]> arthur.barton.de Git - bup.git/blobdiff - lib/bup/git.py
Combine and speed up idx->midx and bupindex merge
[bup.git] / lib / bup / git.py
index 6a51354708f3dcaa61eac472a8a38216499d612d..cd1682afc009b2f3b09d46fb6a8cf0b00e0c3cf7 100644 (file)
@@ -2,10 +2,9 @@
 bup repositories are in Git format. This library allows us to
 interact with the Git data structures.
 """
-import os, zlib, time, subprocess, struct, stat, re, tempfile
-import heapq
+import os, sys, zlib, time, subprocess, struct, stat, re, tempfile
 from bup.helpers import *
-from bup import _helpers
+from bup import _helpers, path
 
 MIDX_VERSION = 2
 
@@ -40,9 +39,13 @@ def repo(sub = ''):
 
 
 def auto_midx(objdir):
-    main_exe = os.environ.get('BUP_MAIN_EXE') or sys.argv[0]
-    args = [main_exe, 'midx', '--auto', '--dir', objdir]
-    rv = subprocess.call(args, stdout=open('/dev/null', 'w'))
+    args = [path.exe(), 'midx', '--auto', '--dir', objdir]
+    try:
+        rv = subprocess.call(args, stdout=open('/dev/null', 'w'))
+    except OSError, e:
+        # make sure 'args' gets printed to help with debugging
+        add_error('%r: exception: %s' % (args, e))
+        raise
     if rv:
         add_error('%r: returned %d' % (args, rv))
 
@@ -140,7 +143,7 @@ def _decode_packobj(buf):
 class PackIdx:
     def __init__(self):
         assert(0)
-    
+
     def find_offset(self, hash):
         """Get the offset of an object inside the index file."""
         idx = self._idx_from_hash(hash)
@@ -222,7 +225,7 @@ class PackIdxV2(PackIdx):
         ofs = struct.unpack('!I', str(buffer(self.ofstable, idx*4, 4)))[0]
         if ofs & 0x80000000:
             idx64 = ofs & 0x7fffffff
-            ofs = struct.unpack('!I',
+            ofs = struct.unpack('!Q',
                                 str(buffer(self.ofs64table, idx64*8, 8)))[0]
         return ofs
 
@@ -399,7 +402,10 @@ class PackIdxList:
                                 log(('warning: index %s missing\n' +
                                     '  used by %s\n') % (n, mxf))
                                 broken += 1
-                        if not broken:
+                        if broken:
+                            del mx
+                            unlink(full)
+                        else:
                             midxl.append(mx)
                 midxl.sort(lambda x,y: -cmp(len(x),len(y)))
                 for ix in midxl:
@@ -420,12 +426,32 @@ class PackIdxList:
             for f in os.listdir(self.dir):
                 full = os.path.join(self.dir, f)
                 if f.endswith('.idx') and not d.get(full):
-                    ix = open_idx(full)
+                    try:
+                        ix = open_idx(full)
+                    except GitError, e:
+                        add_error(e)
+                        continue
                     d[full] = ix
             self.packs = list(set(d.values()))
         debug1('PackIdxList: using %d index%s.\n'
             % (len(self.packs), len(self.packs)!=1 and 'es' or ''))
 
+    def packname_containing(self, hash):
+        # figure out which pack contains a given hash.
+        # FIXME: if the midx file format would just *store* this information,
+        # we could calculate it a lot more efficiently.  But it's not needed
+        # often, so let's do it like this.
+        for f in os.listdir(self.dir):
+            if f.endswith('.idx'):
+                full = os.path.join(self.dir, f)
+                try:
+                    ix = open_idx(full)
+                except GitError, e:
+                    add_error(e)
+                    continue
+                if ix.exists(hash):
+                    return full
+
     def add(self, hash):
         """Insert an additional object in the list."""
         self.also[hash] = 1
@@ -462,8 +488,10 @@ def open_idx(filename):
             else:
                 raise GitError('%s: expected idx file version 2, got %d'
                                % (filename, version))
-        else:
+        elif len(header) == 8 and header[0:4] < '\377tOc':
             return PackIdxV1(filename, f)
+        else:
+            raise GitError('%s: unrecognized idx file header' % filename)
     elif filename.endswith('.midx'):
         return PackMidx(filename)
     else:
@@ -472,60 +500,44 @@ def open_idx(filename):
 
 def idxmerge(idxlist, final_progress=True):
     """Generate a list of all the objects reachable in a PackIdxList."""
-    total = sum(len(i) for i in idxlist)
-    iters = (iter(i) for i in idxlist)
-    heap = [(next(it), it) for it in iters]
-    heapq.heapify(heap)
-    count = 0
-    last = None
-    while heap:
-        if (count % 10024) == 0:
-            progress('Reading indexes: %.2f%% (%d/%d)\r'
-                     % (count*100.0/total, count, total))
-        (e, it) = heap[0]
-        if e != last:
-            yield e
-            last = e
-        count += 1
-        e = next(it)
-        if e:
-            heapq.heapreplace(heap, (e, it))
-        else:
-            heapq.heappop(heap)
-    if final_progress:
-        log('Reading indexes: %.2f%% (%d/%d), done.\n' % (100, total, total))
+    def pfunc(count, total):
+        progress('Reading indexes: %.2f%% (%d/%d)\r'
+                 % (count*100.0/total, count, total))
+    def pfinal(count, total):
+        if final_progress:
+            log('Reading indexes: %.2f%% (%d/%d), done.\n' % (100, total, total))
+    return merge_iter(idxlist, 10024, pfunc, pfinal)
+
 
+def _make_objcache():
+    return PackIdxList(repo('objects/pack'))
 
 class PackWriter:
     """Writes Git objects insid a pack file."""
-    def __init__(self, objcache_maker=None):
+    def __init__(self, objcache_maker=_make_objcache):
         self.count = 0
         self.outbytes = 0
         self.filename = None
         self.file = None
+        self.idx = None
         self.objcache_maker = objcache_maker
         self.objcache = None
 
     def __del__(self):
         self.close()
 
-    def _make_objcache(self):
-        if self.objcache == None:
-            if self.objcache_maker:
-                self.objcache = self.objcache_maker()
-            else:
-                self.objcache = PackIdxList(repo('objects/pack'))
-
     def _open(self):
         if not self.file:
-            self._make_objcache()
             (fd,name) = tempfile.mkstemp(suffix='.pack', dir=repo('objects'))
             self.file = os.fdopen(fd, 'w+b')
             assert(name.endswith('.pack'))
             self.filename = name[:-5]
             self.file.write('PACK\0\0\0\2\0\0\0\0')
+            self.idx = list(list() for i in xrange(256))
 
-    def _raw_write(self, datalist):
+    # the 'sha' parameter is used in client.py's _raw_write(), but not needed
+    # in this basic version.
+    def _raw_write(self, datalist, sha):
         self._open()
         f = self.file
         # in case we get interrupted (eg. KeyboardInterrupt), it's best if
@@ -534,15 +546,29 @@ class PackWriter:
         # to our hashsplit algorithm.)  f.write() does its own buffering,
         # but that's okay because we'll flush it in _end().
         oneblob = ''.join(datalist)
-        f.write(oneblob)
-        self.outbytes += len(oneblob)
+        try:
+            f.write(oneblob)
+        except IOError, e:
+            raise GitError, e, sys.exc_info()[2]
+        nw = len(oneblob)
+        crc = zlib.crc32(oneblob) & 0xffffffff
+        self._update_idx(sha, crc, nw)
+        self.outbytes += nw
         self.count += 1
+        return nw, crc
+
+    def _update_idx(self, sha, crc, size):
+        assert(sha)
+        if self.idx:
+            self.idx[ord(sha[0])].append((sha, crc, self.file.tell() - size))
 
-    def _write(self, bin, type, content):
+    def _write(self, sha, type, content):
         if verbose:
             log('>')
-        self._raw_write(_encode_packobj(type, content))
-        return bin
+        if not sha:
+            sha = calc_hash(type, content)
+        size, crc = self._raw_write(_encode_packobj(type, content), sha=sha)
+        return sha
 
     def breakpoint(self):
         """Clear byte and object counts and return the last processed id."""
@@ -554,19 +580,26 @@ class PackWriter:
         """Write an object in this pack file."""
         return self._write(calc_hash(type, content), type, content)
 
+    def _require_objcache(self):
+        if self.objcache is None and self.objcache_maker:
+            self.objcache = self.objcache_maker()
+        if self.objcache is None:
+            raise GitError(
+                    "PackWriter not opened or can't check exists w/o objcache")
+
     def exists(self, id):
         """Return non-empty if an object is found in the object cache."""
-        if not self.objcache:
-            self._make_objcache()
+        self._require_objcache()
         return self.objcache.exists(id)
 
     def maybe_write(self, type, content):
         """Write an object to the pack file if not present and return its id."""
-        bin = calc_hash(type, content)
-        if not self.exists(bin):
-            self._write(bin, type, content)
-            self.objcache.add(bin)
-        return bin
+        self._require_objcache()
+        sha = calc_hash(type, content)
+        if not self.exists(sha):
+            self._write(sha, type, content)
+            self.objcache.add(sha)
+        return sha
 
     def new_blob(self, blob):
         """Create a blob object in the pack with the supplied content."""
@@ -607,15 +640,18 @@ class PackWriter:
         """Remove the pack file from disk."""
         f = self.file
         if f:
+            self.idx = None
             self.file = None
             f.close()
             os.unlink(self.filename + '.pack')
 
-    def _end(self):
+    def _end(self, run_midx=True):
         f = self.file
         if not f: return None
         self.file = None
         self.objcache = None
+        idx = self.idx
+        self.idx = None
 
         # update object count
         f.seek(8)
@@ -626,35 +662,67 @@ class PackWriter:
         # calculate the pack sha1sum
         f.seek(0)
         sum = Sha1()
-        while 1:
-            b = f.read(65536)
+        for b in chunkyreader(f):
             sum.update(b)
-            if not b: break
-        f.write(sum.digest())
-
+        packbin = sum.digest()
+        f.write(packbin)
         f.close()
 
-        p = subprocess.Popen(['git', 'index-pack', '-v',
-                              '--index-version=2',
-                              self.filename + '.pack'],
-                             preexec_fn = _gitenv,
-                             stdout = subprocess.PIPE)
-        out = p.stdout.read().strip()
-        _git_wait('git index-pack', p)
-        if not out:
-            raise GitError('git index-pack produced no output')
-        nameprefix = repo('objects/pack/%s' % out)
+        idx_f = open(self.filename + '.idx', 'wb')
+        obj_list_sha = self._write_pack_idx_v2(idx_f, idx, packbin)
+        idx_f.close()
+
+        nameprefix = repo('objects/pack/pack-%s' % obj_list_sha)
         if os.path.exists(self.filename + '.map'):
             os.unlink(self.filename + '.map')
         os.rename(self.filename + '.pack', nameprefix + '.pack')
         os.rename(self.filename + '.idx', nameprefix + '.idx')
 
-        auto_midx(repo('objects/pack'))
+        if run_midx:
+            auto_midx(repo('objects/pack'))
         return nameprefix
 
-    def close(self):
+    def close(self, run_midx=True):
         """Close the pack file and move it to its definitive path."""
-        return self._end()
+        return self._end(run_midx=run_midx)
+
+    def _write_pack_idx_v2(self, file, idx, packbin):
+        sum = Sha1()
+
+        def write(data):
+            file.write(data)
+            sum.update(data)
+
+        write('\377tOc\0\0\0\2')
+
+        n = 0
+        for part in idx:
+            n += len(part)
+            write(struct.pack('!i', n))
+            part.sort(key=lambda x: x[0])
+
+        obj_list_sum = Sha1()
+        for part in idx:
+            for entry in part:
+                write(entry[0])
+                obj_list_sum.update(entry[0])
+        for part in idx:
+            for entry in part:
+                write(struct.pack('!I', entry[1]))
+        ofs64_list = []
+        for part in idx:
+            for entry in part:
+                if entry[2] & 0x80000000:
+                    write(struct.pack('!I', 0x80000000 | len(ofs64_list)))
+                    ofs64_list.append(struct.pack('!Q', entry[2]))
+                else:
+                    write(struct.pack('!i', entry[2]))
+        for ofs64 in ofs64_list:
+            write(ofs64)
+
+        write(packbin)
+        file.write(sum.digest())
+        return obj_list_sum.hexdigest()
 
 
 def _git_date(date):
@@ -786,7 +854,10 @@ def guess_repo(path=None):
 def init_repo(path=None):
     """Create the Git bare repository for bup in a given path."""
     guess_repo(path)
-    d = repo()
+    d = repo()  # appends a / to the path
+    parent = os.path.dirname(os.path.dirname(d))
+    if parent and not os.path.exists(parent):
+        raise GitError('parent directory "%s" does not exist\n' % parent)
     if os.path.exists(d) and not os.path.isdir(os.path.join(d, '.')):
         raise GitError('"%d" exists but is not a directory\n' % d)
     p = subprocess.Popen(['git', '--bare', 'init'], stdout=sys.stderr,
@@ -925,6 +996,7 @@ class CatPipe:
                                   stdin=subprocess.PIPE,
                                   stdout=subprocess.PIPE,
                                   close_fds = True,
+                                  bufsize = 4096,
                                   preexec_fn = _gitenv)
 
     def _fast_get(self, id):
@@ -941,6 +1013,7 @@ class CatPipe:
         assert(not id.startswith('-'))
         self.inprogress = id
         self.p.stdin.write('%s\n' % id)
+        self.p.stdin.flush()
         hdr = self.p.stdout.readline()
         if hdr.endswith(' missing\n'):
             self.inprogress = None
@@ -1006,3 +1079,16 @@ class CatPipe:
                 yield d
         except StopIteration:
             log('booger!\n')
+
+def tags():
+    """Return a dictionary of all tags in the form {hash: [tag_names, ...]}."""
+    tags = {}
+    for (n,c) in list_refs():
+        if n.startswith('refs/tags/'):
+            name = n[10:]
+            if not c in tags:
+                tags[c] = []
+
+            tags[c].append(name)  # more than one tag can point at 'c'
+
+    return tags