src


Log

Author Commit Date CI Message
Laurence McGlashan b545be3d 2019-09-10T11:14:36 open:fix memory leak when passing NULL to git_repository_open_ext
Edward Thomson c3a7892f 2019-09-09T13:10:24 Merge pull request #5209 from mkostyuk/apply-wrong-patch apply: Fix a patch corruption related to EOFNL handling
Edward Thomson 17d6cd45 2019-09-09T13:06:22 Merge pull request #5210 from buddyspike/master ignore: correct handling of nested rules overriding wild card unignore
Edward Thomson 4d3392dd 2019-09-09T13:03:42 Merge pull request #5214 from pks-t/pks/diff-iterator-allocation-fixes Memory allocation fixes for diff generator
Edward Thomson 39028eb6 2019-09-09T13:00:53 Merge pull request #5212 from libgit2/ethomson/creds_for_scheme Use an HTTP scheme that supports the given credentials
buddyspike a7b4b639 2019-08-24T12:14:31 ignore: correct handling of nested rules overriding wild card unignore problem: filesystem_iterator loads .gitignore files in top-down order. subsequently, ignore module evaluates them in the order they are loaded. this creates a problem if we have unignored a rule (using a wild card) in a sub dir and ignored it again in a level further below (see the test included in this patch). solution: process ignores in reverse order. closes #4963
Edward Thomson 5fc27aac 2019-08-27T13:38:08 Merge pull request #5208 from mkostyuk/apply-removed-new-file apply: git_apply_to_tree fails to apply patches that add new files
Edward Thomson 6de48085 2019-08-27T11:29:24 Merge pull request #5189 from libgit2/ethomson/attrs_from_head Optionally read `.gitattributes` from HEAD
Edward Thomson aaa48d06 2019-08-27T11:26:50 Merge pull request #5196 from pks-t/pks/config-include-onbranch config: implement "onbranch" conditional
Patrick Steinhardt 699de9c5 2019-08-27T10:36:17 iterator: remove duplicate memset When allocating new tree iterator frames, we zero out the allocated memory twice. Remove one of the `memset` calls.
Patrick Steinhardt 9ca7a60e 2019-08-27T10:36:20 iterator: avoid leaving partially initialized frame on stack When allocating tree iterator entries, we use GIT_ERROR_ALLOC_CHECK` to check whether the allocation has failed. The macro will cause the function to immediately return, though, leaving behind a partially initialized iterator frame. Fix the issue by manually checking for memory allocation errors and using `goto done` in case of an error, popping the iterator frame.
Patrick Steinhardt fe241071 2019-08-27T10:36:19 diff_generate: detect memory allocation errors when preparing opts When preparing options for the two iterators that are about to be diffed, we allocate a common prefix for both iterators depending on the options passed by the user. We do not check whether the allocation was successful, though. In fact, this isn't much of a problem, as using a `NULL` prefix is perfectly fine. But in the end, we probably want to detect that the system doesn't have any memory left, as we're unlikely to be able to continue afterwards anyway. While the issue is being fixed in the newly created function `diff_prepare_iterator_opts`, it has been previously existing in the previous macro `DIFF_FROM_ITERATORS` already.
Patrick Steinhardt 8a23597b 2019-08-27T10:36:18 diff_generate: refactor `DIFF_FROM_ITERATORS` macro of doom While the `DIFF_FROM_ITERATORS` does make it shorter to implement the various `git_diff_foo_to_bar` functions, it is a complex and unreadable beast that implicitly assumes certain local variable names. This is not something desirable to have at all and obstructs understanding and more importantly debugging the code by quite a bit. The `DIFF_FROM_ITERATORS` macro basically removed the burden of having to derive the options for both iterators from a pair of iterator flags and the diff options. This patch introduces a new function that does the that exact and refactors all callers to manage the iterators by themselves. As we potentially need to allocate a shared prefix for the iterator, we need to tell the caller to allocate that prefix as soon as the options aren't required anymore. Thus, the function has a `char **prefix` out pointer that will get set to the allocated string and subsequently be free'd by the caller. While this patch increases the line count, I personally deem this to an acceptable tradeoff for increased readbiblity.
Edward Thomson 4e20c7b1 2019-08-25T22:11:39 Merge pull request #5213 from boardwalk/dskorupski/fix_include_case Fix include casing for case-sensitive filesystems.
Dan Skorupski 44d5e47d 2019-08-24T10:39:56 Fix include casing for case-sensitive filesystems.
Ian Hattendorf 4de51f9e 2019-08-23T16:05:28 http: ensure the scheme supports the credentials When a server responds with multiple scheme support - for example, Negotiate and NTLM are commonly used together - we need to ensure that we choose a scheme that supports the credentials.
Edward Thomson 60319788 2019-08-23T09:58:15 Merge pull request #5054 from tniessen/util-use-64-bit-timer util: use 64 bit timer on Windows
Patrick Steinhardt 8cbef12d 2019-08-08T11:52:54 util: do not perform allocations in insertsort Our hand-rolled fallback sorting function `git__insertsort_r` does an in-place sort of the given array. As elements may not necessarily be pointers, it needs a way of swapping two values of arbitrary size, which is currently implemented by allocating a temporary buffer of the element's size. This is problematic, though, as the emulated `qsort` interface doesn't provide any return values and thus cannot signal an error if allocation of that temporary buffer has failed. Convert the function to swap via a temporary buffer allocated on the stack. Like this, it can `memcpy` contents of both elements in small batches without requiring a heap allocation. The buffer size has been chosen such that in most cases, a single iteration of copying will suffice. Most importantly, it can fully contain `git_oid` structures and pointers. Add a bunch of tests for the `git__qsort_r` interface to verify nothing breaks. Furthermore, this removes the declaration of `git__insertsort_r` and makes it static as it is not used anywhere else.
Patrick Steinhardt f3b3e543 2019-08-08T11:34:01 xdiff: catch memory allocation errors The xdiff code contains multiple call sites where the results of `xdl_malloc` are not being checked for memory allocation errors. Add checks to fix possible segfaults due to `NULL` pointer accesses.
Patrick Steinhardt c2dd895a 2019-08-08T10:47:29 transports: http: check for memory allocation failures When allocating a chunk that is used to write to HTTP streams, we do not check for memory allocation errors. This may lead us to write to a `NULL` pointer and thus cause a segfault. Fix this by adding a call to `GIT_ERROR_CHECK_ALLOC`.
Patrick Steinhardt 08699541 2019-08-08T10:46:42 trailer: check for memory allocation errors The "trailer.c" code has been copied mostly verbatim from git.git with minor adjustments, only. As git.git's `xmalloc` function, which aborts on memory allocation errors, has been swapped out for `git_malloc`, which doesn't abort, we may inadvertently access `NULL` pointers. Add checks to fix this.
Patrick Steinhardt 8c7d9761 2019-08-08T10:45:12 posix: fix direct use of `malloc` In "posix.c" there are multiple callsites which execute `malloc` instead of `git__malloc`. Thus, users of library are not able to track these allocations with a custom allocator. Convert these call sites to use `git__malloc` instead.
Patrick Steinhardt a477bff1 2019-08-08T10:44:57 indexer: catch OOM when adding expected OIDs When adding OIDs to the indexer's map of yet-to-be-seen OIDs to verify that packfiles are complete, we do so by first allocating a new OID and then calling `git_oidmap_set` on it. There was no check for memory allocation errors in place, though, leading to possible segfaults due to trying to copy data to a `NULL` pointer. Verify the result of `git__malloc` with `GIT_ERROR_CHECK_ALLOC` to fix the issue.
Patrick Steinhardt d4fe402b 2019-08-08T10:36:33 merge: check return value of `git_commit_list_insert` The function `git_commit_list_insert` dynamically allocates memory and may thus fail to insert a given commit, but we didn't check for that in several places in "merge.c". Convert surrounding functions to return error codes and check whether `git_commit_list_insert` was successful, returning an error if not.
Patrick Steinhardt c0486188 2019-08-08T10:28:09 blame_git: detect memory allocation errors The code in "blame_git.c" was mostly imported from git.git with only minor changes. One of these changes was to use our own allocators instead of git's `xmalloc`, but there's a subtle difference: `xmalloc` would abort the program if unable to allocate any memory, bit `git__malloc` doesn't. As we didn't check for memory allocation errors in some places, we might inadvertently dereference a `NULL` pointer in out-of-memory situations. Convert multiple functions to return proper error codes and add calls to `GIT_ERROR_CHECK_ALLOC` to fix this.
Edward Thomson 1c847169 2019-08-21T16:38:59 http: allow dummy negotiation scheme to fail to act The dummy negotiation scheme is used for known authentication strategies that do not wish to act. For example, when a server requests the "Negotiate" scheme but libgit2 is not built with Negotiate support, and will use the "dummy" strategy which will simply not act. Instead of setting `out` to NULL and returning a successful code, return `GIT_PASSTHROUGH` to indicate that it did not act and catch that error code.
Max Kostyukevich de4bc2bd 2019-08-20T03:29:45 apply: git_apply_to_tree fails to apply patches that add new files git_apply_to_tree() cannot be used apply patches with new files. An attempt to apply such a patch fails because git_apply_to_tree() tries to remove a non-existing file from an old index. The solution is to modify git_apply_to_tree() to git_index_remove() when the patch states that the modified files is removed.
Max Kostyukevich 630127e3 2019-08-20T03:08:32 apply: Fix a patch corruption related to EOFNL handling Use of apply's API can lead to an improper patch application and a corruption of the modified file. The issue is caused by mishandling of the end of file changes if there are several hunks to apply. The new line character is added to a line from a wrong hunk. The solution is to modify apply_hunk() to add the newline character at the end of a line from a right hunk.
Tobias Nießen 071750a3 2019-08-15T14:18:26 cmake: move _WIN32_WINNT definitions to root
Edward Thomson 0f40e68e 2019-08-14T09:05:07 Merge pull request #5187 from ianhattendorf/fix/clone-whitespace clone: don't decode URL percent encodings
Patrick Steinhardt 57a9ccd5 2019-06-21T15:53:54 commit_list: fix possible buffer overflow in `commit_quick_parse` The function `commit_quick_parse` provides a way to quickly parse parts of a commit without storing or verifying most of its metadata. The first thing it does is calculating the number of parents by skipping "parent " lines until it finds the first non-parent line. Afterwards, this parent count is passed to `alloc_parents`, which will allocate an array to store all the parent. To calculate the amount of storage required for the parents array, `alloc_parents` simply multiplicates the number of parents with the respective elements's size. This already screams "buffer overflow", and in fact this problem is getting worse by the result being cast to an `uint32_t`. In fact, triggering this is possible: git-hash-object(1) will happily write a commit with multiple millions of parents for you. I've stopped at 67,108,864 parents as git-hash-object(1) unfortunately soaks up the complete object without streaming anything to disk and thus will cause an OOM situation at a later point. The point here is: this commit was about 4.1GB of size but compressed down to 24MB and thus easy to distribute. The above doesn't yet trigger the buffer overflow, thus. As the array's elements are all pointers which are 8 bytes on 64 bit, we need a total of 536,870,912 parents to trigger the overflow to `0`. The effect is that we're now underallocating the array and do an out-of-bound writes. As the buffer is kindly provided by the adversary, this may easily result in code execution. Extrapolating from the test file with 67m commits to the one with 536m commits results in a factor of 8. Thus the uncompressed contents would be about 32GB in size and the compressed ones 192MB. While still easily distributable via the network, only servers will have that amount of RAM and not cause an out-of-memory condition previous to triggering the overflow. This at least makes this attack not an easy vector for client-side use of libgit2.
Johannes Schindelin cb1439c9 2019-06-19T12:59:27 config: validate ownership of C:\ProgramData\Git\config before using it When the VirtualStore feature is in effect, it is safe to let random users write into C:\ProgramData because other users won't see those files. This seemed to be the case when we introduced support for C:\ProgramData\Git\config. However, when that feature is not in effect (which seems to be the case in newer Windows 10 versions), we'd rather not use those files unless they come from a trusted source, such as an administrator. This change imitates the strategy chosen by PowerShell's native OpenSSH port to Windows regarding host key files: if a system file is owned neither by an administrator, a system account, or the current user, it is ignored.
Edward Thomson 5774b2b1 2019-08-11T23:42:45 Merge pull request #5113 from pks-t/pks/stash-perf stash: avoid recomputing tree when committing worktree
Edward Thomson 42bacbc6 2019-08-11T21:06:19 Merge pull request #5121 from pks-t/pks/variadic-errors Variadic macros
Edward Thomson fba3bf79 2019-07-21T14:15:12 blob: optionally read attributes from repository When `GIT_BLOB_FILTER_ATTTRIBUTES_FROM_HEAD` is passed to `git_blob_filter`, read attributes from `gitattributes` files that are checked in to the repository at the HEAD revision. This passes the flag `GIT_FILTER_ATTRIBUTES_FROM_HEAD` to the filter functions.
Edward Thomson f0f27c1c 2019-07-21T14:13:25 filter: optionally read attributes from repository When `GIT_FILTER_ATTRIBUTES_FROM_HEAD` is specified, configure the filter to read filter attributes from `gitattributes` files that are checked in to the repository at the HEAD revision. This passes the flag `GIT_ATTR_CHECK_INCLUDE_HEAD` to the attribute reading functions.
Edward Thomson 4fd5748c 2019-07-21T14:11:03 attr: optionally read attributes from repository When `GIT_ATTR_CHECK_INCLUDE_HEAD` is specified, read `gitattribute` files that are checked into the repository at the HEAD revision.
Edward Thomson a5392eae 2019-07-21T12:13:07 blob: allow blob filtering to ignore system gitattributes Introduce `GIT_BLOB_FILTER_NO_SYSTEM_ATTRIBUTES`, which tells `git_blob_filter` to ignore the system-wide attributes file, usually `/etc/gitattributes`. This simply passes the appropriate flag to the attribute loading code.
Edward Thomson 22eb12af 2019-07-21T12:12:05 filter: add GIT_FILTER_NO_SYSTEM_ATTRIBUTES option Allow system-wide attributes (the ones specified in `/etc/gitattributes`) to be ignored if the flag `GIT_FILTER_NO_SYSTEM_ATTRIBUTES` is specified.
Edward Thomson fa1a4c77 2019-07-21T11:03:01 blob: deprecate `git_blob_filtered_content` Users should now use `git_blob_filter`.
Edward Thomson a32ab076 2019-07-21T10:56:42 blob: introduce git_blob_filter Provide a function to filter blobs that allows for more functionality than the existing `git_blob_filtered_content` function.
Patrick Steinhardt b0692d6b 2019-08-09T09:01:56 Merge pull request #4913 from implausible/feature/signing-rebase-commits Add sign capability to git_rebase_commit
Tyler Ang-Wanek 998f9c15 2019-08-07T07:21:27 fixup: strange indentation
Edward Thomson f627ba6c 2019-08-02T13:18:07 Merge pull request #5197 from pks-t/pks/remote-ifdeffed-block remote: remove unused block of code
Patrick Steinhardt e23c0b18 2019-08-02T07:52:58 remote: remove unused block of code In "remote.c", we have a chunk of code that is #ifdef'fed out via `#if 0` with a comment that we could export it as a helper function. The code was implemented in 2013 and ifdef'fed in 2014, which shows that there's clearly no interest in having such a helper at all. As this block has recently created some confusion about `p_getenv` due to it containing the only reference to that function in our codebase, let's remove this block altogether.
Patrick Steinhardt d588de7c 2019-08-02T07:51:02 Merge pull request #5191 from eaigner/master config: check if we are running in a sandboxed environment
Erik Aigner 952fbbfb 2019-08-01T20:04:11 config: check if we are running in a sandboxed environment On macOS the $HOME environment variable returns the path to the sandbox container instead of the actual user $HOME for sandboxed apps. To get the correct path, we have to get it from the password file entry.
Patrick Steinhardt 722ba93f 2019-08-01T15:14:06 config: implement "onbranch" conditional With Git v2.23.0, the conditional include mechanism gained another new conditional "onbranch". As the name says, it will cause a file to be included if the "onbranch" pattern matches the currently checked out branch. Implement this new condition and add a bunch of tests.
Patrick Steinhardt 1721ab04 2019-06-16T11:25:47 unix: posix: avoid use of variadic macro `p_snprintf` The macro `p_snprintf` is implemented as a variadic macro that calls `snprintf` directly with `__VA_ARGS__`. In C89, variadic macros are not allowed, but as the arguments of `p_snprintf` and `snprintf` are matching 1:1, we can fix this by simply removing the parameter list from `p_snprintf`.
Patrick Steinhardt 63d8cd18 2019-06-16T11:17:17 apply: remove use of variadic error macro The macro `apply_err` is implemented as a variadic macro, which are not defined by C89. Convert it to a variadic function, instead.
Patrick Steinhardt 27b8b31e 2019-08-01T11:57:03 parse: remove use of variadic macros which are not C89 compliant The macro `git_parse_error` is implemented in a variadic way so that it's possible to pass printf-style parameters. Unfortunately, variadic macros are not defined by C89 and thus we cannot use that functionality. But as we have implemented `git_error_vset` in the previous commit, we can now just use that instead. Convert `git_parse_error` to a variadic function and use `git_error_vset` to fix the compliance violation. While at it, move the function to "patch_parse.c".
Patrick Steinhardt c8e63812 2019-06-16T11:03:08 errors: introduce `git_error_vset` function Right now, we only provide a `git_error_set` that has a variadic function signature. It's impossible to drive this function in a C89-compliant way from other functions that have a variadic signature, though, like for example `git_parse_error`. Implement a new `git_error_vset` function that gets a `va_list` as parameter, fixing the above problem.
Patrick Steinhardt e8f63411 2019-08-01T11:29:58 Merge pull request #5186 from pks-t/pks/config-snapshot-separation config: separate file and snapshot backends
Tobias Nießen fb0730f1 2019-04-16T23:49:16 util: use 64 bit timer on Windows git__timer was originally implemented using a 32 bit timer since Windows XP did not support GetTickCount64. Windows XP was discontinued five years ago, so it should be safe to use the new API. As a benefit, we do not need to care about overflows for the next 585 million years.
Carlos Martín Nieto c8e249b0 2019-07-29T10:51:22 object: deprecate git_object__size for removal In #5118 we remove the double-underscore to make it a normally-named public function. However, this is not an interesting function outside of the library and it takes up a name for something that could be more useful. Remove the single-underscore version as we have not done any releases with it.
Patrick Steinhardt 37ebe9ad 2019-07-24T18:49:08 config_backend: rename internal structures The internal backend structures are kind-of legacy and do not really speak for themselves. Rename them accordingly to make them easier to understand.
Patrick Steinhardt 2bff84ba 2019-07-26T21:02:56 config_file: separate out read-only backend To further distinguish the file writeable and readonly backends, separate the readonly backend into its own "config_snapshot.c" implementation. The snapshot backend can be generically used to snapshot any type of backend.
Patrick Steinhardt f0b10066 2019-07-24T18:37:14 config_file: fix cast of readonly backend In `backend_readonly_free`, the passed in config backend is being cast to a `diskfile_backend` instead of to a `diskfile_readonly_backend`. While this works out just fine because we only access its header values, which were shared between both backends, it is undefined behaviour. Use the correct type to fix this.
Patrick Steinhardt a3159df8 2019-07-24T18:31:43 config_file: remove shared `diskfile_header` struct The `diskfile_header` structure is shared between both `diskfile_backend` and `diskfile_readonly_backend`. The separation and resulting casting is confusing at times and a source for programming errors. Remove the shared structure and inline them directly.
Patrick Steinhardt 271e5fba 2019-07-24T18:18:18 config_file: duplicate accessors for readonly backend While most functions of the readonly configuration backend are implemented separately from the writeable configuration backend, the two functions `config_iterator_new` and `config_get` are shared between both. This sharing makes it necessary to have some shared data structures, which is the `diskfile_header` structure. Unfortunately, this makes the backends harder to grasp than necessary due to all the casting between structs and also quite error prone. Reimplement those functions for the readonly backends. As readonly backends cannot be refreshed anyway, we can remove the calls to `config_refresh` in there.
Patrick Steinhardt 4e7ce1fb 2019-07-24T18:13:52 config_file: reimplement `config_readonly_open` generically The `config_readonly_open` function currently receives as input a diskfile backend and will copy its entries to a new snapshot. This is rather intimate, as we need to assume that the source config backend is in fact a diskfile entry. We can do better than this though by using generic methods to copy contents of the provided backend, e.g. by using a config iterator. This also allows us to decouple the read-only backend from the read-write backend.
Patrick Steinhardt 76182e84 2019-07-24T18:04:38 config_entries: fix possible segfault when duplicating entries When duplicating a configuration entry, we allocate a new entry but do not verify that we get a valid pointer back. As we're dereferencing the pointer afterwards, we might thus run into a segfault in out-of-memory situations. Extract a new function `git_config_entries_dup_entry` that handles the complete entry duplication. Fix the error by using `GIT_ERROR_CHECK_ALLOC`.
Ian Hattendorf ba2885da 2019-07-24T18:05:28 git_net_url_parse: don't git_buf_decode_percent for path
Patrick Steinhardt 2766b92d 2019-07-21T15:10:34 config_file: refresh when creating an iterator When creating a new iterator for a config file backend, then we should always make sure that we're up to date by calling `config_refresh`. Otherwise, we might not notice when another process has modified the configuration file and thus will represent outdated values. Add two tests to config::stress that verify that we get up-to-date values when reading configuration entries via `git_config_iterator`.
Patrick Steinhardt 9fac8b78 2019-07-21T15:08:22 config_file: do not refresh read-only backends If calling `config_refresh` on a read-only configuration file backend, then we will segfault when comparing the timestamp of the file due to `path` being uninitialized. As a read-only snapshot should not be refreshed anyway and stay consistent, we can simply return early when calling `config_refresh` on a read-only snapshot.
Patrick Steinhardt 28d11b59 2019-07-21T14:41:21 config_file: consistently use `GIT_CONTAINER_OF`
Patrick Steinhardt 6be5ac23 2019-07-11T15:30:51 checkout: postpone creation of symlinks to the end On most platforms it's fine to create symlinks to nonexisting files. Not so on Windows, where the type of a symlink (file or directory) needs to be set at creation time. So depending on whether the target file exists or not, we may end up with different symlink types. This creates a problem when performing checkouts, where we simply iterate over all blobs that need to be updated without treating symlinks any special. If the target file of the symlink is going to be checked out after the symlink itself, then the symlink will be created as directory symlink and not as file symlink. Fix the issue by iterating over blobs twice: once to perform postponed deletions and updates to non-symlink blobs, and once to perform updates to symlink blobs.
Patrick Steinhardt 50194dcd 2019-07-11T15:14:42 win32: fix symlinks to relative file targets When creating a symlink in Windows, one needs to tell Windows whether the symlink should be a file or directory symlink. To determine which flag to pass, we call `GetFileAttributesW` on the target file to see whether it is a directory and then pass the flag accordingly. The problem though is if create a symlink with a relative target path, then we will check that relative path while not necessarily being inside of the working directory where the symlink is to be created. Thus, getting its attributes will either fail or return attributes of the wrong target. Fix this by resolving the target path relative to the directory in which the symlink is to be created.
Patrick Steinhardt a00842c4 2019-06-29T09:59:14 win32: correctly unlink symlinks to directories When deleting a symlink on Windows, then the way to delete it depends on whether it is a directory symlink or a file symlink. In the first case, we need to use `DeleteFile`, in the second `RemoveDirectory`. Right now, `p_unlink` will only ever try to use `DeleteFile`, though, and thus fail to remove directory symlinks. This mismatches how unlink(3P) is expected to behave, though, as it shall remove any symlink disregarding whether it is a file or directory symlink. In order to correctly unlink a symlink, we thus need to check what kind of file this is. If we were to first query file attributes of every file upon calling `p_unlink`, then this would penalize the common case though. Instead, we can try to first delete the file with `DeleteFile` and only if the error returned is `ERROR_ACCESS_DENIED` will we query file attributes and determine whether it is a directory symlink to use `RemoveDirectory` instead.
Patrick Steinhardt ded77bb1 2019-06-29T09:58:34 path: extract function to check whether a path supports symlinks When initializing a repository, we need to check whether its working directory supports symlinks to correctly set the initial value of the "core.symlinks" config variable. The code to check the filesystem is reusable in other parts of our codebase, like for example in our tests to determine whether certain tests can be expected to succeed or not. Extract the code into a new function `git_path_supports_symlinks` to avoid duplicate implementations. Remove a duplicate implementation in the repo test helper code.
Patrick Steinhardt e54343a4 2019-06-29T09:17:32 fileops: rename to "futils.h" to match function signatures Our file utils functions all have a "futils" prefix, e.g. `git_futils_touch`. One would thus naturally guess that their definitions and implementation would live in files "futils.h" and "futils.c", respectively, but in fact they live in "fileops.h". Rename the files to match expectations.
Patrick Steinhardt a7d32d60 2019-07-20T18:46:32 stash: avoid recomputing tree when committing worktree When creating a new stash, we need to create there separate commits storing differences stored in the index, untracked changes as well as differences in the working directory. The first two will only be done conditionally if the equivalent options "git stash --keep-index --include-untracked" are being passed to `git_stash_save`, but even when only creating a stash of worktree changes we're much slower than git.git. Using our new stash example: $ time git stash Saved working directory and index state WIP on (no branch): 2f7d9d47575e Linux 5.1.7 real 0m0.528s user 0m0.309s sys 0m0.381s $ time lg2 stash real 0m27.165s user 0m13.645s sys 0m6.403s As can be seen, libgit2 is more than 50x slower than git.git! When creating the stash commit that includes all worktree changes, we create a completely new index to prepare for the new commit and populate it with the entries contained in the index' tree. Here comes the catch: by populating the index with a tree's contents, we do not have any stat caches in the index. This means that we have to re-validate every single file from the worktree and see whether it has changed. The issue can be fixed by populating the new index with the repo's existing index instead of with the tree. This retains all stat cache information, and thus we really only need to check files that have changed stat information. This is semantically equivalent to what we previously did: previously, we used the tree of the commit computed from the index. Now we're just using the index directly. And, in fact, the cache is doing wonders: time lg2 stash real 0m1.836s user 0m1.166s sys 0m0.663s We're now performing 15x faster than before and are only 3x slower than git.git now.
Patrick Steinhardt a613832e 2019-07-20T18:49:48 patch_parse: fix segfault due to line containing static contents With commit dedf70ad2 (patch_parse: do not depend on parsed buffer's lifetime, 2019-07-05), all lines of the patch are allocated with `strdup` to make lifetime of the parsed patch independent of the buffer that is currently being parsed. In patch b08932824 (patch_parse: ensure valid patch output with EOFNL, 2019-07-11), we introduced another code location where we add lines to the parsed patch. But as that one was implemented via a separate pull request, it wasn't converted to use `strdup`, as well. As a consequence, we generate a segfault when trying to deallocate the potentially static buffer that's now in some of the lines. Use `git__strdup` to fix the issue.
Edward Thomson e07dbc92 2019-07-20T11:26:00 Merge pull request #5173 from pks-t/pks/gitignore-wildmatch-error ignore: fix determining whether a shorter pattern negates another
Edward Thomson fd7a384b 2019-07-20T11:24:37 Merge pull request #5159 from pks-t/pks/patch-parse-old-missing-nl patch_parse: handle missing newline indicator in old file
Edward Thomson f33ca472 2019-07-20T11:06:23 Merge pull request #5158 from pks-t/pks/patch-parsed-lifetime patch_parse: do not depend on parsed buffer's lifetime
Edward Thomson d78a1b18 2019-07-20T11:04:53 Merge pull request #5174 from pks-t/pks/winhttp-hash sha1: fix compilation of WinHTTP backend
Edward Thomson 964c1c60 2019-07-20T11:02:30 Merge pull request #5176 from pks-t/pks/repo-template-head repository: do not initialize HEAD if it's provided by templates
Patrick Steinhardt 9d46f167 2019-07-19T10:50:51 repository: do not initialize HEAD if it's provided by templates When using templates to initialize a git repository, then git-init(1) will copy over all contents of the template directory. These will be preferred over the default ones created by git-init(1). While we mostly do the same, there is the exception of "HEAD". While we do copy over the template's HEAD file, afterwards we'll immediately re-initialize its contents with either the default "ref: refs/origin/master" or the init option's `initial_head` field. Let's fix the inconsistency with upstream git-init(1) by not overwriting the template HEAD, but only if the user hasn't set `opts.initial_head`. If the `initial_head` field has been supplied, we should use that indifferent from whether the template contained a HEAD file or not. Add tests to verify we correctly use the template directory's HEAD file and that `initial_head` overrides the template.
Patrick Steinhardt f3134a84 2019-07-19T10:41:10 repository: update error handling in `init_ext` Update `git_repository_init_ext` to use our typical style of error handling. The function had multiple statements which didn't `goto out` immediately but instead deferred it to later calls combined with `if` statements.
Patrick Steinhardt 869ae5a3 2019-07-19T10:15:43 repository: avoid swallowing error codes in `create_head` The error handling in `git_repository_create_head` completely swallows all error codes. While probably not too much of a problem, this also violates our usual coding style. Refactor the code to use a local `error` variable with the typical `goto out` statements.
Patrick Steinhardt 3424c210 2019-07-19T08:00:13 Merge pull request #5138 from libgit2/ethomson/cvar configuration: cvar -> configmap
Patrick Steinhardt a33c0de2 2019-07-18T19:17:40 Merge pull request #5172 from bk2204/cache-efficient-eviction Evict cache items more efficiently
Patrick Steinhardt 658022c4 2019-07-18T13:53:41 configuration: cvar -> configmap `cvar` is an unhelpful name. Refactor its usage to `configmap` for more clarity.
Patrick Steinhardt 343fb83a 2019-07-18T13:50:47 Merge pull request #5156 from pks-t/pks/attr-macros-in-subdir gitattributes: ignore macros defined in subdirectories
Patrick Steinhardt 7574564e 2019-07-18T13:40:34 sha1: win32: fix compilation due to unknown type In commit bbf034ab9 (hash: move `git_hash_prov` into Win32 backend, 2019-02-22), the `git_hash_prov`'s structure name has been removed in favour of its typedef'ed name. But as we have no CI that compiles with the WinHTTPS hashing backend right now, it wasn't noticed that the implementation that uses this struct wasn't changed correctly. Fix the struct type to make it compile again.
Patrick Steinhardt 6f6340af 2019-07-18T11:57:55 ignore: fix determining whether a shorter pattern negates another When computing whether we need to store a negative pattern, we iterate through all previously known patterns and check whether the negative pattern undoes any of the previous ones. In doing so we call `wildmatch` and check it's return for any negative error values. If there was a negative return, we will abort and bubble up that error to the caller. In fact, this check for negative values stems from the time where we still used `fnmatch` instead of `wildmatch`. For `fnmatch`, negative values indicate a "real" error, while for `wildmatch` a negative value may be returned if the matching was prematurely aborted. A premature abort may for example also happen if the pattern matches a prefix of the haystack if the pattern is shorter. Returning an error in that case is the wrong thing to do. Fix the code to compare for equality with `WM_MATCH`, only. Negative values returned by `wildmatch` are perfectly fine and thus should be ignored. Add a test that verifies we do not see the error.
brian m. carlson 770b91b1 2019-07-17T15:59:54 cache: evict items more efficiently When our object cache is full, we pick eight items (or the whole cache, if there are fewer) and evict them. For small cache sizes, this is fine, but when we're dealing with a large number of objects, we can repeatedly exhaust the cache and spend a large amount of time in git_oidmap_iterate trying to find items to evict. Instead, let's assume that if the cache gets full, we have a large number of objects that we're handling, and be more aggressive about evicting items. Let's remove one item for every 2048 items, but not less than 8. This causes us to scale our evictions in proportion to the size of the cache and significantly reduces the time we spend in git_oidmap_iterate. Before this change, a full pack of all the non-blob objects in the Linux repository took in excess of 30 minutes and spent 62.3% of total runtime in odb_read_1 and its children, and 44.3% of the time in git_oidmap_iterate. With this change, the same operation now takes 14 minutes and 44 seconds, and odb_read_1 accounts for only 35.9% of total time, whereas git_oidmap_iterate consists of 6.2%. Note that we do spend a little more time inflating objects and a decent amount more time in memcmp. However, overall, the time taken is significantly improved, and time in pack building is now dominated by git_delta_create_from_index (33.7%), which is what we would expect.
brian m. carlson c4df926b 2019-07-16T21:54:10 pack-objects: allocate memory more efficiently The packbuilder code allocates memory in chunks. When it needs to allocate, it tries to add 1024 to the number of objects and multiply by 3/2. However, it actually multiplies by 1 instead, since it performs an integral division in the expression "3 / 2" and only then multiplies by the increased number of objects. The current behavior causes the code to waste massive amounts of time copying memory when it reallocates, causing inserting all non-blob objects in the Linux repository into a new pack to take some indeterminate time greater than 5 minutes instead of 52 seconds. Correct this error by first dividing by two, and only then multiplying by 3. We still check for overflow for the multiplication, which is the only part that can overflow. This appears to be the only place in the code base which has this problem.
Patrick Steinhardt f8346905 2019-07-12T09:03:33 attr_file: ignore macros defined in subdirectories Right now, we are unconditionally applying all macros found in a gitatttributes file. But quoting gitattributes(5): Custom macro attributes can be defined only in top-level gitattributes files ($GIT_DIR/info/attributes, the .gitattributes file at the top level of the working tree, or the global or system-wide gitattributes files), not in .gitattributes files in working tree subdirectories. The built-in macro attribute "binary" is equivalent to: So gitattribute files in subdirectories of the working tree may explicitly _not_ contain macro definitions, but we do not currently enforce this limitation. This patch introduces a new parameter to the gitattributes parser that tells whether macros are allowed in the current file or not. If set to `false`, we will still parse macros, but silently ignore them instead of adding them to the list of defined macros. Update all callers to correctly determine whether the to-be-parsed file may contain macros or not. Most importantly, when walking up the directory hierarchy, we will only set it to `true` once it reaches the root directory of the repo itself. Add a test that verifies that we are indeed not applying macros from subdirectories. Previous to these changes, the test would've failed.
Patrick Steinhardt 97968529 2019-07-05T08:05:16 attr_file: refactor `parse_buffer` function The gitattributes code is one of our oldest and most-untouched codebases in libgit2, and as such its code style doesn't quite match our current best practices. Refactor the function `git_attr_file__parse_buffer` to better match them.
Patrick Steinhardt dbc7e4b1 2019-07-05T07:53:02 attr_file: refactor `load_standalone` function The gitattributes code is one of our oldest and most-untouched codebases in libgit2, and as such its code style doesn't quite match our current best practices. Refactor the function `git_attr_file__lookup_standalone` to better match them.
Patrick Steinhardt be8f9bb1 2019-07-05T13:33:10 attrcache: fix memory leak if inserting invalid macro to cache A macro without any assignments is considered an invalid macro by the attributes cache and is thus not getting added to the macro map at all. But as `git_attr_cache__insert_macro` returns success with neither free'ing nor adopting the macro into its map, this will cause a memory leak. Fix this by freeing the macro in the function if it's not going to be added. This is perfectly fine to do, as callers assume that the attrcache will have the macro adopted on success anyway.
Patrick Steinhardt 7277bf83 2019-07-05T13:33:05 attrcache: fix multiple memory leaks when inserting macros The function `git_attr_cache__insert_macro` is responsible for adopting macros in the per-repo macro cache. When adding a macro that replaces an already existing macro (e.g. because of re-parsing gitattributes files), then we do not free the previous macro and thus cause a memory leak. Fix this leak by first checking if the cache already has a macro defined with the same name. If so, free it before replacing the cache entry with the new instance.
Patrick Steinhardt 5ae22a63 2019-06-21T08:13:31 fileops: fix creation of directory in filesystem root In commit 45f24e787 (git_repository_init: stop traversing at windows root, 2019-04-12), we have fixed `git_futils_mkdir` to correctly handle the case where we create a directory in Windows-style filesystem roots like "C:\repo". The problem here is an off-by-one: previously, to that commit, we've been checking wether the parent directory's length is equal to the root directory's length incremented by one. When we call the function with "/example", then the parent directory's length ("/") is 1, but the root directory offset is 0 as the path is directly rooted without a drive prefix. This resulted in `1 == 0 + 1`, which was true. With the change, we've stopped incrementing the root directory length, and thus now compare `1 <= 0`, which is false. The previous way of doing it was kind of finicky any non-obvious, which is also why the error was introduced. So instead of just re-adding the increment, let's explicitly add a condition that aborts finding the parent if the current parent path is "/". Making this change causes Azure Pipelines to fail the testcase repo::init::nonexistent_paths on Unix-based systems. This is because we have just fixed creating directories in the filesystem root, which previously didn't work. As Docker-based tests are running as root user, we are thus able to create the non-existing path and will now succeed to create the repository that was expected to actually fail. Let's split this up into three different tests: - A test to verify that we do not create repos in a non-existing parent directoy if the flag `GIT_REPOSITORY_INIT_MKPATH` is not set. - A test to verify that we fail if the root directory does not exist. As there is a common root directory on Unix-based systems that always exist, we can only test for this on Windows-based systems. - A test to verify that we fail if trying to create a repository in an unwriteable parent directory. We can only test this if not running tests as root user, as CAP_DAC_OVERRIDE will cause us to ignore permissions when creating files.
Erik Aigner b0893282 2019-07-11T12:12:04 patch_parse: ensure valid patch output with EOFNL
Patrick Steinhardt 3f855fe8 2019-07-05T11:06:33 patch_parse: handle missing newline indicator in old file When either the old or new file contents have no newline at the end of the file, then git-diff(1) will print out a "\ No newline at end of file" indicator. While we do correctly handle this in the case where the new file has this indcator, we fail to parse patches where the old file is missing a newline at EOF. Fix this bug by handling and missing newline indicators in the old file. Add tests to verify that we can parse such files.
Patrick Steinhardt b30dab8f 2019-07-11T12:10:48 apply: refactor to use a switch statement
Patrick Steinhardt 001d76e1 2019-07-11T11:34:40 diff: ignore EOFNL for computing patch IDs The patch ID is supposed to be mostly context-insignificant and thus only includes added or deleted lines. As such, we shouldn't honor end-of-file-without-newline markers in diffs. Ignore such lines to fix how we compute the patch ID for such diffs.
Patrick Steinhardt dbeadf8a 2019-07-11T10:56:05 config_parse: provide parser init and dispose functions Right now, all configuration file backends are expected to directly mess with the configuration parser's internals in order to set it up. Let's avoid doing that by implementing both a `git_config_parser_init` and `git_config_parser_dispose` function to clearly define the interface between configuration backends and the parser. Ideally, we would make the `git_config_parser` structure definition private to its implementation. But as that would require an additional memory allocation that was not required before we just live with it being visible to others.