bf4f0ef840fc8f6f283b62ee167d6211edb575e6 rentzsch jwr.git@redshed.net http://github.com/geoffgarside/cocoagit/commit/fff7b28027f485df1842310b6bb4db70723b2d42 fff7b28027f485df1842310b6bb4db70723b2d42 2009-07-20T03:40:49-07:00 2009-07-16T17:25:47-07:00 [DEV] Make GITRepo+Enumerators.h a public header. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 721d809ad1d46be437c958cf1c2be7b803ad3556 Geoff Garside geoff@geoffgarside.co.uk 7fb78de88518e180672a895698591d26cb5f7ad6 rentzsch jwr.git@redshed.net http://github.com/geoffgarside/cocoagit/commit/bf4f0ef840fc8f6f283b62ee167d6211edb575e6 bf4f0ef840fc8f6f283b62ee167d6211edb575e6 2009-07-20T03:40:48-07:00 2009-07-16T13:19:19-07:00 [NEW] Add -[GITDateTime compare:] for commit-time sortablity. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> c7456667f1b8d3388d1efb78a9d60a9d0f3a8671 Geoff Garside geoff@geoffgarside.co.uk 5f139e017b418e7e45804f85672c121dc17fd8eb Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/7fb78de88518e180672a895698591d26cb5f7ad6 7fb78de88518e180672a895698591d26cb5f7ad6 2009-06-17T05:43:43-07:00 2009-04-30T13:44:22-07:00 Refactor packSHA1 routine to use a lookup table instead of strchr() Also, since this method now operates directly on bytes, create a new method 'packSHA1FromBytes' and call it from 'packSHA1FromString' passing a pointer to the bytes in the string. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> ebf7f53926e3a756cd4e1d9d1458167c6dee9c5b Geoff Garside geoff@geoffgarside.co.uk c45fb8296758c589d436ab94d098725f7f3c3273 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/5f139e017b418e7e45804f85672c121dc17fd8eb 5f139e017b418e7e45804f85672c121dc17fd8eb 2009-06-17T05:43:41-07:00 2009-04-30T13:45:58-07:00 Added testcase for packSHA1FromBytes Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> fa2119e9acc5d04707a5de8ce02d129640ebc118 Geoff Garside geoff@geoffgarside.co.uk 9940ed58a9d3b96f3cc8745d38990fca3a14f863 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/c45fb8296758c589d436ab94d098725f7f3c3273 c45fb8296758c589d436ab94d098725f7f3c3273 2009-06-17T05:43:39-07:00 2009-04-23T15:52:36-07:00 Refactor GITPackReverseIndex to be simpler and faster. Remove the slow, convoluted reverse index creation scheme and replace it with a CFArray of struct pointers that store the offset and the index of the entry in the pack index. This is similar to the git implementation, except we use a CFArray instead of a straight C solution. Note that we need to use a CFArray instead of NSArray because we want to store pointers to structs, not objects. NSArray can only store objects. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 8555160ff4c721788adbe977751bd584ddcadbc5 Geoff Garside geoff@geoffgarside.co.uk 475ae99c0281eb4f21933fe11c93f6b07b70e2a3 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/9940ed58a9d3b96f3cc8745d38990fca3a14f863 9940ed58a9d3b96f3cc8745d38990fca3a14f863 2009-06-17T05:43:38-07:00 2009-04-30T13:32:24-07:00 optimization[2/2]: only parse essential commit info upon initialization Minimal information needed to traverse the graph: + commit-id (sha1) to uniquely identify the commit. + tree-id (sha1) to access the tree and object-ids associated with the commit. + list of parent-id (sha1)s. + raw committer date, ignoring the timezone. For better efficiency, we extract only what we need, and cache the rest in an NSData object for later extraction. Lazily fetched properties are: author + authored + committer + committed + message Currently lazy-fetching is as simple as possible, a call to any of the above methods will cause all of the cached data to be parsed, filling in the rest of the data. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> b1de83a44551db0d8ca3bc7655210a4070d99f5f Geoff Garside geoff@geoffgarside.co.uk fbbd05f0609203f2cfa9defcdda8db507f6a1d53 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/475ae99c0281eb4f21933fe11c93f6b07b70e2a3 475ae99c0281eb4f21933fe11c93f6b07b70e2a3 2009-06-17T05:43:36-07:00 2009-04-30T13:35:29-07:00 Initialize a GITActor from partially processed object header data Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> c794625d52f6809430a19282847daa8aef1abcfe Geoff Garside geoff@geoffgarside.co.uk 5c0ba8c0a748ad0d4255156e9674297a4dccf156 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/fbbd05f0609203f2cfa9defcdda8db507f6a1d53 fbbd05f0609203f2cfa9defcdda8db507f6a1d53 2009-06-17T05:43:33-07:00 2009-04-30T13:39:51-07:00 Add GITDateTime initializaer that accepts a BSD date and (int)timezone offset Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 13d0eb4571e2f531d03e96810e2f3c5144707d2b Geoff Garside geoff@geoffgarside.co.uk 80c98eb0eed1df3dcb34fcb3738c14ee3473320d Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/5c0ba8c0a748ad0d4255156e9674297a4dccf156 5c0ba8c0a748ad0d4255156e9674297a4dccf156 2009-06-17T05:43:32-07:00 2009-04-29T23:16:59-07:00 optimization[1/2]: faster routines for extracting data from core objects When walking the commit graph, parsing essential information from commits can be a potential bottleneck. This is not a big deal for small repositories, or in cases where the entire graph not traversed. However, for large repos such as git or the linux kernel, traversing the graph starting from HEAD takes several seconds. Shark profiling shows that the NSScanner-based GITCommit#parseRawData routine is slow, taking the same or more time than (zlib) inflate. In constrast, zlib inflate is the major limiting step in git-rev-list. This change-set introduces some custom parsing code in an attempt to make this go as fast as possible. The general approach is outlined below, with step 1 implemented in this change-set. 1. Remove NSScanner and drop straight down to raw C. The slow steps seemed to be loading the scanner and allocating NSStrings. We now access raw bytes directly and do not create an NSString unless we need it. To drive the 'parser', define an objectRecord struct for each line or substring that needs to be extracted. In some cases (i.e. sha1 strings) we know the length, so specifying it explicitly saves some scanning. This code is mostly used through a single-method Objective-C interface defined in a "Parsing" category on GITObject. 2. Do _not_ parse the entire commit when the object is created. Minimal nformation needed to traverse the graph: + commit-id (sha1) to uniquely identify the commit. + tree-id (sha1) to access the tree and object-ids associated with the commit. + list of parent-id (sha1)s. + raw committer date, ignoring the timezone. Thus, for better efficiency, we extract only what we need, and cache the rest in an NSData object for later extraction. This means that properties are now fetched lazily: + author + authored + committer + committed + message Currently lazy-fetching is as simple as possible, a call to any of the above methods will cause all of the cached data to be parsed, filling in the rest of the properties and releasing the cached data. This is simple to implement, and the rationale is that long lists of commits/objects are likely to be needed quickly during network transfer or pack file creation. If the information associated with a commit is required, it is likely being displayed on demand for a user who will not need to look at 18000 commits at once. To summarize, it boils down to: Extract structural information required for building a graph as fast as possible. Fill in the rest when it is needed. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 08f347cc9537f313b3e64091ddd88ec280a86671 Geoff Garside geoff@geoffgarside.co.uk 073acf6721b8f747389f22d64dd6a4b6768734a0 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/80c98eb0eed1df3dcb34fcb3738c14ee3473320d 80c98eb0eed1df3dcb34fcb3738c14ee3473320d 2009-06-17T05:43:29-07:00 2009-04-30T22:18:05-07:00 update xcode project Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 3338906a67537d570eb82cba286129733b528a3f Geoff Garside geoff@geoffgarside.co.uk f4c80f2cec1f2248feb8404ede55ae3cd6a436c8 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/073acf6721b8f747389f22d64dd6a4b6768734a0 073acf6721b8f747389f22d64dd6a4b6768734a0 2009-06-17T05:43:27-07:00 2009-04-30T20:36:39-07:00 Update test repo with new branches for testing revision walking and sorting. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 481e4f0571ab0e313ce8e9263546a9b2b93d1d5e Geoff Garside geoff@geoffgarside.co.uk 9b8861ecfbbdb1779fec4ec92ef5e56d8d6e29c6 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/f4c80f2cec1f2248feb8404ede55ae3cd6a436c8 f4c80f2cec1f2248feb8404ede55ae3cd6a436c8 2009-06-17T05:43:23-07:00 2009-04-30T21:52:04-07:00 Add fixture generator and .plist file for GITGraph sort order tests Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 510c48430647a2f7e672e5268d82d3f309800c42 Geoff Garside geoff@geoffgarside.co.uk e6ad9f14b76b0f073355c970d734975c929ab208 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/9b8861ecfbbdb1779fec4ec92ef5e56d8d6e29c6 9b8861ecfbbdb1779fec4ec92ef5e56d8d6e29c6 2009-06-17T05:43:20-07:00 2009-04-30T20:40:10-07:00 Update count of branches in GITRepo test Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 12a95e547c55a2032007acb04f9ab7c87d2a2131 Geoff Garside geoff@geoffgarside.co.uk e8b62cca0574299d47f32a46c5014af17290a758 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/e6ad9f14b76b0f073355c970d734975c929ab208 e6ad9f14b76b0f073355c970d734975c929ab208 2009-06-17T05:43:18-07:00 2009-04-30T20:37:45-07:00 Add tests for GITGraph sort methods. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> a52e2db8e13aeaa14edcbe2e8c66af09d1b4fda5 Geoff Garside geoff@geoffgarside.co.uk 0706a1358153eae8ffeafaca548d050dc32bae22 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/e8b62cca0574299d47f32a46c5014af17290a758 e8b62cca0574299d47f32a46c5014af17290a758 2009-06-17T05:43:16-07:00 2009-04-30T22:06:14-07:00 git-rev-list CLI tool: uses API to reproduce output from 'git rev-list' The tool currently supports output identical[1] to the following commands: git rev-list <commit-id> git rev-list --date-order <commit-id> git rev-list --topo-order <commit-id> Note that listing multiple commits, ranges or excluding commits is not supported, but branch heads can be specified in place of <commit-id>. See use for more info. This tool is exceptionally useful for profiling and comparison with the real git-rev-list. It plays well with Shark and Instruments as long as you setup the command arguments properly (hint: see the "Executables" grouping in Xcode). The '-r' flag allows you to specify any repo, so you can clone out the linux kernel and use that for testing, ect.. [1] This is based on my testing with the git repository, comparing output using HEAD as the commit with the options specified above. This was actually a PITA to sort out, especially since --date-order is almost always identical to the default ordering, except in some rare cases. The test repo now contains a small version of a branch with enough strange date-madness edge-cases that will produce a unique sort ordering in each case. This should be useful for further development of these routines. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> dab9805965a815ccdbde0d23ab1510e619555b48 Geoff Garside geoff@geoffgarside.co.uk ba4c0eb86e0982c95d617802101b7c3a4d26d208 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/0706a1358153eae8ffeafaca548d050dc32bae22 0706a1358153eae8ffeafaca548d050dc32bae22 2009-06-17T05:43:13-07:00 2009-04-30T22:34:24-07:00 Original Nu prototype for graph structure and topological sort rev-list.nu This is superseded by the 'git-rev-list' cli tool, but spared from death for educational value + links to some useful references. set-diff.nu Simple brute force tool to test whether sets of lines from 2 files identical. I was a bit lost when I first started trying to sort out graph traversal and topological ordering. This helped compare output from the real git-rev-list to my incorrect output to at least make sure that I wasn't doing something really stupid. Since we're comparing sets, we don't care about order, only unique content. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 33a8da3ac20f9b2b8810d191c4c796d60566716b Geoff Garside geoff@geoffgarside.co.uk 0745dc596e70518713f7468b745fddbb5643bed4 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/ba4c0eb86e0982c95d617802101b7c3a4d26d208 ba4c0eb86e0982c95d617802101b7c3a4d26d208 2009-06-17T05:43:11-07:00 2009-04-30T22:15:40-07:00 Enumerator subclasses for BFS/DFS traversal of GITCommit objects in a repo. This also includes a GITRepo category that defines the enumerator factor methods. These are currently not used by the graph class, but they will likely be used eventually for some a few revision-walking tasks. Stay tuned. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> d63dde64942eb233a5d8b87071a99f2097e87570 Geoff Garside geoff@geoffgarside.co.uk 097e9a3fd23ffff36a0579d445c65c16f71d6577 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/0745dc596e70518713f7468b745fddbb5643bed4 0745dc596e70518713f7468b745fddbb5643bed4 2009-06-17T05:43:10-07:00 2009-04-30T21:43:33-07:00 Add GITNode class to represent a GITObject in a graph. A GITNode is a simple wrapper class that store a pointer to a GITObject and also caches/stores information related to the graph structure. A GITObject (currently a GITCommit) is required to create a node, since the nodes key is derived from the object's sha1. This key is used to uniquely identify the node in a graph. The idea behind separating the actual objects from the graph representation is: 1. convenience of not added extra properties to GITObject/GITCommit 2. encapsulation of graph storage/traversal algorithms 3. possibility to create thread-safe graph structure for multi-threaded processing. Once constructed, we can still do traversals and sorting operations without the objects. They could be released to save space. In addition, this also makes it possible to create a minimal copy of the graph for multi-threaded operations, should the need arise. On the other hand, it may turn out that the overhead of creating extra objects is a bad idea. I'm also aware that this is a fairly naive way to handle a graph. There are almost certainly better/more efficient ways of storing and accessing the same information. This seems to work for now, but is subject to change should the need arise. Rough estimate of memory overhead for a graph of the git repository starting from HEAD, containing ~18400 commits: graph + nodes (with objects) = 23 MB graph + nodes (objects released) = 8 MB Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> ea320116d588e95d097354dede8f3219ec294f4a Geoff Garside geoff@geoffgarside.co.uk d0d16b13d2d31e02afab18ae1ac8259f6010cbd5 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/097e9a3fd23ffff36a0579d445c65c16f71d6577 097e9a3fd23ffff36a0579d445c65c16f71d6577 2009-06-17T05:43:07-07:00 2009-04-30T20:39:25-07:00 Add helper methods to GITRepo for retrieving commits associated with branch heads Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 0919ba5b8fccbb5e350018bbca3bc0df92550f4d Geoff Garside geoff@geoffgarside.co.uk ab2e783fa442cd73221741ac43338ace3f050a6f Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/d0d16b13d2d31e02afab18ae1ac8259f6010cbd5 d0d16b13d2d31e02afab18ae1ac8259f6010cbd5 2009-06-17T05:43:06-07:00 2009-04-30T20:24:44-07:00 Simple GITGraph class to represent a directed acyclic graph (DAG) The graph class is a wrapper around an NSMutableDictionary of node objects. Each dictionary entry consists of a sha1 key (40-char NSString) and a GITObjectNode value that wraps the corresponding GITObject (GITCommit for now). Because of this arrangement, the graph acts like a standard Foundation collection in the sense that it retains objects that are added, and released objects upon removal. In addition to methods for adding/removing and querying nodes, the graph also supports specialized sort methods that return NSArrays of nodes sorted by date or topology. These lists are identical to the lists of objects obtained from git-rev-list with the default, --topo-order or --date-order flags. The interface currently lacks much documentation, but is relatively straightforward. For use examples, see the unit tests or the source 'git-rev-list' tool. NOTES ----- Each of the 'sort' utilities requires the entire graph to be built and held in memory prior to sorting. I think this is required for the topological sort, but not for the date-based sort. I think it should be possible to abstract out some of the functionality into enumerators or generators that can spit out objects on demand, without requiring as much overhead. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 1d2b23f9f781d7db652728c680292c51cdf179a3 Geoff Garside geoff@geoffgarside.co.uk 98178c0a410a19b86c9339b8c5d8eac8ffd717ad Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/ab2e783fa442cd73221741ac43338ace3f050a6f ab2e783fa442cd73221741ac43338ace3f050a6f 2009-06-17T05:43:04-07:00 2009-04-25T11:41:49-07:00 Fix leaking GITCommit#parentShas. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 5a6fa69ce0ff63a6f491d7f6b1536d254a942e3b Geoff Garside geoff@geoffgarside.co.uk 981b4021248ac9c3b42c291721cf7611c100b0b2 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/98178c0a410a19b86c9339b8c5d8eac8ffd717ad 98178c0a410a19b86c9339b8c5d8eac8ffd717ad 2009-06-17T05:43:02-07:00 2009-04-23T16:39:15-07:00 Fix refStore memleak in GITRepo. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> aaa986a1f1988c413b518056f76324ca2b321711 Geoff Garside geoff@geoffgarside.co.uk 84af0e465440b50a040eaf86aaff6f358c9c12a8 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/981b4021248ac9c3b42c291721cf7611c100b0b2 981b4021248ac9c3b42c291721cf7611c100b0b2 2009-06-17T05:43:01-07:00 2009-04-23T16:40:15-07:00 Fix memleak in symbolic-ref handling. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> ca708c457949d29bc0942f4ee9841537b6fca055 Geoff Garside geoff@geoffgarside.co.uk cdfdb2a8679e9d1eeaab90d3028fe20ec4886dbc Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/84af0e465440b50a040eaf86aaff6f358c9c12a8 84af0e465440b50a040eaf86aaff6f358c9c12a8 2009-06-17T05:42:59-07:00 2009-04-18T16:48:06-07:00 testing ways to optimize GITCommit#parseRawData: Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> a83012758d1577042fe0afa51717a05b4b4da8fb Geoff Garside geoff@geoffgarside.co.uk 6c37c491727fc96d1d7e20ac01a3929959a6d87f Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/cdfdb2a8679e9d1eeaab90d3028fe20ec4886dbc cdfdb2a8679e9d1eeaab90d3028fe20ec4886dbc 2009-06-17T05:42:57-07:00 2009-04-16T22:01:53-07:00 Use NSArray for GITCommit#parents to preverse the order of parents of a commit object. In order to consistantly produce the same topological ordering of a graph of commit objects, the order of the parent objects must be preserved. If the parent order is not preservedr, then commits with muliple parents will result in an arbitrarily chosen topological sort. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> ef9b661b526b129d6862fdb0d4604b55fd26c122 Geoff Garside geoff@geoffgarside.co.uk 352a2d9addef621e2ab46919c3f839240af732ad Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/6c37c491727fc96d1d7e20ac01a3929959a6d87f 6c37c491727fc96d1d7e20ac01a3929959a6d87f 2009-06-17T05:42:52-07:00 2009-04-16T21:37:27-07:00 GITObject#hash calls [[self sha1] hash] to match GITObject#isEqual: The previous implementation was creating an NSScanner to scan [self sha1]. This turned out to be a _massive_ bottleneck during preliminary testing of revision walking routines that iterate over a large set of commit objects. Calling [[self sha1] hash] here is faster by at least an order of magnitude. Signed-off-by: Geoff Garside <geoff@geoffgarside.co.uk> 9541ca1c9b3cb6b31a368bfee166853880671594 Geoff Garside geoff@geoffgarside.co.uk 466f0a1ceee4a97a63eb4ea86d6e3fb640a45da7 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/352a2d9addef621e2ab46919c3f839240af732ad 352a2d9addef621e2ab46919c3f839240af732ad 2009-04-15T12:18:19-07:00 2009-04-15T00:52:15-07:00 Refactor index lookup methods to use a binary search algorithm for sha1 lookups. Start cleaning up the interface by adding -indexOfSha1: method that finds the index of a sha1 in the sha1(v2) or offset+sha1(v1) table and either returns the index or NSNotFound. This is the same concept as NSArray#indexOfObject: The main difference is that right now indexOfSha1: is intended to be a private method, since other classes should not really care about the 'index' in the table. I think index lookups should be handled as follows: Internal methods that require an index as a parameter should throw an exception if the index is out of bounds, since it should never be passed an illegal index unless there is a programming error. The public methods for accessing properties associated with a sha1 (i.e. offset, CRC) just check for an NSNotFound and bail (optionally with an error). Otherwise, the lookup succeeds. If somehow -indexOfSha1: returns a value outside of the index range, something bad has happened, and having the exception will help track down the bug. ed920247fef270cd53e6d4db4dd8e8c3f2a0b94d Brian Chapados chapados@sciencegeeks.org ec75fc7cfce9bc688e450d63873001c1cc9e9019 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/466f0a1ceee4a97a63eb4ea86d6e3fb640a45da7 466f0a1ceee4a97a63eb4ea86d6e3fb640a45da7 2009-04-14T01:47:44-07:00 2009-04-14T01:21:48-07:00 Add test cases and assertions for version 1 pack index files. NOTE: The actual pack file is the same. The only thing that has changed is that the .v1.idx file is a version 1 index for the pack file. The tests make sure that we get the same objects/sha1/offsets from both versions. d751304453f646a61dc6230c94d74f68aa3854b7 Brian Chapados chapados@sciencegeeks.org 933056be6dc18561b0d379581452406e811cb9a0 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/ec75fc7cfce9bc688e450d63873001c1cc9e9019 ec75fc7cfce9bc688e450d63873001c1cc9e9019 2009-04-14T01:47:44-07:00 2009-04-13T22:13:16-07:00 Add sha1/offset lookup & GITPackReverseIndex support to GITPackIndexVersion1 This changeset provides Pack Index V1 implementations of: - (packIndexEntry *) packEntryWithIndex:(NSUInteger)i; - (off_t) packOffsetWithIndex:(NSUInteger)i; - (NSData *)packedSha1WithIndex:(NSUInteger)i; packOffsetWithIndex is used to build the reverse index, which provides: - (off_t)baseOffsetWithOffset:(off_t)offset; - (off_t)nextOffsetWithOffset:(off_t)offset; - (NSString *)sha1WithOffset:(off_t)offset; fb2ea5e183d8433a899adc75daa1a8f80771f6ee Brian Chapados chapados@sciencegeeks.org e37c9f497303dfecee69e769c2e8dbade5256ab5 Brian Chapados chapados@sciencegeeks.org http://github.com/geoffgarside/cocoagit/commit/933056be6dc18561b0d379581452406e811cb9a0 933056be6dc18561b0d379581452406e811cb9a0 2009-04-14T01:47:28-07:00 2009-04-13T22:06:31-07:00 only import GITPackReverseIndex in source file 4e3e5ea0a30541bfeca4aae6b31e7e9ef3562b67 Brian Chapados chapados@sciencegeeks.org