]> Kevux Git Server - rit/log
rit
11 years agot9001: refactor sendmail aliases test infrastructure
Eric Sunshine [Sun, 31 May 2015 22:29:30 +0000 (18:29 -0400)]
t9001: refactor sendmail aliases test infrastructure

Several new tests of sendmail aliases parsing will be added in a
subsequent patch, so factor out functionality common to all of them
into a new helper function.

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: implement sendmail aliases line continuation support
Eric Sunshine [Sun, 31 May 2015 22:29:29 +0000 (18:29 -0400)]
send-email: implement sendmail aliases line continuation support

Logical lines in sendmail aliases files can be spread over multiple
physical lines[1]. A line beginning with whitespace is folded into the
preceding line. A line ending with '\' consumes the following line.

[1]: https://www.freebsd.org/cgi/man.cgi?query=aliases&sektion=5

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: simplify sendmail aliases comment and blank line recognizer
Eric Sunshine [Sun, 31 May 2015 22:29:28 +0000 (18:29 -0400)]
send-email: simplify sendmail aliases comment and blank line recognizer

Replace unnecessarily complex regular expression for recognizing comment
and blank lines in sendmail aliases with idiomatic expressions which
can be easily understood at a glance.

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: refactor sendmail aliases parser
Eric Sunshine [Sun, 31 May 2015 22:29:27 +0000 (18:29 -0400)]
send-email: refactor sendmail aliases parser

The sendmail aliases parser inlined into %parse_alias is already
uncomfortably large and is expected to grow as additional functionality
is implemented, so extract it to improve manageability.

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: fix style: cuddle 'elsif' and 'else' with closing brace
Eric Sunshine [Sun, 31 May 2015 22:29:26 +0000 (18:29 -0400)]
send-email: fix style: cuddle 'elsif' and 'else' with closing brace

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: drop noise comments which merely repeat what code says
Eric Sunshine [Sun, 31 May 2015 22:29:25 +0000 (18:29 -0400)]
send-email: drop noise comments which merely repeat what code says

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: visually distinguish sendmail aliases parser warnings
Eric Sunshine [Sun, 31 May 2015 22:29:24 +0000 (18:29 -0400)]
send-email: visually distinguish sendmail aliases parser warnings

Although emitted to stderr, warnings from the sendmail aliases parser
are not visually distinguished as such, and thus can easily be
overlooked in the normal noisy send-email output.

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: further document missing sendmail aliases functionality
Eric Sunshine [Sun, 31 May 2015 22:29:23 +0000 (18:29 -0400)]
send-email: further document missing sendmail aliases functionality

Sendmail aliases[1] supports expansion to a file ("/path/name") or
pipe ("|command"), as well as file inclusion (":include: /path/name"),
however, our implementation does not support such functionality.

[1]: https://www.freebsd.org/cgi/man.cgi?query=aliases&sektion=5

Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoblame: add blame.showEmail configuration
Quentin Neill [Sun, 31 May 2015 19:27:37 +0000 (14:27 -0500)]
blame: add blame.showEmail configuration

Complement existing --show-email option with fallback
configuration variable, with tests.

Signed-off-by: Quentin Neill <quentin.neill@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoSixth batch for 2.5 cycle
Junio C Hamano [Mon, 1 Jun 2015 19:47:56 +0000 (12:47 -0700)]
Sixth batch for 2.5 cycle

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMerge branch 'sb/test-bitmap-free-at-end'
Junio C Hamano [Mon, 1 Jun 2015 19:45:21 +0000 (12:45 -0700)]
Merge branch 'sb/test-bitmap-free-at-end'

An earlier leakfix to bitmap testing code was incomplete.

* sb/test-bitmap-free-at-end:
  test_bitmap_walk: free bitmap with bitmap_free

11 years agoMerge branch 'mm/rebase-i-post-rewrite-exec'
Junio C Hamano [Mon, 1 Jun 2015 19:45:20 +0000 (12:45 -0700)]
Merge branch 'mm/rebase-i-post-rewrite-exec'

"git rebase -i" fired post-rewrite hook when it shouldn't (namely,
when it was told to stop sequencing with 'exec' insn).

* mm/rebase-i-post-rewrite-exec:
  t5407: use <<- to align the expected output
  rebase -i: fix post-rewrite hook with failed exec command
  rebase -i: demonstrate incorrect behavior of post-rewrite

11 years agoMerge branch 'fm/fetch-raw-sha1'
Junio C Hamano [Mon, 1 Jun 2015 19:45:19 +0000 (12:45 -0700)]
Merge branch 'fm/fetch-raw-sha1'

"git upload-pack" that serves "git fetch" can be told to serve
commits that are not at the tip of any ref, as long as they are
reachable from a ref, with uploadpack.allowReachableSHA1InWant
configuration variable.

* fm/fetch-raw-sha1:
  upload-pack: optionally allow fetching reachable sha1
  upload-pack: prepare to extend allow-tip-sha1-in-want
  config.txt: clarify allowTipSHA1InWant with camelCase

11 years agoMerge branch 'sg/help-group'
Junio C Hamano [Mon, 1 Jun 2015 19:45:18 +0000 (12:45 -0700)]
Merge branch 'sg/help-group'

Group list of commands shown by "git help" along the workflow
elements to help early learners.

* sg/help-group:
  help: respect new common command grouping
  command-list.txt: drop the "common" tag
  generate-cmdlist: parse common group commands
  command-list.txt: add the common groups block
  command-list: prepare machinery for upcoming "common groups" section

11 years agoMerge branch 'mm/log-format-raw-doc'
Junio C Hamano [Mon, 1 Jun 2015 19:45:17 +0000 (12:45 -0700)]
Merge branch 'mm/log-format-raw-doc'

Clarify that "log --raw" and "log --format=raw" are unrelated
concepts.

* mm/log-format-raw-doc:
  Documentation/log: clarify sha1 non-abbreviation in log --raw
  Documentation/log: clarify what --raw means

11 years agoMerge branch 'dt/cat-file-follow-symlinks'
Junio C Hamano [Mon, 1 Jun 2015 19:45:16 +0000 (12:45 -0700)]
Merge branch 'dt/cat-file-follow-symlinks'

"git cat-file --batch(-check)" learned the "--follow-symlinks"
option that follows an in-tree symbolic link when asked about an
object via extended SHA-1 syntax, e.g. HEAD:RelNotes that points at
Documentation/RelNotes/2.5.0.txt.  With the new option, the command
behaves as if HEAD:Documentation/RelNotes/2.5.0.txt was given as
input instead.

* dt/cat-file-follow-symlinks:
  cat-file: add --follow-symlinks to --batch
  sha1_name: get_sha1_with_context learns to follow symlinks
  tree-walk: learn get_tree_entry_follow_symlinks

11 years agoMerge branch 'rs/janitorial'
Junio C Hamano [Mon, 1 Jun 2015 19:45:15 +0000 (12:45 -0700)]
Merge branch 'rs/janitorial'

Code clean-up.

* rs/janitorial:
  dir: remove unused variable sb
  clean: remove unused variable buf
  use file_exists() to check if a file exists in the worktree

11 years agoMerge branch 'jk/stash-options'
Junio C Hamano [Mon, 1 Jun 2015 19:45:14 +0000 (12:45 -0700)]
Merge branch 'jk/stash-options'

Make "git stash something --help" error out, so that users can
safely say "git stash drop --help".

* jk/stash-options:
  stash: recognize "--help" for subcommands
  stash: complain about unknown flags

11 years agoMerge branch 'da/mergetool-winmerge'
Junio C Hamano [Mon, 1 Jun 2015 19:45:14 +0000 (12:45 -0700)]
Merge branch 'da/mergetool-winmerge'

"git mergetool" learned to drive WinMerge as a backend.

* da/mergetool-winmerge:
  mergetools: add winmerge as a builtin tool
  mergetool--lib: set IFS for difftool and mergetool

11 years agoMerge branch 'mc/commit-doc-grammofix'
Junio C Hamano [Mon, 1 Jun 2015 19:45:12 +0000 (12:45 -0700)]
Merge branch 'mc/commit-doc-grammofix'

Doc grammar fix.

* mc/commit-doc-grammofix:
  Documentation/git-commit: grammofix

11 years agoMerge branch 'jh/filter-empty-contents'
Junio C Hamano [Mon, 1 Jun 2015 19:45:10 +0000 (12:45 -0700)]
Merge branch 'jh/filter-empty-contents'

The clean/smudge interface did not work well when filtering an
empty contents (failed and then passed the empty input through).
It can be argued that a filter that produces anything but empty for
an empty input is nonsense, but if the user wants to do strange
things, then why not?

* jh/filter-empty-contents:
  sha1_file: pass empty buffer to index empty file

11 years agoMerge branch 'ah/usage-strings'
Junio C Hamano [Mon, 1 Jun 2015 19:45:09 +0000 (12:45 -0700)]
Merge branch 'ah/usage-strings'

A few usage string updates.

* ah/usage-strings:
  blame, log: format usage strings similarly to those in documentation

11 years agoMerge branch 'jk/http-backend-deadlock'
Junio C Hamano [Mon, 1 Jun 2015 19:45:09 +0000 (12:45 -0700)]
Merge branch 'jk/http-backend-deadlock'

Communication between the HTTP server and http_backend process can
lead to a dead-lock when relaying a large ref negotiation request.
Diagnose the situation better, and mitigate it by reading such a
request first into core (to a reasonable limit).

* jk/http-backend-deadlock:
  http-backend: spool ref negotiation requests to buffer
  t5551: factor out tag creation
  http-backend: fix die recursion with custom handler

11 years agoMerge branch 'dt/clean-pathspec-filter-then-lstat'
Junio C Hamano [Mon, 1 Jun 2015 19:45:08 +0000 (12:45 -0700)]
Merge branch 'dt/clean-pathspec-filter-then-lstat'

"git clean pathspec..." tried to lstat(2) and complain even for
paths outside the given pathspec.

* dt/clean-pathspec-filter-then-lstat:
  clean: only lstat files in pathspec

11 years agoapply: reject a hunk that does not do anything
Junio C Hamano [Mon, 1 Jun 2015 19:12:04 +0000 (12:12 -0700)]
apply: reject a hunk that does not do anything

A hunk like this in a hand-edited patch without correctly adjusting
the line counts:

     @@ -660,2 +660,2 @@ inline struct sk_buff *ieee80211_authentic...
             auth = (struct ieee80211_authentication *)
                     skb_put(skb, sizeof(struct ieee80211_authentication));
     -       some old text
     +       some new text
     --
     2.1.0

     dev mailing list

at the end of the input does not have a good way for us to diagnose
it as a corrupt patch.  We just read two context lines and discard
the remainder as cruft, which we must do in order to ignore the
e-mail footer.  Notice that the patch does not change anything and
signal an error.

Note that this fix will not help if the hand-edited hunk header were
"@@ -660,3, +660,2" to include the removal.  We would just remove
the old text without adding the new one, and treat "+ some new text"
and everything after that line as trailing cruft.  So it is dubious
that this patch alone would help very much in practice, but it may
be better than nothing.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosuppress errors on missing UNINTERESTING links
Jeff King [Mon, 1 Jun 2015 09:56:40 +0000 (05:56 -0400)]
suppress errors on missing UNINTERESTING links

When we are traversing commit parents along the
UNINTERESTING side of a revision walk, we do not care if
the parent turns out to be missing. That lets us limit
traversals using unreachable and possibly incomplete
sections of history. However, we do still print error
messages about the missing commits; this patch suppresses
the error, as well.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosilence broken link warnings with revs->ignore_missing_links
Jeff King [Mon, 1 Jun 2015 09:56:37 +0000 (05:56 -0400)]
silence broken link warnings with revs->ignore_missing_links

We set revs->ignore_missing_links to instruct the
revision-walking machinery that we know the history graph
may be incomplete. For example, we use it when walking
unreachable but recent objects; we want to add what we can,
but it's OK if the history is incomplete.

However, we still print error messages for the missing
objects, which can be confusing. This is not an error, but
just a normal situation when transitioning from a repository
last pruned by an older git (which can leave broken segments
of history) to a more recent one (where we try to preserve
whole reachable segments).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoadd quieter versions of parse_{tree,commit}
Jeff King [Mon, 1 Jun 2015 09:56:26 +0000 (05:56 -0400)]
add quieter versions of parse_{tree,commit}

When we call parse_commit, it will complain to stderr if the
object does not exist or cannot be read. This means that we
may produce useless error messages if this situation is
expected (e.g., because the object is marked UNINTERESTING,
or because revs->ignore_missing_links is set).

We can fix this by adding a new "parse_X_gently" form that
takes a flag to suppress the messages. The existing
"parse_X" form is already gentle in the sense that it
returns an error rather than dying, and we could in theory
just add a "quiet" flag to it (with existing callers passing
"0"). But doing it this way means we do not have to disturb
existing callers.

Note also that the new flag is "quiet_on_missing", and not
just "quiet". We could add a flag to suppress _all_ errors,
but besides being a more invasive change (we would have to
pass the flag down to sub-functions, too), there is a good
reason not to: we would never want to use it. Missing a
linked object is expected in some circumstances, but it is
never expected to have a malformed commit, or to get a tree
when we wanted a commit.  We should always complain about
these corruptions.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agocompletion: suggest sequencer commands for revert
Thomas Braun [Mon, 25 May 2015 09:59:35 +0000 (11:59 +0200)]
completion: suggest sequencer commands for revert

Signed-off-by: Thomas Braun <thomas.braun@virtuell-zuhause.de>
Acked-by: Ramkumar Ramachandra <artagnon@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoglossary: add "remote", "submodule", "superproject"
Stefan Beller [Fri, 29 May 2015 18:23:56 +0000 (11:23 -0700)]
glossary: add "remote", "submodule", "superproject"

Noticed-by: Philip Oakley <philipoakley@iee.org>
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosetup_git_directory: delay core.bare/core.worktree errors
Jeff King [Fri, 29 May 2015 06:49:10 +0000 (02:49 -0400)]
setup_git_directory: delay core.bare/core.worktree errors

If both core.bare and core.worktree are set, we complain
about the bogus config and die. Dying is good, because it
avoids commands running and doing damage in a potentially
incorrect setup. But dying _there_ is bad, because it means
that commands which do not even care about the work tree
cannot run. This can make repairing the situation harder:

  [setup]
  $ git config core.bare true
  $ git config core.worktree /some/path

  [OK, expected.]
  $ git status
  fatal: core.bare and core.worktree do not make sense

  [Hrm...]
  $ git config --unset core.worktree
  fatal: core.bare and core.worktree do not make sense

  [Nope...]
  $ git config --edit
  fatal: core.bare and core.worktree do not make sense

  [Gaaah.]
  $ git help config
  fatal: core.bare and core.worktree do not make sense

Instead, let's issue a warning about the bogus config when
we notice it (i.e., for all commands), but only die when the
command tries to use the work tree (by calling setup_work_tree).
So we now get:

  $ git status
  warning: core.bare and core.worktree do not make sense
  fatal: unable to set up work tree using invalid config

  $ git config --unset core.worktree
  warning: core.bare and core.worktree do not make sense

We have to update t1510 to accomodate this; it uses
symbolic-ref to check whether the configuration works or
not, but of course that command does not use the working
tree. Instead, we switch it to use `git status`, as it
requires a work-tree, does not need any special setup, and
is read-only (so a failure will not adversely affect further
tests).

In addition, we add a new test that checks the desired
behavior (i.e., that running "git config" with the bogus
config does in fact work).

Reported-by: SZEDER Gábor <szeder@ira.uka.de>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMakefile: silence perl/PM.stamp recipe
Jeff King [Fri, 29 May 2015 07:26:48 +0000 (03:26 -0400)]
Makefile: silence perl/PM.stamp recipe

Every time we run "make", we update perl/PM.stamp, which
contains a list of all of the perl module files (if it's
updated, we need to rebuild perl/perl.mak, since the
Makefile will not otherwise know about the new files).

This means that every time "make" is run, we see:

      GEN perl/PM.stamp

in the output, even though it is not likely to have changed.
Let's make this recipe completely silent, as we do for other
auto-generated dependency files (e.g., GIT-CFLAGS).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMakefile: avoid timestamp updates to GIT-BUILD-OPTIONS
Jeff King [Fri, 29 May 2015 07:26:30 +0000 (03:26 -0400)]
Makefile: avoid timestamp updates to GIT-BUILD-OPTIONS

We force the GIT-BUILD-OPTIONS recipe to run every time
"make" is invoked. We must do this to catch new options
which may have come from the command-line or environment.

However, we actually update the file's timestamp each time
the recipe is run, whether anything changed or not. As a
result, any files which depend on it (for example, all of
the perl scripts, which need to know whether NO_PERL was
set) will be re-built every time.

Let's do our usual trick of writing to a tempfile, then
doing a "cmp || mv" to update the file only when something
changed.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMakefile: drop dependency between git-instaweb and gitweb
Jeff King [Fri, 29 May 2015 07:25:45 +0000 (03:25 -0400)]
Makefile: drop dependency between git-instaweb and gitweb

The rule for "git-instaweb" depends on "gitweb". This makes
no sense, because:

  1. git-instaweb has no build-time dependency on gitweb; it
     is a run-time dependency

  2. gitweb is a directory that we want to recursively make
     in. As a result, its recipe is marked .PHONY, which
     causes "make" to rebuild git-instaweb every time it is
     run.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot5520: check reflog action in fast-forward merge
Paul Tan [Fri, 29 May 2015 11:44:45 +0000 (19:44 +0800)]
t5520: check reflog action in fast-forward merge

When testing a fast-forward merge with git-pull, check to see if the
reflog action is "pull" with the arguments passed to git-pull.

While we are in the vicinity, remove the empty line as well.

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot5521: test --dry-run does not make any changes
Paul Tan [Fri, 29 May 2015 11:44:44 +0000 (19:44 +0800)]
t5521: test --dry-run does not make any changes

Test that when --dry-run is provided to git-pull, it does not make any
changes, namely:

* --dry-run gets passed to git-fetch, so no FETCH_HEAD will be created
  and no refs will be fetched.

* The index and work tree will not be modified.

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot5520: test --rebase failure on unborn branch with index
Paul Tan [Fri, 29 May 2015 11:44:43 +0000 (19:44 +0800)]
t5520: test --rebase failure on unborn branch with index

Commit 19a7fcb (allow pull --rebase on branch yet to be born,
2009-08-11) special cases git-pull on an unborn branch in a different
code path such that git-pull --rebase is still valid even though there
is no HEAD yet.

This code path still ensures that there is no index in order not to lose
any staged changes. Implement a test to ensure that this check is
triggered.

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot5520: test --rebase with multiple branches
Paul Tan [Fri, 29 May 2015 11:44:42 +0000 (19:44 +0800)]
t5520: test --rebase with multiple branches

Since rebasing on top of multiple upstream branches does not make sense,
since 51b2ead (disallow providing multiple upstream branches to rebase,
pull --rebase, 2009-02-18), git-pull explicitly disallowed specifying
multiple branches in the rebase case.

Implement tests to ensure that git-pull fails and prints out the
user-friendly error message in such a case.

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot5520: test work tree fast-forward when fetch updates head
Paul Tan [Fri, 29 May 2015 11:44:41 +0000 (19:44 +0800)]
t5520: test work tree fast-forward when fetch updates head

Since b10ac50 (Fix pulling into the same branch., 2005-08-25), git-pull,
upon detecting that git-fetch updated the current head, will
fast-forward the working tree to the updated head commit.

Implement tests to ensure that the fast-forward occurs in such a case,
as well as to ensure that the user-friendly advice is printed upon
failure.

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot5520: test for failure if index has unresolved entries
Paul Tan [Fri, 29 May 2015 11:44:40 +0000 (19:44 +0800)]
t5520: test for failure if index has unresolved entries

Commit d38a30d (Be more user-friendly when refusing to do something
because of conflict., 2010-01-12) introduced code paths to git-pull
which will error out with user-friendly advices if the user is in the
middle of a merge or has unmerged files.

Implement tests to ensure that git-pull will not run, and will print
these advices, if the user is in the middle of a merge or has unmerged
files in the index.

Signed-off-by: Paul Tan <pyokagan@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosubmodule doc: reorder introductory paragraphs
Stefan Beller [Wed, 27 May 2015 19:48:01 +0000 (12:48 -0700)]
submodule doc: reorder introductory paragraphs

It's better to start the man page with a description of what
submodules actually are, instead of saying what they are not.

Reorder the paragraphs such that

 - the first short paragraph introduces the submodule concept,
 - the second paragraph highlights the usage of the submodule command,
 - the third paragraph giving background information, and finally
 - the fourth paragraph discusing alternatives such as subtrees and
   remotes, which we don't want to be confused with.

This ordering deepens the knowledge on submodules with each paragraph.
First the basic questions like "How/what" will be answered, while the
underlying concepts will be taught at a later time.

Making sure it is not confused with subtrees and remotes is not really
enhancing knowledge of submodules itself, but rather painting the big
picture of git concepts, so you could also argue to have it as the second
paragraph. Personally I think this may confuse readers, specially
newcomers though.

Additionally to reordering the paragraphs, they have been slightly
reworded.

Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoDocumentation: include 'merge.branchdesc' for merge and config as well
SZEDER Gábor [Wed, 27 May 2015 21:52:23 +0000 (23:52 +0200)]
Documentation: include 'merge.branchdesc' for merge and config as well

'merge.branchdesc' is only mentioned in the docs of 'git fmt-merge-msg'.

The description of 'merge.log' is already duplicated between
'merge-config.txt' and 'git-fmt-merge-msg.txt'; instead of duplicating the
description of another config variable, extract the descriptions of both
of these variables from 'git-fmt-merge-msg.txt' into a separate file and
include it there and in 'merge-config.txt'.

Leave 'merge.summary' only in git-fmt-merge-msg.txt, as it is marked
as deprecated.

Signed-off-by: SZEDER Gábor <szeder@ira.uka.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoxmmap(): drop "Out of memory?"
Junio C Hamano [Wed, 27 May 2015 20:30:29 +0000 (13:30 -0700)]
xmmap(): drop "Out of memory?"

We show that message with die_errno(), but the OS is ought to know
why mmap(2) failed much better than we do.  There is no reason for
us to say "Out of memory?" here.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoconfig.c: rewrite ENODEV into EISDIR when mmap fails
Jeff King [Thu, 28 May 2015 08:03:01 +0000 (04:03 -0400)]
config.c: rewrite ENODEV into EISDIR when mmap fails

If we try to mmap a directory, we'll get ENODEV. This
translates to "no such device" for the user, which is not
very helpful. Since we've just fstat()'d the file, we can
easily check whether the problem was a directory to give a
better message.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoconfig.c: avoid xmmap error messages
Jeff King [Thu, 28 May 2015 07:56:15 +0000 (03:56 -0400)]
config.c: avoid xmmap error messages

The config-writing code uses xmmap to map the existing
config file, which will die if the map fails. This has two
downsides:

  1. The error message is not very helpful, as it lacks any
     context about the file we are mapping:

       $ mkdir foo
       $ git config --file=foo some.key value
       fatal: Out of memory? mmap failed: No such device

  2. We normally do not die in this code path; instead, we'd
     rather report the error and return an appropriate exit
     status (which is part of the public interface
     documented in git-config.1).

This patch introduces a "gentle" form of xmmap which lets us
produce our own error message. We do not want to use mmap
directly, because we would like to use the other
compatibility elements of xmmap (e.g., handling 0-length
maps portably).

The end result is:

    $ git.compile config --file=foo some.key value
    error: unable to mmap 'foo': No such device
    $ echo $?
    3

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoconfig.c: fix mmap leak when writing config
Jeff King [Thu, 28 May 2015 07:54:43 +0000 (03:54 -0400)]
config.c: fix mmap leak when writing config

We mmap the existing config file, but fail to unmap it if we
hit an error. The function already has a shared exit path,
so we can fix this by moving the mmap pointer to the
function scope and clearing it in the shared exit.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoread-cache.c: drop PROT_WRITE from mmap of index
Jeff King [Thu, 28 May 2015 07:54:00 +0000 (03:54 -0400)]
read-cache.c: drop PROT_WRITE from mmap of index

Once upon a time, git's in-memory representation of a cache
entry actually pointed to the mmap'd on-disk data. So in
520fc24 (Allow writing to the private index file mapping.,
2005-04-26), we specified PROT_WRITE so that we could tweak
the entries while we run (in our own MAP_PRIVATE copy-on-write
version, of course).

Later, 7a51ed6 (Make on-disk index representation separate
from in-core one, 2008-01-14) stopped doing this; we copy
the data into our in-core representation, and then drop the
mmap immediately. We can therefore drop the PROT_WRITE flag.
It's probably not hurting anything as it is, but it's
potentially confusing.

Note that we could also mark the mapping as "const" to
verify that we never write to it. However, we don't
typically do that for our other maps, as it then requires
casting to munmap() it.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agop4: retrieve the right revision of the file in UTF-16 codepath
Miguel Torroja [Wed, 27 May 2015 23:14:39 +0000 (01:14 +0200)]
p4: retrieve the right revision of the file in UTF-16 codepath

Fixing bug with UTF-16 files when they are retrieved by git-p4.  It
was always getting the tip version of the file and the history of
the file was lost.

Signed-off-by: Miguel Torroja <miguel.torroja@gmail.com>
Acked-by: Luke Diamand <luke@diamand.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoref_transaction_commit(): do not capitalize error messages
Michael Haggerty [Fri, 22 May 2015 23:34:57 +0000 (01:34 +0200)]
ref_transaction_commit(): do not capitalize error messages

Our convention is for error messages to start with a lower-case
letter.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoverify_lock(): do not capitalize error messages
Michael Haggerty [Fri, 22 May 2015 23:34:56 +0000 (01:34 +0200)]
verify_lock(): do not capitalize error messages

Our convention is for error messages to start with a lower-case
letter.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoverify_lock(): report errors via a strbuf
Michael Haggerty [Fri, 22 May 2015 23:34:55 +0000 (01:34 +0200)]
verify_lock(): report errors via a strbuf

Instead of writing error messages directly to stderr, write them to
a "strbuf *err".  The caller, lock_ref_sha1_basic(), uses this error
reporting convention with all the other callees, and reports its
error this way to its callers.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agodiff.h: rename DIFF_PLAIN color slot to DIFF_CONTEXT
Jeff King [Wed, 27 May 2015 20:48:46 +0000 (16:48 -0400)]
diff.h: rename DIFF_PLAIN color slot to DIFF_CONTEXT

The latter is a much more descriptive name (and we support
"color.diff.context" now). This also updates the name of any
local variables which were used to store the color.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agodiff: accept color.diff.context as a synonym for "plain"
Jeff King [Wed, 27 May 2015 07:22:19 +0000 (03:22 -0400)]
diff: accept color.diff.context as a synonym for "plain"

The term "plain" is a bit ambiguous; let's allow the more
specific "context", but keep "plain" around for
compatibility.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMerge branch 'nd/untracked-cache'
Junio C Hamano [Wed, 27 May 2015 20:14:38 +0000 (13:14 -0700)]
Merge branch 'nd/untracked-cache'

* nd/untracked-cache:
  t7063: hide stderr from setup inside prereq

11 years agot7063: hide stderr from setup inside prereq
Jeff King [Wed, 27 May 2015 09:34:58 +0000 (05:34 -0400)]
t7063: hide stderr from setup inside prereq

When t7063 starts, it runs "update-index --untracked-cache"
to see if we support the untracked cache. Its output goes
straight to stderr, even if the test is not run with "-v".
Let's wrap it in a prereq that will hide the output by
default, but show it with "-v".

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot9001: write $HOME/, not ~/, to help shells without tilde expansion
Junio C Hamano [Sat, 23 May 2015 18:07:50 +0000 (11:07 -0700)]
t9001: write $HOME/, not ~/, to help shells without tilde expansion

Even though it is in POSIX, we do not have to use it, only to hurt
shells that may lack the support.

The .mailrc test tries to define an alias in .mailrc in the home
directory by shell redirection, and then tries to see ~/.mailrc in
config is tilde-expanded by Git without help from shell.  So the
creation should become $HOME/ to be portable for shells that may
lack tilde expansion but the reference should be done as "~/.mailrc".

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend-email: add sendmail email aliases format
Allen Hubbe [Tue, 26 May 2015 21:32:03 +0000 (17:32 -0400)]
send-email: add sendmail email aliases format

Teach send-email to read aliases in the sendmail aliases format, i.e.

<alias>: <address|alias>[, <address|alias>...]

Examples:

alice: Alice W Land <awol@example.com>
bob: Robert Bobbyton <bob@example.com>
# this is a comment
   # this is also a comment
chloe: chloe@example.com
abgroup: alice, bob
bcgrp: bob, chloe, Other <o@example.com>

 - Quoted aliases and quoted addresses are not supported.
 - Line continuations are not supported.

Warnings are printed for explicitly unsupported constructs, and any
other lines that are not matched by the parser.

Signed-off-by: Allen Hubbe <allenbh@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoverify_lock(): on errors, let the caller unlock the lock
Michael Haggerty [Fri, 22 May 2015 23:34:54 +0000 (01:34 +0200)]
verify_lock(): on errors, let the caller unlock the lock

The caller already knows how to do it, so always do it in the same
place.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoverify_lock(): return 0/-1 rather than struct ref_lock *
Michael Haggerty [Fri, 22 May 2015 23:34:53 +0000 (01:34 +0200)]
verify_lock(): return 0/-1 rather than struct ref_lock *

Its return value wasn't conveying any extra information, but it made
the reader wonder whether the ref_lock that it returned might be
different than the one that was passed to it. So change the function
to the traditional "return 0 on success or a negative value on error".

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoclone: reorder --dissociate and --reference options
Jeff King [Thu, 21 May 2015 04:16:04 +0000 (00:16 -0400)]
clone: reorder --dissociate and --reference options

These options are intimately related, so it makes sense to
list them nearby in the "-h" output (they are already
adjacent in the manpage).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoclone: use OPT_STRING_LIST for --reference
Jeff King [Thu, 21 May 2015 04:15:19 +0000 (00:15 -0400)]
clone: use OPT_STRING_LIST for --reference

Not only does this save us having to implement a custom
callback, but it handles "--no-reference" in the usual way
(to clear the list).

The generic callback does copy the string, which we don't
technically need, but that should not hurt anything.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agodiff.c: --ws-error-highlight=<kind> option
Junio C Hamano [Tue, 26 May 2015 17:11:28 +0000 (10:11 -0700)]
diff.c: --ws-error-highlight=<kind> option

Traditionally, we only cared about whitespace breakages introduced
in new lines.  Some people want to paint whitespace breakages on old
lines, too.  When they see a whitespace breakage on a new line, they
can spot the same kind of whitespace breakage on the corresponding
old line and want to say "Ah, those breakages are there but they
were inherited from the original, so let's not touch them for now."

Introduce `--ws-error-highlight=<kind>` option, that lets them pass
a comma separated list of `old`, `new`, and `context` to specify
what lines to highlight whitespace errors on.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agodiff.c: add emit_del_line() and emit_context_line()
Junio C Hamano [Tue, 26 May 2015 16:56:33 +0000 (09:56 -0700)]
diff.c: add emit_del_line() and emit_context_line()

Traditionally, we only had emit_add_line() helper, which knows how
to find and paint whitespace breakages on the given line, because we
only care about whitespace breakages introduced in new lines.  The
context lines and old (i.e. deleted) lines are emitted with a
simpler emit_line_0() that paints the entire line in plain or old
colors.

Identify callers of emit_line_0() that show deleted lines and
context lines, have them call new helpers, emit_del_line() and
emit_context_line(), so that we can later tweak what is done to
these two classes of lines.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoSync with 2.4.2
Junio C Hamano [Tue, 26 May 2015 20:50:39 +0000 (13:50 -0700)]
Sync with 2.4.2

11 years agoGit 2.4.2
Junio C Hamano [Tue, 26 May 2015 20:49:59 +0000 (13:49 -0700)]
Git 2.4.2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMerge branch 'jk/still-interesting' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:26 +0000 (13:49 -0700)]
Merge branch 'jk/still-interesting' into maint

"git rev-list --objects $old --not --all" to see if everything that
is reachable from $old is already connected to the existing refs
was very inefficient.

* jk/still-interesting:
  limit_list: avoid quadratic behavior from still_interesting

11 years agoMerge branch 'jc/hash-object' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:24 +0000 (13:49 -0700)]
Merge branch 'jc/hash-object' into maint

"hash-object --literally" introduced in v2.2 was not prepared to
take a really long object type name.

* jc/hash-object:
  write_sha1_file(): do not use a separate sha1[] array
  t1007: add hash-object --literally tests
  hash-object --literally: fix buffer overrun with extra-long object type
  git-hash-object.txt: document --literally option

11 years agoMerge branch 'jk/rebase-quiet-noop' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:23 +0000 (13:49 -0700)]
Merge branch 'jk/rebase-quiet-noop' into maint

"git rebase --quiet" was not quite quiet when there is nothing to
do.

* jk/rebase-quiet-noop:
  rebase: silence "git checkout" for noop rebase

11 years agoMerge branch 'sg/complete-decorate-full-not-long' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:22 +0000 (13:49 -0700)]
Merge branch 'sg/complete-decorate-full-not-long' into maint

The completion for "log --decorate=" parameter value was incorrect.

* sg/complete-decorate-full-not-long:
  completion: fix and update 'git log --decorate=' options

11 years agoMerge branch 'jk/filter-branch-use-of-sed-on-incomplete-line' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:20 +0000 (13:49 -0700)]
Merge branch 'jk/filter-branch-use-of-sed-on-incomplete-line' into maint

"filter-branch" corrupted commit log message that ends with an
incomplete line on platforms with some "sed" implementations that
munge such a line.  Work it around by avoiding to use "sed".

* jk/filter-branch-use-of-sed-on-incomplete-line:
  filter-branch: avoid passing commit message through sed

11 years agoMerge branch 'jc/daemon-no-ipv6-for-2.4.1' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:19 +0000 (13:49 -0700)]
Merge branch 'jc/daemon-no-ipv6-for-2.4.1' into maint

"git daemon" fails to build from the source under NO_IPV6
configuration (regression in 2.4).

* jc/daemon-no-ipv6-for-2.4.1:
  daemon: unbreak NO_IPV6 build regression

11 years agoMerge branch 'jk/stash-require-clean-index' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:19 +0000 (13:49 -0700)]
Merge branch 'jk/stash-require-clean-index' into maint

"git stash pop/apply" forgot to make sure that not just the working
tree is clean but also the index is clean. The latter is important
as a stash application can conflict and the index will be used for
conflict resolution.

* jk/stash-require-clean-index:
  stash: require a clean index to apply
  t3903: avoid applying onto dirty index
  t3903: stop hard-coding commit sha1s

11 years agoMerge branch 'jk/git-no-more-argv0-path-munging' into maint
Junio C Hamano [Tue, 26 May 2015 20:49:18 +0000 (13:49 -0700)]
Merge branch 'jk/git-no-more-argv0-path-munging' into maint

We have prepended $GIT_EXEC_PATH and the path "git" is installed in
(typically "/usr/bin") to $PATH when invoking subprograms and hooks
for almost eternity, but the original use case the latter tried to
support was semi-bogus (i.e. install git to /opt/foo/git and run it
without having /opt/foo on $PATH), and more importantly it has
become less and less relevant as Git grew more mainstream (i.e. the
users would _want_ to have it on their $PATH).  Stop prepending the
path in which "git" is installed to users' $PATH, as that would
interfere the command search order people depend on (e.g. they may
not like versions of programs that are unrelated to Git in /usr/bin
and want to override them by having different ones in /usr/local/bin
and have the latter directory earlier in their $PATH).

* jk/git-no-more-argv0-path-munging:
  stop putting argv[0] dirname at front of PATH

11 years agoFifth batch for 2.5 cycle
Junio C Hamano [Tue, 26 May 2015 20:33:35 +0000 (13:33 -0700)]
Fifth batch for 2.5 cycle

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMerge branch 'nd/untracked-cache'
Junio C Hamano [Tue, 26 May 2015 20:24:45 +0000 (13:24 -0700)]
Merge branch 'nd/untracked-cache'

Teach the index to optionally remember already seen untracked files
to speed up "git status" in a working tree with tons of cruft.

* nd/untracked-cache: (24 commits)
  git-status.txt: advertisement for untracked cache
  untracked cache: guard and disable on system changes
  mingw32: add uname()
  t7063: tests for untracked cache
  update-index: test the system before enabling untracked cache
  update-index: manually enable or disable untracked cache
  status: enable untracked cache
  untracked-cache: temporarily disable with $GIT_DISABLE_UNTRACKED_CACHE
  untracked cache: mark index dirty if untracked cache is updated
  untracked cache: print stats with $GIT_TRACE_UNTRACKED_STATS
  untracked cache: avoid racy timestamps
  read-cache.c: split racy stat test to a separate function
  untracked cache: invalidate at index addition or removal
  untracked cache: load from UNTR index extension
  untracked cache: save to an index extension
  ewah: add convenient wrapper ewah_serialize_strbuf()
  untracked cache: don't open non-existent .gitignore
  untracked cache: mark what dirs should be recursed/saved
  untracked cache: record/validate dir mtime and reuse cached output
  untracked cache: make a wrapper around {open,read,close}dir()
  ...

11 years agoMerge branch 'rs/plug-leak-in-pack-bitmaps'
Junio C Hamano [Tue, 26 May 2015 20:24:44 +0000 (13:24 -0700)]
Merge branch 'rs/plug-leak-in-pack-bitmaps'

The code to read pack-bitmap wanted to allocate a few hundred
pointers to a structure, but by mistake allocated and leaked memory
enough to hold that many actual structures.  Correct the allocation
size and also have it on stack, as it is small enough.

* rs/plug-leak-in-pack-bitmaps:
  pack-bitmaps: plug memory leak, fix allocation size for recent_bitmaps

11 years agoMerge branch 'pt/pull-ff-vs-merge-ff'
Junio C Hamano [Tue, 26 May 2015 20:24:43 +0000 (13:24 -0700)]
Merge branch 'pt/pull-ff-vs-merge-ff'

The pull.ff configuration was supposed to override the merge.ff
configuration, but it didn't.

* pt/pull-ff-vs-merge-ff:
  pull: parse pull.ff as a bool or string
  pull: make pull.ff=true override merge.ff

11 years agoMerge branch 'pt/pull-log-n'
Junio C Hamano [Tue, 26 May 2015 20:24:43 +0000 (13:24 -0700)]
Merge branch 'pt/pull-log-n'

"git pull --log" and "git pull --no-log" worked as expected, but
"git pull --log=20" did not.

* pt/pull-log-n:
  pull: handle --log=<n>

11 years agoMerge branch 'jk/rerere-forget-check-enabled'
Junio C Hamano [Tue, 26 May 2015 20:24:42 +0000 (13:24 -0700)]
Merge branch 'jk/rerere-forget-check-enabled'

"git rerere forget" in a repository without rerere enabled gave a
cryptic error message; it should be a silent no-op instead.

* jk/rerere-forget-check-enabled:
  rerere: exit silently on "forget" when rerere is disabled

11 years agogit-p4: tests: use test-chmtime in place of touch
Luke Diamand [Tue, 19 May 2015 22:23:18 +0000 (23:23 +0100)]
git-p4: tests: use test-chmtime in place of touch

Using "touch" for P4EDITOR means that the tests can be a bit
racy, since git-p4 checks the timestamp has been updated and
fails if the timestamp is not updated.

Use test-chmtime instead, which is designed for this.

Signed-off-by: Luke Diamand <luke@diamand.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot4015: separate common setup and per-test expectation
Junio C Hamano [Tue, 26 May 2015 18:43:08 +0000 (11:43 -0700)]
t4015: separate common setup and per-test expectation

The last two tests in the script were to

 - set up color.diff.* slots
 - set up an expectation for a single test
 - run that test and check the result

but split in a wrong way.  It did the first two in the first test
and the third one in the second test.  The latter two belong to each
other.  This matters when you plan to add more of these tests that
share the common coloring.

While at it, make sure we use a color different from old, which is
also red.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agot4015: modernise style
Junio C Hamano [Tue, 26 May 2015 18:19:52 +0000 (11:19 -0700)]
t4015: modernise style

Move the preparatory steps that create the expected output inside
the test bodies, remove unnecessary blank lines before and after the
test bodies, and drop SP between redirection operator and its target.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoMerge branch 'jk/http-backend-deadlock-2.3' into jk/http-backend-deadlock
Junio C Hamano [Tue, 26 May 2015 03:44:42 +0000 (20:44 -0700)]
Merge branch 'jk/http-backend-deadlock-2.3' into jk/http-backend-deadlock

* jk/http-backend-deadlock-2.3:
  http-backend: spool ref negotiation requests to buffer
  t5551: factor out tag creation
  http-backend: fix die recursion with custom handler

11 years agoMerge branch 'jk/http-backend-deadlock-2.2' into jk/http-backend-deadlock-2.3
Junio C Hamano [Tue, 26 May 2015 03:44:04 +0000 (20:44 -0700)]
Merge branch 'jk/http-backend-deadlock-2.2' into jk/http-backend-deadlock-2.3

* jk/http-backend-deadlock-2.2:
  http-backend: spool ref negotiation requests to buffer
  t5551: factor out tag creation
  http-backend: fix die recursion with custom handler

11 years agohttp-backend: spool ref negotiation requests to buffer
Jeff King [Wed, 20 May 2015 07:37:09 +0000 (03:37 -0400)]
http-backend: spool ref negotiation requests to buffer

When http-backend spawns "upload-pack" to do ref
negotiation, it streams the http request body to
upload-pack, who then streams the http response back to the
client as it reads. In theory, git can go full-duplex; the
client can consume our response while it is still sending
the request.  In practice, however, HTTP is a half-duplex
protocol. Even if our client is ready to read and write
simultaneously, we may have other HTTP infrastructure in the
way, including the webserver that spawns our CGI, or any
intermediate proxies.

In at least one documented case[1], this leads to deadlock
when trying a fetch over http. What happens is basically:

  1. Apache proxies the request to the CGI, http-backend.

  2. http-backend gzip-inflates the data and sends
     the result to upload-pack.

  3. upload-pack acts on the data and generates output over
     the pipe back to Apache. Apache isn't reading because
     it's busy writing (step 1).

This works fine most of the time, because the upload-pack
output ends up in a system pipe buffer, and Apache reads
it as soon as it finishes writing. But if both the request
and the response exceed the system pipe buffer size, then we
deadlock (Apache blocks writing to http-backend,
http-backend blocks writing to upload-pack, and upload-pack
blocks writing to Apache).

We need to break the deadlock by spooling either the input
or the output. In this case, it's ideal to spool the input,
because Apache does not start reading either stdout _or_
stderr until we have consumed all of the input. So until we
do so, we cannot even get an error message out to the
client.

The solution is fairly straight-forward: we read the request
body into an in-memory buffer in http-backend, freeing up
Apache, and then feed the data ourselves to upload-pack. But
there are a few important things to note:

  1. We limit the in-memory buffer to prevent an obvious
     denial-of-service attack. This is a new hard limit on
     requests, but it's unlikely to come into play. The
     default value is 10MB, which covers even the ridiculous
     100,000-ref negotation in the included test (that
     actually caps out just over 5MB). But it's configurable
     on the off chance that you don't mind spending some
     extra memory to make even ridiculous requests work.

  2. We must take care only to buffer when we have to. For
     pushes, the incoming packfile may be of arbitrary
     size, and we should connect the input directly to
     receive-pack. There's no deadlock problem here, though,
     because we do not produce any output until the whole
     packfile has been read.

     For upload-pack's initial ref advertisement, we
     similarly do not need to buffer. Even though we may
     generate a lot of output, there is no request body at
     all (i.e., it is a GET, not a POST).

[1] http://article.gmane.org/gmane.comp.version-control.git/269020

Test-adapted-from: Dennis Kaarsemaker <dennis@kaarsemaker.net>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agostruct ref_lock: convert old_sha1 member to object_id
Michael Haggerty [Mon, 25 May 2015 18:39:22 +0000 (18:39 +0000)]
struct ref_lock: convert old_sha1 member to object_id

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agowarn_if_dangling_symref(): convert local variable "junk" to object_id
Michael Haggerty [Mon, 25 May 2015 18:39:21 +0000 (18:39 +0000)]
warn_if_dangling_symref(): convert local variable "junk" to object_id

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoeach_ref_fn_adapter(): remove adapter
Michael Haggerty [Mon, 25 May 2015 18:39:20 +0000 (18:39 +0000)]
each_ref_fn_adapter(): remove adapter

All of the callers of the for_each_ref family of functions have now
been rewritten to work with object_ids, so this adapter is no longer
needed.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agorev_list_insert_ref(): remove unneeded arguments
Michael Haggerty [Mon, 25 May 2015 18:39:19 +0000 (18:39 +0000)]
rev_list_insert_ref(): remove unneeded arguments

Now that the function is not being used as an each_ref_sha1_fn, we can
delete the unused arguments in its signature.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agorev_list_insert_ref_oid(): new function, taking an object_oid
Michael Haggerty [Mon, 25 May 2015 18:39:18 +0000 (18:39 +0000)]
rev_list_insert_ref_oid(): new function, taking an object_oid

This function can be used with for_each_ref() without having to be
wrapped.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agomark_complete(): remove unneeded arguments
Michael Haggerty [Mon, 25 May 2015 18:39:17 +0000 (18:39 +0000)]
mark_complete(): remove unneeded arguments

Now that the function is not being used as an each_ref_sha1_fn, we can
delete the unused arguments in its signature.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agomark_complete_oid(): new function, taking an object_oid
Michael Haggerty [Mon, 25 May 2015 18:39:16 +0000 (18:39 +0000)]
mark_complete_oid(): new function, taking an object_oid

This function can be used with for_each_ref() without having to be
wrapped.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoclear_marks(): rewrite to take an object_id argument
Michael Haggerty [Mon, 25 May 2015 18:39:15 +0000 (18:39 +0000)]
clear_marks(): rewrite to take an object_id argument

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agomark_complete(): rewrite to take an object_id argument
Michael Haggerty [Mon, 25 May 2015 18:39:14 +0000 (18:39 +0000)]
mark_complete(): rewrite to take an object_id argument

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agosend_ref(): convert local variable "peeled" to object_id
Michael Haggerty [Mon, 25 May 2015 18:39:13 +0000 (18:39 +0000)]
send_ref(): convert local variable "peeled" to object_id

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agoupload-pack: rewrite functions to take object_id arguments
Michael Haggerty [Mon, 25 May 2015 18:39:12 +0000 (18:39 +0000)]
upload-pack: rewrite functions to take object_id arguments

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agofind_symref(): convert local variable "unused" to object_id
Michael Haggerty [Mon, 25 May 2015 18:39:11 +0000 (18:39 +0000)]
find_symref(): convert local variable "unused" to object_id

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agofind_symref(): rewrite to take an object_id argument
Michael Haggerty [Mon, 25 May 2015 18:39:10 +0000 (18:39 +0000)]
find_symref(): rewrite to take an object_id argument

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agowrite_one_ref(): rewrite to take an object_id argument
Michael Haggerty [Mon, 25 May 2015 18:39:09 +0000 (18:39 +0000)]
write_one_ref(): rewrite to take an object_id argument

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
11 years agowrite_refs_to_temp_dir(): convert local variable sha1 to object_id
Michael Haggerty [Mon, 25 May 2015 18:39:08 +0000 (18:39 +0000)]
write_refs_to_temp_dir(): convert local variable sha1 to object_id

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>