OSDN Git Service

drop support for "experimental" loose objects
authorJeff King <peff@peff.net>
Thu, 21 Nov 2013 11:48:37 +0000 (06:48 -0500)
committerJunio C Hamano <gitster@pobox.com>
Thu, 21 Nov 2013 19:43:42 +0000 (11:43 -0800)
In git v1.4.3, we introduced a new loose object format that
encoded some object information outside of the zlib stream.
Ultimately the format was dropped in v1.5.3, but we kept the
reading side around to help people migrate objects. Each
time we open a loose object, we use a heuristic to check
whether it is in the normal loose format, or the
experimental one.

This heuristic is robust in the face of valid data, but it
tends to treat corrupted or garbage data as an experimental
object. With the regular format, we would notice quickly
that zlib's crc does not check out and complain. With the
experimental object, we are likely to extract a nonsensical
object size and try to allocate a huge buffer, resulting in
xmalloc calling "die".

This latter behavior is much worse, for two reasons. One,
git reports an allocation error when the real error is
corruption. And two, the program dies unconditionally, so
you cannot even run fsck (which would otherwise ignore the
broken object and keep going).

We could try to improve the heuristic to err on the side of
normal objects in the face of corruption, but there is
really little point. The experimental format is long-dead,
and was never enabled by default to begin with. We can
instead simply remove it. The only affected repository would
be one that explicitly set core.legacyheaders in 2007, and
then never repacked in the intervening 6 years.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
19 files changed:
sha1_file.c
t/t1013-loose-object-format.sh [deleted file]
t/t1013/objects/14/9cedb5c46929d18e0f118e9fa31927487af3b6 [deleted file]
t/t1013/objects/16/56f9233d999f61ef23ef390b9c71d75399f435 [deleted file]
t/t1013/objects/1e/72a6b2c4a577ab0338860fa9fe87f761fc9bbd [deleted file]
t/t1013/objects/25/7cc5642cb1a054f08cc83f2d943e56fd3ebe99 [deleted file]
t/t1013/objects/2e/65efe2a145dda7ee51d1741299f848e5bf752e [deleted file]
t/t1013/objects/6b/aee0540ea990d9761a3eb9ab183003a71c3696 [deleted file]
t/t1013/objects/70/e6a83d8dcb26fc8bc0cf702e2ddeb6adca18fd [deleted file]
t/t1013/objects/76/e7fa9941f4d5f97f64fea65a2cba436bc79cbb [deleted file]
t/t1013/objects/78/75c6237d3fcdd0ac2f0decc7d3fa6a50b66c09 [deleted file]
t/t1013/objects/7a/37b887a73791d12d26c0d3e39568a8fb0fa6e8 [deleted file]
t/t1013/objects/85/df50785d62d3b05ab03d9cbf7e4a0b49449730 [deleted file]
t/t1013/objects/8d/4e360d6c70fbd72411991c02a09c442cf7a9fa [deleted file]
t/t1013/objects/95/b1625de3ba8b2214d1e0d0591138aea733f64f [deleted file]
t/t1013/objects/9a/e9e86b7bd6cb1472d9373702d8249973da0832 [deleted file]
t/t1013/objects/bd/15045f6ce8ff75747562173640456a394412c8 [deleted file]
t/t1013/objects/e6/9de29bb2d1d6434b8b29ae775ad8c2e48c5391 [deleted file]
t/t1013/objects/f8/16d5255855ac160652ee5253b06cd8ee14165a [deleted file]

index 613839d..406aac4 100644 (file)
@@ -1372,51 +1372,6 @@ void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
        return map;
 }
 
-/*
- * There used to be a second loose object header format which
- * was meant to mimic the in-pack format, allowing for direct
- * copy of the object data.  This format turned up not to be
- * really worth it and we no longer write loose objects in that
- * format.
- */
-static int experimental_loose_object(unsigned char *map)
-{
-       unsigned int word;
-
-       /*
-        * We must determine if the buffer contains the standard
-        * zlib-deflated stream or the experimental format based
-        * on the in-pack object format. Compare the header byte
-        * for each format:
-        *
-        * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
-        * Experimental pack-based : Stttssss : ttt = 1,2,3,4
-        *
-        * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
-        * in standard loose-object format, UNLESS it is a Git-pack
-        * format object *exactly* 8 bytes in size when inflated.
-        *
-        * However, RFC1950 also specifies that the 1st 16-bit word
-        * must be divisible by 31 - this checksum tells us our buffer
-        * is in the standard format, giving a false positive only if
-        * the 1st word of the Git-pack format object happens to be
-        * divisible by 31, ie:
-        *      ((byte0 * 256) + byte1) % 31 = 0
-        *   =>        0ttt10000www1000 % 31 = 0
-        *
-        * As it happens, this case can only arise for www=3 & ttt=1
-        * - ie, a Commit object, which would have to be 8 bytes in
-        * size. As no Commit can be that small, we find that the
-        * combination of these two criteria (bitmask & checksum)
-        * can always correctly determine the buffer format.
-        */
-       word = (map[0] << 8) + map[1];
-       if ((map[0] & 0x8F) == 0x08 && !(word % 31))
-               return 0;
-       else
-               return 1;
-}
-
 unsigned long unpack_object_header_buffer(const unsigned char *buf,
                unsigned long len, enum object_type *type, unsigned long *sizep)
 {
@@ -1444,14 +1399,6 @@ unsigned long unpack_object_header_buffer(const unsigned char *buf,
 
 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
 {
-       unsigned long size, used;
-       static const char valid_loose_object_type[8] = {
-               0, /* OBJ_EXT */
-               1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
-               0, /* "delta" and others are invalid in a loose object */
-       };
-       enum object_type type;
-
        /* Get the data stream */
        memset(stream, 0, sizeof(*stream));
        stream->next_in = map;
@@ -1459,27 +1406,6 @@ int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long ma
        stream->next_out = buffer;
        stream->avail_out = bufsiz;
 
-       if (experimental_loose_object(map)) {
-               /*
-                * The old experimental format we no longer produce;
-                * we can still read it.
-                */
-               used = unpack_object_header_buffer(map, mapsize, &type, &size);
-               if (!used || !valid_loose_object_type[type])
-                       return -1;
-               map += used;
-               mapsize -= used;
-
-               /* Set up the stream for the rest.. */
-               stream->next_in = map;
-               stream->avail_in = mapsize;
-               git_inflate_init(stream);
-
-               /* And generate the fake traditional header */
-               stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
-                                                typename(type), size);
-               return 0;
-       }
        git_inflate_init(stream);
        return git_inflate(stream, 0);
 }
diff --git a/t/t1013-loose-object-format.sh b/t/t1013-loose-object-format.sh
deleted file mode 100755 (executable)
index fbf5f2f..0000000
+++ /dev/null
@@ -1,66 +0,0 @@
-#!/bin/sh
-#
-# Copyright (c) 2011 Roberto Tyley
-#
-
-test_description='Correctly identify and parse loose object headers
-
-There are two file formats for loose objects - the original standard
-format, and the experimental format introduced with Git v1.4.3, later
-deprecated with v1.5.3. Although Git no longer writes the
-experimental format, objects in both formats must be read, with the
-format for a given file being determined by the header.
-
-Detecting file format based on header is not entirely trivial, not
-least because the first byte of a zlib-deflated stream will vary
-depending on how much memory was allocated for the deflation window
-buffer when the object was written out (for example 4KB on Android,
-rather that 32KB on a normal PC).
-
-The loose objects used as test vectors have been generated with the
-following Git versions:
-
-standard format: Git v1.7.4.1
-experimental format: Git v1.4.3 (legacyheaders=false)
-standard format, deflated with 4KB window size: Agit/JGit on Android
-'
-
-. ./test-lib.sh
-
-assert_blob_equals() {
-       printf "%s" "$2" >expected &&
-       git cat-file -p "$1" >actual &&
-       test_cmp expected actual
-}
-
-test_expect_success setup '
-       cp -R "$TEST_DIRECTORY/t1013/objects" .git/ &&
-       git --version
-'
-
-test_expect_success 'read standard-format loose objects' '
-       git cat-file tag 8d4e360d6c70fbd72411991c02a09c442cf7a9fa &&
-       git cat-file commit 6baee0540ea990d9761a3eb9ab183003a71c3696 &&
-       git ls-tree 7a37b887a73791d12d26c0d3e39568a8fb0fa6e8 &&
-       assert_blob_equals "257cc5642cb1a054f08cc83f2d943e56fd3ebe99" "foo$LF"
-'
-
-test_expect_success 'read experimental-format loose objects' '
-       git cat-file tag 76e7fa9941f4d5f97f64fea65a2cba436bc79cbb &&
-       git cat-file commit 7875c6237d3fcdd0ac2f0decc7d3fa6a50b66c09 &&
-       git ls-tree 95b1625de3ba8b2214d1e0d0591138aea733f64f &&
-       assert_blob_equals "2e65efe2a145dda7ee51d1741299f848e5bf752e" "a" &&
-       assert_blob_equals "9ae9e86b7bd6cb1472d9373702d8249973da0832" "ab" &&
-       assert_blob_equals "85df50785d62d3b05ab03d9cbf7e4a0b49449730" "abcd" &&
-       assert_blob_equals "1656f9233d999f61ef23ef390b9c71d75399f435" "abcdefgh" &&
-       assert_blob_equals "1e72a6b2c4a577ab0338860fa9fe87f761fc9bbd" "abcdefghi" &&
-       assert_blob_equals "70e6a83d8dcb26fc8bc0cf702e2ddeb6adca18fd" "abcdefghijklmnop" &&
-       assert_blob_equals "bd15045f6ce8ff75747562173640456a394412c8" "abcdefghijklmnopqrstuvwx"
-'
-
-test_expect_success 'read standard-format objects deflated with smaller window buffer' '
-       git cat-file tag f816d5255855ac160652ee5253b06cd8ee14165a &&
-       git cat-file tag 149cedb5c46929d18e0f118e9fa31927487af3b6
-'
-
-test_done
diff --git a/t/t1013/objects/14/9cedb5c46929d18e0f118e9fa31927487af3b6 b/t/t1013/objects/14/9cedb5c46929d18e0f118e9fa31927487af3b6
deleted file mode 100644 (file)
index 472fd14..0000000
Binary files a/t/t1013/objects/14/9cedb5c46929d18e0f118e9fa31927487af3b6 and /dev/null differ
diff --git a/t/t1013/objects/16/56f9233d999f61ef23ef390b9c71d75399f435 b/t/t1013/objects/16/56f9233d999f61ef23ef390b9c71d75399f435
deleted file mode 100644 (file)
index c379d74..0000000
Binary files a/t/t1013/objects/16/56f9233d999f61ef23ef390b9c71d75399f435 and /dev/null differ
diff --git a/t/t1013/objects/1e/72a6b2c4a577ab0338860fa9fe87f761fc9bbd b/t/t1013/objects/1e/72a6b2c4a577ab0338860fa9fe87f761fc9bbd
deleted file mode 100644 (file)
index 9370630..0000000
Binary files a/t/t1013/objects/1e/72a6b2c4a577ab0338860fa9fe87f761fc9bbd and /dev/null differ
diff --git a/t/t1013/objects/25/7cc5642cb1a054f08cc83f2d943e56fd3ebe99 b/t/t1013/objects/25/7cc5642cb1a054f08cc83f2d943e56fd3ebe99
deleted file mode 100644 (file)
index bdcf704..0000000
Binary files a/t/t1013/objects/25/7cc5642cb1a054f08cc83f2d943e56fd3ebe99 and /dev/null differ
diff --git a/t/t1013/objects/2e/65efe2a145dda7ee51d1741299f848e5bf752e b/t/t1013/objects/2e/65efe2a145dda7ee51d1741299f848e5bf752e
deleted file mode 100644 (file)
index ad62c43..0000000
Binary files a/t/t1013/objects/2e/65efe2a145dda7ee51d1741299f848e5bf752e and /dev/null differ
diff --git a/t/t1013/objects/6b/aee0540ea990d9761a3eb9ab183003a71c3696 b/t/t1013/objects/6b/aee0540ea990d9761a3eb9ab183003a71c3696
deleted file mode 100644 (file)
index 3d2f033..0000000
Binary files a/t/t1013/objects/6b/aee0540ea990d9761a3eb9ab183003a71c3696 and /dev/null differ
diff --git a/t/t1013/objects/70/e6a83d8dcb26fc8bc0cf702e2ddeb6adca18fd b/t/t1013/objects/70/e6a83d8dcb26fc8bc0cf702e2ddeb6adca18fd
deleted file mode 100644 (file)
index b3f71a6..0000000
Binary files a/t/t1013/objects/70/e6a83d8dcb26fc8bc0cf702e2ddeb6adca18fd and /dev/null differ
diff --git a/t/t1013/objects/76/e7fa9941f4d5f97f64fea65a2cba436bc79cbb b/t/t1013/objects/76/e7fa9941f4d5f97f64fea65a2cba436bc79cbb
deleted file mode 100644 (file)
index af4e9a7..0000000
+++ /dev/null
@@ -1,2 +0,0 @@
\vx\9c%ÌA\ e\820\10@Ñ}O1{cSZ(\98\18ãνá\ 2Ãthª\94\92Z\8cÜÞ Ëÿ\16?\r\ f¦\ 2m×6dµi\9d\19É9\85¤Gå\98h\a´Ø¨ÁZR'Q¶\85\81R\8c¡\88\82\1eø³p\ e\91ç\82ÓqL9âÏ=g¸§\81sIÐo\13opÎÿ\94eÏ«_1»\80³¤$×ç\ 5*Si«ëNwpP\95RBôûÅÁú
\87[(ð®d-\8dø\ 2ÁL9á
\ No newline at end of file
diff --git a/t/t1013/objects/78/75c6237d3fcdd0ac2f0decc7d3fa6a50b66c09 b/t/t1013/objects/78/75c6237d3fcdd0ac2f0decc7d3fa6a50b66c09
deleted file mode 100644 (file)
index 3dd28be..0000000
Binary files a/t/t1013/objects/78/75c6237d3fcdd0ac2f0decc7d3fa6a50b66c09 and /dev/null differ
diff --git a/t/t1013/objects/7a/37b887a73791d12d26c0d3e39568a8fb0fa6e8 b/t/t1013/objects/7a/37b887a73791d12d26c0d3e39568a8fb0fa6e8
deleted file mode 100644 (file)
index 2b97b26..0000000
Binary files a/t/t1013/objects/7a/37b887a73791d12d26c0d3e39568a8fb0fa6e8 and /dev/null differ
diff --git a/t/t1013/objects/85/df50785d62d3b05ab03d9cbf7e4a0b49449730 b/t/t1013/objects/85/df50785d62d3b05ab03d9cbf7e4a0b49449730
deleted file mode 100644 (file)
index 6dff746..0000000
Binary files a/t/t1013/objects/85/df50785d62d3b05ab03d9cbf7e4a0b49449730 and /dev/null differ
diff --git a/t/t1013/objects/8d/4e360d6c70fbd72411991c02a09c442cf7a9fa b/t/t1013/objects/8d/4e360d6c70fbd72411991c02a09c442cf7a9fa
deleted file mode 100644 (file)
index cb41e92..0000000
Binary files a/t/t1013/objects/8d/4e360d6c70fbd72411991c02a09c442cf7a9fa and /dev/null differ
diff --git a/t/t1013/objects/95/b1625de3ba8b2214d1e0d0591138aea733f64f b/t/t1013/objects/95/b1625de3ba8b2214d1e0d0591138aea733f64f
deleted file mode 100644 (file)
index 7ac46b4..0000000
Binary files a/t/t1013/objects/95/b1625de3ba8b2214d1e0d0591138aea733f64f and /dev/null differ
diff --git a/t/t1013/objects/9a/e9e86b7bd6cb1472d9373702d8249973da0832 b/t/t1013/objects/9a/e9e86b7bd6cb1472d9373702d8249973da0832
deleted file mode 100644 (file)
index 9d8316d..0000000
Binary files a/t/t1013/objects/9a/e9e86b7bd6cb1472d9373702d8249973da0832 and /dev/null differ
diff --git a/t/t1013/objects/bd/15045f6ce8ff75747562173640456a394412c8 b/t/t1013/objects/bd/15045f6ce8ff75747562173640456a394412c8
deleted file mode 100644 (file)
index eebf239..0000000
Binary files a/t/t1013/objects/bd/15045f6ce8ff75747562173640456a394412c8 and /dev/null differ
diff --git a/t/t1013/objects/e6/9de29bb2d1d6434b8b29ae775ad8c2e48c5391 b/t/t1013/objects/e6/9de29bb2d1d6434b8b29ae775ad8c2e48c5391
deleted file mode 100644 (file)
index 134cf19..0000000
Binary files a/t/t1013/objects/e6/9de29bb2d1d6434b8b29ae775ad8c2e48c5391 and /dev/null differ
diff --git a/t/t1013/objects/f8/16d5255855ac160652ee5253b06cd8ee14165a b/t/t1013/objects/f8/16d5255855ac160652ee5253b06cd8ee14165a
deleted file mode 100644 (file)
index 26b75ae..0000000
+++ /dev/null
@@ -1 +0,0 @@
-H\89\15ÌÁ\ e\820\f\80aÏ{\8aÞ\rI»e\1d&Æø*¥\1d\88\ 1\17ß^¸ýù\ e¿Ë\ 4DåÒ\86wU\87Ò\97¬\1cS±4ª\19\8aÆ\11­ª\9e ,\19\afÅ[ðßVAÛºÎ\1eüxÈÇö6[wtG§Lu\a¸?\97¦²¼Ú×\1f@\89"gì{\86+\12b\b\7fy¾%M
\ No newline at end of file