'A' && c <= 'F')
-               return c - 'A' + 10;
-       return ~0;
-}
-
-int get_sha1_hex(const char *hex, unsigned char *sha1)
-{
-       int i;
-       for (i = 0; i < 20; i++) {
-               unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
-               if (val & ~0xff)
-                       return -1;
-               *sha1++ = val;
-               hex += 2;
-       }
-       return 0;
-}
-
-char * sha1_to_hex(const unsigned char *sha1)
-{
-       static char buffer[50];
-       static const char hex[] = "0123456789abcdef";
-       char *buf = buffer;
-       int i;
-
-       for (i = 0; i < 20; i++) {
-               unsigned int val = *sha1++;
-               *buf++ = hex[val >> 4];
-               *buf++ = hex[val & 0xf];
-       }
-       return buffer;
-}
-
-/*
- * NOTE! This returns a statically allocated buffer, so you have to be
- * careful about using it. Do a "strdup()" if you need to save the
- * filename.
- */
-char *sha1_file_name(const unsigned char *sha1)
-{
-       int i;
-       static char *name, *base;
-
-       if (!base) {
-               char *sha1_file_directory = getenv(DB_ENVIRONMENT) ? : DEFAULT_DB_ENVIRONMENT;
-               int len = strlen(sha1_file_directory);
-               base = malloc(len + 60);
-               memcpy(base, sha1_file_directory, len);
-               memset(base+len, 0, 60);
-               base[len] = '/';
-               base[len+3] = '/';
-               name = base + len + 1;
-       }
-       for (i = 0; i < 20; i++) {
-               static char hex[] = "0123456789abcdef";
-               unsigned int val = sha1[i];
-               char *pos = name + i*2 + (i > 0);
-               *pos++ = hex[val >> 4];
-               *pos = hex[val & 0xf];
-       }
-       return base;
-}
-
-int check_sha1_signature(unsigned char *sha1, void *map, unsigned long size)
-{
-       unsigned char real_sha1[20];
-       SHA_CTX c;
-
-       SHA1_Init(&c);
-       SHA1_Update(&c, map, size);
-       SHA1_Final(real_sha1, &c);
-       return memcmp(sha1, real_sha1, 20) ? -1 : 0;
-}
-
-void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
-{
-       char *filename = sha1_file_name(sha1);
-       int fd = open(filename, O_RDONLY);
-       struct stat st;
-       void *map;
-
-       if (fd < 0) {
-               perror(filename);
-               return NULL;
-       }
-       if (fstat(fd, &st) < 0) {
-               close(fd);  
-               return NULL;
-       }
-       map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
-       close(fd);
-       if (-1 == (int)(long)map)
-               return NULL;
-       *size = st.st_size;
-       return map;
-}
-
-void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
-{
-       int ret, bytes;
-       z_stream stream;
-       char buffer[8192];
-       char *buf;
-
-       /* Get the data stream */
-       memset(&stream, 0, sizeof(stream));
-       stream.next_in = map;
-       stream.avail_in = mapsize;
-       stream.next_out = buffer;
-       stream.avail_out = sizeof(buffer);
-
-       inflateInit(&stream);
-       ret = inflate(&stream, 0);
-       if (sscanf(buffer, "%10s %lu", type, size) != 2)
-               return NULL;
-
-       bytes = strlen(buffer) + 1;
-       buf = malloc(*size);
-       if (!buf)
-               return NULL;
-
-       memcpy(buf, buffer + bytes, stream.total_out - bytes);
-       bytes = stream.total_out - bytes;
-       if (bytes < *size && ret == Z_OK) {
-               stream.next_out = buf + bytes;
-               stream.avail_out = *size - bytes;
-               while (inflate(&stream, Z_FINISH) == Z_OK)
-                       /* nothing */;
-       }
-       inflateEnd(&stream);
-       return buf;
-}
-
-void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
-{
-       unsigned long mapsize;
-       void *map, *buf;
-
-       map = map_sha1_file(sha1, &mapsize);
-       if (map) {
-               buf = unpack_sha1_file(map, mapsize, type, size);
-               munmap(map, mapsize);
-               return buf;
-       }
-       return NULL;
-}
-
-int write_sha1_file(char *buf, unsigned len, unsigned char *returnsha1)
-{
-       int size;
-       char *compressed;
-       z_stream stream;
-       unsigned char sha1[20];
-       SHA_CTX c;
-
-       /* Set it up */
-       memset(&stream, 0, sizeof(stream));
-       deflateInit(&stream, Z_BEST_COMPRESSION);
-       size = deflateBound(&stream, len);
-       compressed = malloc(size);
-
-       /* Compress it */
-       stream.next_in = buf;
-       stream.avail_in = len;
-       stream.next_out = compressed;
-       stream.avail_out = size;
-       while (deflate(&stream, Z_FINISH) == Z_OK)
-               /* nothing */;
-       deflateEnd(&stream);
-       size = stream.total_out;
-
-       /* Sha1.. */
-       SHA1_Init(&c);
-       SHA1_Update(&c, compressed, size);
-       SHA1_Final(sha1, &c);
-
-       if (write_sha1_buffer(sha1, compressed, size) < 0)
-               return -1;
-       if (returnsha1)
-               memcpy(returnsha1, sha1, 20);
-       return 0;
-}
-
-int write_sha1_buffer(const unsigned char *sha1, void *buf, unsigned int size)
-{
-       char *filename = sha1_file_name(sha1);
-       int fd;
-
-       fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
-       if (fd < 0)
-               return (errno == EEXIST) ? 0 : -1;
-       write(fd, buf, size);
-       close(fd);
-       return 0;
-}
-
-static int error(const char * string)
-{
-       fprintf(stderr, "error: %s\n", string);
-       return -1;
-}
+unsigned int active_nr = 0, active_alloc = 0, active_cache_changed = 0;
 
 int cache_match_stat(struct cache_entry *ce, struct stat *st)
 {
        unsigned int changed = 0;
 
-       if (ce->mtime.sec  != (unsigned int)st->st_mtim.tv_sec ||
-           ce->mtime.nsec != (unsigned int)st->st_mtim.tv_nsec)
+       switch (ntohl(ce->ce_mode) & S_IFMT) {
+       case S_IFREG:
+               changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
+               /* We consider only the owner x bit to be relevant for "mode changes" */
+               if (0100 & (ntohl(ce->ce_mode) ^ st->st_mode))
+                       changed |= MODE_CHANGED;
+               break;
+       case S_IFLNK:
+               changed |= !S_ISLNK(st->st_mode) ? TYPE_CHANGED : 0;
+               break;
+       default:
+               die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
+       }
+       if (ce->ce_mtime.sec != htonl(st->st_mtime))
+               changed |= MTIME_CHANGED;
+       if (ce->ce_ctime.sec != htonl(st->st_ctime))
+               changed |= CTIME_CHANGED;
+
+#ifdef NSEC
+       /*
+        * nsec seems unreliable - not all filesystems support it, so
+        * as long as it is in the inode cache you get right nsec
+        * but after it gets flushed, you get zero nsec.
+        */
+       if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
                changed |= MTIME_CHANGED;
-       if (ce->ctime.sec  != (unsigned int)st->st_ctim.tv_sec ||
-           ce->ctime.nsec != (unsigned int)st->st_ctim.tv_nsec)
+       if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
                changed |= CTIME_CHANGED;
-       if (ce->st_uid != (unsigned int)st->st_uid ||
-           ce->st_gid != (unsigned int)st->st_gid)
+#endif 
+
+       if (ce->ce_uid != htonl(st->st_uid) ||
+           ce->ce_gid != htonl(st->st_gid))
                changed |= OWNER_CHANGED;
-       if (ce->st_mode != (unsigned int)st->st_mode)
-               changed |= MODE_CHANGED;
-       if (ce->st_dev != (unsigned int)st->st_dev ||
-           ce->st_ino != (unsigned int)st->st_ino)
+       if (ce->ce_dev != htonl(st->st_dev) ||
+           ce->ce_ino != htonl(st->st_ino))
                changed |= INODE_CHANGED;
-       if (ce->st_size != (unsigned int)st->st_size)
+       if (ce->ce_size != htonl(st->st_size))
                changed |= DATA_CHANGED;
        return changed;
 }
 
-int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
+int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 {
+       int len1 = flags1 & CE_NAMEMASK;
+       int len2 = flags2 & CE_NAMEMASK;
        int len = len1 < len2 ? len1 : len2;
        int cmp;
 
@@ -262,6 +68,10 @@ int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
                return -1;
        if (len1 > len2)
                return 1;
+       if (flags1 < flags2)
+               return -1;
+       if (flags1 > flags2)
+               return 1;
        return 0;
 }
 
@@ -274,49 +84,187 @@ int cache_name_pos(const char *name, int namelen)
        while (last > first) {
                int next = (last + first) >> 1;
                struct cache_entry *ce = active_cache[next];
-               int cmp = cache_name_compare(name, namelen, ce->name, ce->namelen);
+               int cmp = cache_name_compare(name, namelen, ce->name, htons(ce->ce_flags));
                if (!cmp)
-                       return -next-1;
+                       return next;
                if (cmp < 0) {
                        last = next;
                        continue;
                }
                first = next+1;
        }
-       return first;
+       return -first-1;
+}
+
+/* Remove entry, return true if there are more entries to go.. */
+int remove_entry_at(int pos)
+{
+       active_cache_changed = 1;
+       active_nr--;
+       if (pos >= active_nr)
+               return 0;
+       memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
+       return 1;
 }
 
 int remove_file_from_cache(char *path)
 {
        int pos = cache_name_pos(path, strlen(path));
-       if (pos < 0) {
+       if (pos < 0)
                pos = -pos-1;
-               active_nr--;
-               if (pos < active_nr)
-                       memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos - 1) * sizeof(struct cache_entry *));
-       }
+       while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
+               remove_entry_at(pos);
        return 0;
 }
 
-int add_cache_entry(struct cache_entry *ce, int ok_to_add)
+int same_name(struct cache_entry *a, struct cache_entry *b)
 {
-       int pos;
+       int len = ce_namelen(a);
+       return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
+}
+
+/* We may be in a situation where we already have path/file and path
+ * is being added, or we already have path and path/file is being
+ * added.  Either one would result in a nonsense tree that has path
+ * twice when git-write-tree tries to write it out.  Prevent it.
+ * 
+ * If ok-to-replace is specified, we remove the conflicting entries
+ * from the cache so the caller should recompute the insert position.
+ * When this happens, we return non-zero.
+ */
+static int check_file_directory_conflict(const struct cache_entry *ce,
+                                        int ok_to_replace)
+{
+       int pos, replaced = 0;
+       const char *path = ce->name;
+       int namelen = strlen(path);
+       int stage = ce_stage(ce);
+       char *pathbuf = xmalloc(namelen + 1);
+       char *cp;
+
+       memcpy(pathbuf, path, namelen + 1);
+
+       /*
+        * We are inserting path/file.  Do they have path registered at
+        * the same stage?  We need to do this for all the levels of our
+        * subpath.
+        */
+       cp = pathbuf;
+       while (1) {
+               char *ep = strchr(cp, '/');
+               if (!ep)
+                       break;
+               *ep = 0;    /* first cut it at slash */
+               pos = cache_name_pos(pathbuf,
+                                    htons(create_ce_flags(ep-cp, stage)));
+               if (0 <= pos) {
+                       /* Our leading path component is registered as a file,
+                        * and we are trying to make it a directory.  This is
+                        * bad.
+                        */
+                       if (!ok_to_replace) {
+                               free(pathbuf);
+                               return -1;
+                       }
+                       fprintf(stderr, "removing file '%s' to replace it with a directory to create '%s'.\n", pathbuf, path);
+                       remove_entry_at(pos);
+                       replaced = 1;
+               }
+               *ep = '/';  /* then restore it and go downwards */
+               cp = ep + 1;
+       }
+       free(pathbuf);
+
+       /* Do we have an entry in the cache that makes our path a prefix
+        * of it?  That is, are we creating a file where they already expect
+        * a directory there?
+        */
+       pos = cache_name_pos(path,
+                            htons(create_ce_flags(namelen, stage)));
+
+       /* (0 <= pos) cannot happen because add_cache_entry()
+        * should have taken care of that case.
+        */
+       pos = -pos-1;
+
+       /* pos would point at an existing entry that would come immediately
+        * after our path.  It could be the same as our path in higher stage,
+        * or different path but in a lower stage.
+        *
+        * E.g. when we are inserting path at stage 2,
+        *
+        *        1 path
+        * pos->  3 path
+        *        2 path/file1
+        *        3 path/file1
+        *        2 path/file2
+        *        2 patho
+        *
+        * We need to examine pos, ignore it because it is at different
+        * stage, examine next to find the path/file at stage 2, and
+        * complain.  We need to do this until we are not the leading
+        * path of an existing entry anymore.
+        */
+
+       while (pos < active_nr) {
+               struct cache_entry *other = active_cache[pos];
+               if (strncmp(other->name, path, namelen))
+                       break; /* it is not our "subdirectory" anymore */
+               if ((ce_stage(other) == stage) &&
+                   other->name[namelen] == '/') {
+                       if (!ok_to_replace)
+                               return -1;
+                       fprintf(stderr, "removing file '%s' under '%s' to be replaced with a file\n", other->name, path);
+                       remove_entry_at(pos);
+                       replaced = 1;
+                       continue; /* cycle without updating pos */
+               }
+               pos++;
+       }
+       return replaced;
+}
 
-       pos = cache_name_pos(ce->name, ce->namelen);
+int add_cache_entry(struct cache_entry *ce, int option)
+{
+       int pos;
+       int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
+       int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
+       pos = cache_name_pos(ce->name, htons(ce->ce_flags));
 
        /* existing match? Just replace it */
-       if (pos < 0) {
-               active_cache[-pos-1] = ce;
+       if (pos >= 0) {
+               active_cache_changed = 1;
+               active_cache[pos] = ce;
                return 0;
        }
+       pos = -pos-1;
+
+       /*
+        * Inserting a merged entry ("stage 0") into the index
+        * will always replace all non-merged entries..
+        */
+       if (pos < active_nr && ce_stage(ce) == 0) {
+               while (same_name(active_cache[pos], ce)) {
+                       ok_to_add = 1;
+                       if (!remove_entry_at(pos))
+                               break;
+               }
+       }
 
        if (!ok_to_add)
                return -1;
 
+       if (check_file_directory_conflict(ce, ok_to_replace)) {
+               if (!ok_to_replace)
+                       return -1;
+               pos = cache_name_pos(ce->name, htons(ce->ce_flags));
+               pos = -pos-1;
+       }
+
        /* Make sure the array is big enough .. */
        if (active_nr == active_alloc) {
                active_alloc = alloc_nr(active_alloc);
-               active_cache = realloc(active_cache, active_alloc * sizeof(struct cache_entry *));
+               active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
        }
 
        /* Add it in.. */
@@ -324,6 +272,7 @@ int add_cache_entry(struct cache_entry *ce, int ok_to_add)
        if (active_nr > pos)
                memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
        active_cache[pos] = ce;
+       active_cache_changed = 1;
        return 0;
 }
 
@@ -332,16 +281,15 @@ static int verify_hdr(struct cache_header *hdr, unsigned long size)
        SHA_CTX c;
        unsigned char sha1[20];
 
-       if (hdr->signature != CACHE_SIGNATURE)
+       if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
                return error("bad signature");
-       if (hdr->version != 1)
-               return error("bad version");
+       if (hdr->hdr_version != htonl(2))
+               return error("bad index version");
        SHA1_Init(&c);
-       SHA1_Update(&c, hdr, offsetof(struct cache_header, sha1));
-       SHA1_Update(&c, hdr+1, size - sizeof(*hdr));
+       SHA1_Update(&c, hdr, size - 20);
        SHA1_Final(sha1, &c);
-       if (memcmp(sha1, hdr->sha1, 20))
-               return error("bad header sha1");
+       if (memcmp(sha1, (void *)hdr + size - 20, 20))
+               return error("bad index file sha1 signature");
        return 0;
 }
 
@@ -357,12 +305,7 @@ int read_cache(void)
        if (active_cache)
                return error("more than one cachefile");
        errno = ENOENT;
-       sha1_file_directory = getenv(DB_ENVIRONMENT);
-       if (!sha1_file_directory)
-               sha1_file_directory = DEFAULT_DB_ENVIRONMENT;
-       if (access(sha1_file_directory, X_OK) < 0)
-               return error("no access to SHA1 file directory");
-       fd = open(".dircache/index", O_RDONLY);
+       fd = open(get_index_file(), O_RDONLY);
        if (fd < 0)
                return (errno == ENOENT) ? 0 : error("open failed");
 
@@ -371,8 +314,8 @@ int read_cache(void)
        if (!fstat(fd, &st)) {
                size = st.st_size;
                errno = EINVAL;
-               if (size >= sizeof(struct cache_header))
-                       map = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
+               if (size >= sizeof(struct cache_header) + 20)
+                       map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
        }
        close(fd);
        if (-1 == (int)(long)map)
@@ -382,12 +325,12 @@ int read_cache(void)
        if (verify_hdr(hdr, size) < 0)
                goto unmap;
 
-       active_nr = hdr->entries;
+       active_nr = ntohl(hdr->hdr_entries);
        active_alloc = alloc_nr(active_nr);
        active_cache = calloc(active_alloc, sizeof(struct cache_entry *));
 
        offset = sizeof(*hdr);
-       for (i = 0; i < hdr->entries; i++) {
+       for (i = 0; i < active_nr; i++) {
                struct cache_entry *ce = map + offset;
                offset = offset + ce_size(ce);
                active_cache[i] = ce;
@@ -400,33 +343,67 @@ unmap:
        return error("verify header failed");
 }
 
+#define WRITE_BUFFER_SIZE 8192
+static char write_buffer[WRITE_BUFFER_SIZE];
+static unsigned long write_buffer_len;
+
+static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
+{
+       while (len) {
+               unsigned int buffered = write_buffer_len;
+               unsigned int partial = WRITE_BUFFER_SIZE - buffered;
+               if (partial > len)
+                       partial = len;
+               memcpy(write_buffer + buffered, data, partial);
+               buffered += partial;
+               if (buffered == WRITE_BUFFER_SIZE) {
+                       SHA1_Update(context, write_buffer, WRITE_BUFFER_SIZE);
+                       if (write(fd, write_buffer, WRITE_BUFFER_SIZE) != WRITE_BUFFER_SIZE)
+                               return -1;
+                       buffered = 0;
+               }
+               write_buffer_len = buffered;
+               len -= partial;
+               data += partial;
+       }
+       return 0;
+}
+
+static int ce_flush(SHA_CTX *context, int fd)
+{
+       unsigned int left = write_buffer_len;
+
+       if (left) {
+               write_buffer_len = 0;
+               SHA1_Update(context, write_buffer, left);
+       }
+
+       /* Append the SHA1 signature at the end */
+       SHA1_Final(write_buffer + left, context);
+       left += 20;
+       if (write(fd, write_buffer, left) != left)
+               return -1;
+       return 0;
+}
+
 int write_cache(int newfd, struct cache_entry **cache, int entries)
 {
        SHA_CTX c;
        struct cache_header hdr;
        int i;
 
-       hdr.signature = CACHE_SIGNATURE;
-       hdr.version = 1;
-       hdr.entries = entries;
+       hdr.hdr_signature = htonl(CACHE_SIGNATURE);
+       hdr.hdr_version = htonl(2);
+       hdr.hdr_entries = htonl(entries);
 
        SHA1_Init(&c);
-       SHA1_Update(&c, &hdr, offsetof(struct cache_header, sha1));
-       for (i = 0; i < entries; i++) {
-               struct cache_entry *ce = cache[i];
-               int size = ce_size(ce);
-               SHA1_Update(&c, ce, size);
-       }
-       SHA1_Final(hdr.sha1, &c);
-
-       if (write(newfd, &hdr, sizeof(hdr)) != sizeof(hdr))
+       if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
                return -1;
 
        for (i = 0; i < entries; i++) {
                struct cache_entry *ce = cache[i];
-               int size = ce_size(ce);
-               if (write(newfd, ce, size) != size)
+               if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
                        return -1;
        }
-       return 0;
+       return ce_flush(&c, newfd);
 }