Documentation / RelNotes / 2.7.0.txton commit update-index: fix xgetcwd() related memory leak (c105f56)
   1Git 2.7 Release Notes
   2=====================
   3
   4Updates since v2.6
   5------------------
   6
   7UI, Workflows & Features
   8
   9 * "git remote" learned "get-url" subcommand to show the URL for a
  10   given remote name used for fetching and pushing.
  11
  12 * There was no way to defeat a configured rebase.autostash variable
  13   from the command line, as "git rebase --no-autostash" was missing.
  14
  15 * "git log --date=local" used to only show the normal (default)
  16   format in the local timezone.  The command learned to take 'local'
  17   as an instruction to use the local timezone with other formats,
  18
  19 * The refs used during a "git bisect" session is now per-worktree so
  20   that independent bisect sessions can be done in different worktrees
  21   created with "git worktree add".
  22
  23 * Users who are too busy to type three extra keystrokes to ask for
  24   "git stash show -p" can now set stash.showPatch configuration
  25   varible to true to always see the actual patch, not just the list
  26   of paths affected with feel for the extent of damage via diffstat.
  27
  28 * "quiltimport" allows to specify the series file by honoring the
  29   $QUILT_SERIES environment and also --series command line option.
  30
  31 * The use of 'good/bad' in "git bisect" made it confusing to use when
  32   hunting for a state change that is not a regression (e.g. bugfix).
  33   The command learned 'old/new' and then allows the end user to
  34   say e.g. "bisect start --term-old=fast --term-new=slow" to find a
  35   performance regression.
  36
  37 * "git interpret-trailers" can now run outside of a Git repository.
  38
  39 * "git p4" learned to reencode the pathname it uses to communicate
  40   with the p4 depot with a new option.
  41
  42 * Give progress meter to "git filter-branch".
  43
  44 * Allow a later "!/abc/def" to override an earlier "/abc" that
  45   appears in the same .gitignore file to make it easier to express
  46   "everything in /abc directory is ignored, except for ...".
  47
  48 * Teach "git p4" to send large blobs outside the repository by
  49   talking to Git LFS.
  50
  51 * Prepare for Git on-disk repository representation to undergo
  52   backward incompatible changes by introducing a new repository
  53   format version "1", with an extension mechanism.
  54
  55 * "git worktree" learned a "list" subcommand.
  56
  57 * "git clone --dissociate" learned that it can be used even when
  58   "--reference" was not used at the same time.
  59
  60 * "git blame" learnt to take "--first-parent" and "--reverse" at the
  61   same time when it makes sense.
  62
  63 * "git checkout" did not follow the usual "--[no-]progress"
  64   convention and implemented only "--quiet" that is essentially
  65   a superset of "--no-progress".  Extend the command to support the
  66   usual "--[no-]progress".
  67
  68 * The semantics of tranfer.hideRefs configuration variable have been
  69   extended to work better with the ref "namespace" feature that lets
  70   you throw unrelated bunches of repositories in a single physical
  71   repository and virtually serve them as separate ones.
  72
  73 * send-email config variables whose values are pathnames now go
  74   through the ~username/ expansion.
  75
  76 * bash completion learnt to TAB-complete recipient addresses given
  77   to send-email.
  78
  79 * The credential-cache daemon can be told to ignore SIGHUP to work
  80   around issue when running Git from inside emacs.
  81
  82
  83Performance, Internal Implementation, Development Support etc.
  84
  85 * The infrastructure to rewrite "git submodule" in C is being built
  86   incrementally.  Let's polish these early parts well enough and make
  87   them graduate to 'next' and 'master', so that the more involved
  88   follow-up can start cooking on a solid ground.
  89
  90 * Some features from "git tag -l" and "git branch -l" have been made
  91   available to "git for-each-ref" so that eventually the unified
  92   implementation can be shared across all three.  The version merged
  93   to the 'master' branch earlier had a performance regression in "tag
  94   --contains", which has since been corrected.
  95
  96 * Because "test_when_finished" in our test framework queues the
  97   clean-up tasks to be done in a shell variable, it should not be
  98   used inside a subshell.  Add a mechanism to allow 'bash' to catch
  99   such uses, and fix the ones that were found.
 100
 101 * The debugging infrastructure for pkt-line based communication has
 102   been improved to mark the side-band communication specifically.
 103
 104 * Update "git branch" that list existing branches, using the
 105   ref-filter API that is shared with "git tag" and "git
 106   for-each-ref".
 107
 108 * The test for various line-ending conversions has been enhanced.
 109
 110 * A few test scripts around "git p4" have been improved for
 111   portability.
 112
 113 * Many allocations that is manually counted (correctly) that are
 114   followed by strcpy/sprintf have been replaced with a less error
 115   prone constructs such as xstrfmt.
 116
 117 * The internal stripspace() function has been moved to where it
 118   logically belongs to, i.e. strbuf API, and the command line parser
 119   of "git stripspace" has been updated to use the parse_options API.
 120
 121 * "git am" used to spawn "git mailinfo" via run_command() API once
 122   per each patch, but learned to make a direct call to mailinfo()
 123   instead.
 124
 125 * The implementation of "git mailinfo" was refactored so that a
 126   mailinfo() function can be directly called from inside a process.
 127
 128 * With a "debug" helper, debugging of a single "git" invocation in
 129   our test scripts has become a lot easier.
 130
 131 * The "configure" script did not test for -lpthread correctly, which
 132   upset some linkers.
 133
 134 * Cross completed task off of subtree project's todo list.
 135
 136 * Test cleanups for the subtree project.
 137
 138 * Clean up style in an ancient test t9300.
 139
 140 * Work around some test flakiness with p4d.
 141
 142 * Fsck did not correctly detect a NUL-truncated header in a tag.
 143
 144 * Use a safer behavior when we hit errors verifying remote certificates.
 145
 146 * Speed up filter-branch for cases where we only care about rewriting
 147   commits, not tree data.
 148
 149 * The parse-options API has been updated to make "-h" command line
 150   option work more consistently in all commands.
 151
 152 * "git svn rebase/mkdirs" got optimized by keeping track of empty
 153   directories better.
 154
 155 * Fix some racy client/server tests by treating SIGPIPE the same as a
 156   normal non-zero exit.
 157
 158 * The necessary infrastructure to build topics using the free Travis
 159   CI has been added. Developers forking from this topic (and enabling
 160   Travis) can do their own builds, and we can turn on auto-builds for
 161   git/git (including build-status for pull requests that people
 162   open).
 163
 164
 165Also contains various documentation updates and code clean-ups.
 166
 167
 168Fixes since v2.6
 169----------------
 170
 171Unless otherwise noted, all the fixes since v2.6 in the maintenance
 172track are contained in this release (see the maintenance releases'
 173notes for details).
 174
 175 * Very small number of options take a parameter that is optional
 176   (which is not a great UI element as they can only appear at the end
 177   of the command line).  Add notice to documentation of each and
 178   every one of them.
 179
 180 * "git blame --first-parent v1.0..v2.0" was not rejected but did not
 181   limit the blame to commits on the first parent chain.
 182
 183 * "git subtree" (in contrib/) now can take whitespaces in the
 184   pathnames, not only in the in-tree pathname but the name of the
 185   directory that the repository is in.
 186
 187 * The ssh transport, just like any other transport over the network,
 188   did not clear GIT_* environment variables, but it is possible to
 189   use SendEnv and AcceptEnv to leak them to the remote invocation of
 190   Git, which is not a good idea at all.  Explicitly clear them just
 191   like we do for the local transport.
 192
 193 * Correct "git p4 --detect-labels" so that it does not fail to create
 194   a tag that points at a commit that is also being imported.
 195
 196 * The Makefile always runs the library archiver with hardcoded "crs"
 197   options, which was inconvenient for exotic platforms on which
 198   people want to use programs with totally different set of command
 199   line options.
 200
 201 * Customization to change the behaviour with "make -w" and "make -s"
 202   in our Makefile was broken when they were used together.
 203
 204 * Allocation related functions and stdio are unsafe things to call
 205   inside a signal handler, and indeed killing the pager can cause
 206   glibc to deadlock waiting on allocation mutex as our signal handler
 207   tries to free() some data structures in wait_for_pager().  Reduce
 208   these unsafe calls.
 209
 210 * The way how --ref/--notes to specify the notes tree reference are
 211   DWIMmed was not clearly documented.
 212
 213 * "git gc" used to barf when a symbolic ref has gone dangling
 214   (e.g. the branch that used to be your upstream's default when you
 215   cloned from it is now gone, and you did "fetch --prune").
 216
 217 * "git clone --dissociate" runs a big "git repack" process at the
 218   end, and it helps to close file descriptors that are open on the
 219   packs and their idx files before doing so on filesystems that
 220   cannot remove a file that is still open.
 221
 222 * Description of the "log.follow" configuration variable in "git log"
 223   documentation is now also copied to "git config" documentation.
 224
 225 * "git rebase -i" had a minor regression recently, which stopped
 226   considering a line that begins with an indented '#' in its insn
 227   sheet not a comment. Further, the code was still too picky on
 228   Windows where CRLF left by the editor is turned into a trailing CR
 229   on the line read via the "read" built-in command of bash.  Both of
 230   these issues are now fixed.
 231
 232 * After "git checkout --detach", "git status" reported a fairly
 233   useless "HEAD detached at HEAD", instead of saying at which exact
 234   commit.
 235
 236 * When "git send-email" wanted to talk over Net::SMTP::SSL,
 237   Net::Cmd::datasend() did not like to be fed too many bytes at the
 238   same time and failed to send messages.  Send the payload one line
 239   at a time to work around the problem.
 240
 241 * When "git am" was rewritten as a built-in, it stopped paying
 242   attention to user.signingkey, which was fixed.
 243
 244 * It was not possible to use a repository-lookalike created by "git
 245   worktree add" as a local source of "git clone".
 246
 247 * On a case insensitive filesystems, setting GIT_WORK_TREE variable
 248   using a random cases that does not agree with what the filesystem
 249   thinks confused Git that it wasn't inside the working tree.
 250
 251 * Performance-measurement tests did not work without an installed Git.
 252
 253 * A test script for the HTTP service had a timing dependent bug,
 254   which was fixed.
 255
 256 * There were some classes of errors that "git fsck" diagnosed to its
 257   standard error that did not cause it to exit with non-zero status.
 258
 259 * Work around "git p4" failing when the P4 depot records the contents
 260   in UTF-16 without UTF-16 BOM.
 261
 262 * When "git gc --auto" is backgrounded, its diagnosis message is
 263   lost.  Save it to a file in $GIT_DIR and show it next time the "gc
 264   --auto" is run.
 265
 266 * The submodule code has been taught to work better with separate
 267   work trees created via "git worktree add".
 268
 269 * "git gc" is safe to run anytime only because it has the built-in
 270   grace period to protect young objects.  In order to run with no
 271   grace period, the user must make sure that the repository is
 272   quiescent.
 273
 274 * A recent "filter-branch --msg-filter" broke skipping of the commit
 275   object header, which is fixed.
 276
 277 * The normalize_ceiling_entry() function does not muck with the end
 278   of the path it accepts, and the real world callers do rely on that,
 279   but a test insisted that the function drops a trailing slash.
 280
 281 * A test for interaction between untracked cache and sparse checkout
 282   added in Git 2.5 days were flaky.
 283
 284 * A couple of commands still showed "[options]" in their usage string
 285   to note where options should come on their command line, but we
 286   spell that "[<options>]" in most places these days.
 287
 288 * The synopsis text and the usage string of subcommands that read
 289   list of things from the standard input are often shown as if they
 290   only take input from a file on a filesystem, which was misleading.
 291
 292 * "git am -3" had a small regression where it is aborted in its error
 293   handling codepath when underlying merge-recursive failed in certain
 294   ways, as it assumed that the internal call to merge-recursive will
 295   never die, which is not the case (yet).
 296
 297 * The linkage order of libraries was wrong in places around libcurl.
 298
 299 * The name-hash subsystem that is used to cope with case insensitive
 300   filesystems keeps track of directories and their on-filesystem
 301   cases for all the paths in the index by holding a pointer to a
 302   randomly chosen cache entry that is inside the directory (for its
 303   ce->ce_name component).  This pointer was not updated even when the
 304   cache entry was removed from the index, leading to use after free.
 305   This was fixed by recording the path for each directory instead of
 306   borrowing cache entries and restructuring the API somewhat.
 307
 308 * "git merge-file" tried to signal how many conflicts it found, which
 309   obviously would not work well when there are too many of them.
 310
 311 * The error message from "git blame --contents --reverse" incorrectly
 312   talked about "--contents --children".
 313
 314 * "git imap-send" did not compile well with older version of cURL library.
 315
 316 * Merging a branch that removes a path and another that changes the
 317   mode bits on the same path should have conflicted at the path, but
 318   it didn't and silently favoured the removal.
 319
 320 * "git --literal-pathspecs add -u/-A" without any command line
 321   argument misbehaved ever since Git 2.0.
 322
 323 * "git daemon" uses "run_command()" without "finish_command()", so it
 324   needs to release resources itself, which it forgot to do.
 325
 326 * "git status --branch --short" accessed beyond the constant string
 327   "HEAD", which has been corrected.
 328
 329 * We peek objects from submodule's object store by linking it to the
 330   list of alternate object databases, but the code to do so forgot to
 331   correctly initialize the list.
 332
 333 * The code to prepare the working tree side of temporary directory
 334   for the "dir-diff" feature forgot that symbolic links need not be
 335   copied (or symlinked) to the temporary area, as the code already
 336   special cases and overwrites them.  Besides, it was wrong to try
 337   computing the object name of the target of symbolic link, which may
 338   not even exist or may be a directory.
 339
 340 * A Range: request can be responded with a full response and when
 341   asked properly libcurl knows how to strip the result down to the
 342   requested range.  However, we were hand-crafting a range request
 343   and it did not kick in.
 344
 345 * Having a leftover .idx file without corresponding .pack file in
 346   the repository hurts performance; "git gc" learned to prune them.
 347
 348 * Apple's common crypto implementation of SHA1_Update() does not take
 349   more than 4GB at a time, and we now have a compile-time workaround
 350   for it.
 351
 352 * Produce correct "dirty" marker for shell prompts, even when we
 353   are on an orphan or an unborn branch.
 354
 355 * A build without NO_IPv6 used to use gethostbyname() when guessing
 356   user's hostname, instead of getaddrinfo() that is used in other
 357   codepaths in such a build.
 358
 359 * The exit code of git-fsck did not reflect some types of errors
 360   found in packed objects, which has been corrected.
 361
 362 * The helper used to iterate over loose object directories to prune
 363   stale objects did not closedir() immediately when it is done with a
 364   directory--a callback such as the one used for "git prune" may want
 365   to do rmdir(), but it would fail on open directory on platforms
 366   such as WinXP.
 367
 368 * "git p4" used to import Perforce CLs that touch only paths outside
 369   the client spec as empty commits.  It has been corrected to ignore
 370   them instead, with a new configuration git-p4.keepEmptyCommits as a
 371   backward compatibility knob.
 372
 373 * Code clean-up, minor fixes etc.