When git invokes a pager that exits with non-zero the common case is
that we'll already return the correct SIGPIPE failure from git itself,
but the exit code logged in trace2 has always been incorrectly
reported[1]. Fix that and log the correct exit code in the logs.
Since this gives us something to test outside of our recently-added
tests needing a !MINGW prerequisite, let's refactor the test to run on
MINGW and actually check for SIGPIPE outside of MINGW.
The wait_or_whine() is only called with a true "in_signal" from from
finish_command_in_signal(), which in turn is only used in pager.c.
The "in_signal && !WIFEXITED(status)" case is not covered by
tests. Let's log the default -1 in that case for good measure.
1. The incorrect logging of the exit code in was seemingly copy/pasted
into finish_command_in_signal() in ee4512ed48 (trace2: create new
combined trace facility, 2019-02-22)
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add braces to an "if" block in the wait_or_whine() function. This
isn't needed now, but will make a subsequent commit easier to read.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Some commands run 'git maintenance run --auto --[no-]quiet' after doing
their normal work, as a way to keep repositories clean as they are used.
Currently, users who do not want this maintenance to occur would set the
'gc.auto' config option to 0 to avoid the 'gc' task from running.
However, this does not stop the extra process invocation. On Windows,
this extra process invocation can be more expensive than necessary.
Allow users to drop this extra process by setting 'maintenance.auto' to
'false'.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The run_auto_gc() method is used in several places to trigger a check
for repo maintenance after some Git commands, such as 'git commit' or
'git fetch'.
To allow for extra customization of this maintenance activity, replace
the 'git gc --auto [--quiet]' call with one to 'git maintenance run
--auto [--quiet]'. As we extend the maintenance builtin with other
steps, users will be able to select different maintenance activities.
Rename run_auto_gc() to run_auto_maintenance() to be clearer what is
happening on this call, and to expose all callers in the current diff.
Rewrite the method to use a struct child_process to simplify the calls
slightly.
Since 'git fetch' already allows disabling the 'git gc --auto'
subprocess, add an equivalent option with a different name to be more
descriptive of the new behavior: '--[no-]maintenance'. Update the
documentation to include these options at the same time.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "argc" and "argv" names made sense when the struct was argv_array,
but now they're just confusing. Let's rename them to "nr" (which we use
for counts elsewhere) and "v" (which is rather terse, but reads well
when combined with typical variable names like "args.v").
Note that we have to update all of the callers immediately. Playing
tricks with the preprocessor is hard here, because we wouldn't want to
rewrite unrelated tokens.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We eventually want to drop the argv_array name and just use strvec
consistently. There's no particular reason we have to do it all at once,
or care about interactions between converted and unconverted bits.
Because of our preprocessor compat layer, the names are interchangeable
to the compiler (so even a definition and declaration using different
names is OK).
This patch converts all of the remaining files, as the resulting diff is
reasonably sized.
The conversion was done purely mechanically with:
git ls-files '*.c' '*.h' |
xargs perl -i -pe '
s/ARGV_ARRAY/STRVEC/g;
s/argv_array/strvec/g;
'
We'll deal with any indentation/style fallouts separately.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This requires updating #include lines across the code-base, but that's
all fairly mechanical, and was done with:
git ls-files '*.c' '*.h' |
xargs perl -i -pe 's/argv-array.h/strvec.h/'
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When an aliased command, whose output is piped to a pager by git,
gets killed by a signal, the pager got into a funny state, which
has been corrected (again).
* ta/wait-on-aliased-commands-upon-signal:
Wait for child on signal death for aliases to externals
Wait for child on signal death for aliases to builtins
When you hit ^C all the processes in the tree receives it. When a git
command uses a pager, git ignores this and waits until the pager quits.
However, when using an alias there is an additional process in the tree
which didn't ignore the signal. That caused it to exit which in turn
caused the pager to exit. This fixes that for aliases to builtins.
This was originally fixed in 46df6906 (execv_dashed_external: wait
for child on signal death, 2017-01-06), but was broken by ee4512ed
(trace2: create new combined trace facility, 2019-02-22) and then
b9140840 (git: avoid calling aliased builtins via their dashed form,
2019-07-29).
Signed-off-by: Trygve Aaberge <trygveaa@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Teach "am", "commit", "merge" and "rebase", when they are run with
the "--quiet" option, to pass "--quiet" down to "gc --auto".
* jc/auto-gc-quiet:
auto-gc: pass --quiet down from am, commit, merge and rebase
auto-gc: extract a reusable helper from "git fetch"
Back in 1991006c (fetch: convert argv_gc_auto to struct argv_array,
2014-08-16), we taught "git fetch --quiet" to pass the "--quiet"
option down to "gc --auto". This issue, however, is not limited to
"fetch":
$ git grep -e 'gc.*--auto' \*.c
finds hits in "am", "commit", "merge", and "rebase" and these
commands do not pass "--quiet" down to "gc --auto" when they
themselves are told to be quiet.
As a preparatory step, let's introduce a helper function
run_auto_gc(), that the caller can pass a boolean "quiet",
and redo the fix to "git fetch" using the helper.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
On Cygwin, the codepath for POSIX-like systems is taken in
run-command.c::start_command(). The prepare_cmd() helper
function is called to decide if the command needs to be looked
up in the PATH. The logic there is to do the PATH-lookup if
and only if it does not have any slash '/' in it. If this test
passes we end up attempting to run the command by appending the
string after each colon-separated component of PATH.
The Cygwin environment supports both Windows and POSIX style
paths, so both forwardslahes '/' and back slashes '\' can be
used as directory separators for any external program the user
supplies.
Examples for path strings which are being incorrectly searched
for in the PATH instead of being executed as is:
- "C:\Program Files\some-program.exe"
- "a\b\c.exe"
To handle these, the PATH lookup detection logic in prepare_cmd()
is taught to know about this Cygwin quirk, by introducing
has_dir_sep(path) helper function to abstract away the difference
between true POSIX and Cygwin systems.
Signed-off-by: Andras Kucsma <r0maikx02b@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In this function, we free the pointer we get from locate_in_PATH and
then check whether it's NULL. However, this is undefined behavior if
the pointer is non-NULL, since the C standard no longer permits us to
use a valid pointer after freeing it.
The only case in which the C standard would permit this to be defined
behavior is if r were NULL, since it states that in such a case "no
action occurs" as a result of calling free.
It's easy to suggest that this is not likely to be a problem, but we
know that GCC does aggressively exploit the fact that undefined
behavior can never occur to optimize and rewrite code, even when that's
contrary to the expectations of the programmer. It is, in fact, very
common for it to omit NULL pointer checks, just as we have here.
Since it's easy to fix, let's do so, and avoid a potential headache in
the future.
Noticed-by: Miriam R. <mirucam@gmail.com>
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Create a new unified tracing facility for git. The eventual intent is to
replace the current trace_printf* and trace_performance* routines with a
unified set of git_trace2* routines.
In addition to the usual printf-style API, trace2 provides higer-level
event verbs with fixed-fields allowing structured data to be written.
This makes post-processing and analysis easier for external tools.
Trace2 defines 3 output targets. These are set using the environment
variables "GIT_TR2", "GIT_TR2_PERF", and "GIT_TR2_EVENT". These may be
set to "1" or to an absolute pathname (just like the current GIT_TRACE).
* GIT_TR2 is intended to be a replacement for GIT_TRACE and logs command
summary data.
* GIT_TR2_PERF is intended as a replacement for GIT_TRACE_PERFORMANCE.
It extends the output with columns for the command process, thread,
repo, absolute and relative elapsed times. It reports events for
child process start/stop, thread start/stop, and per-thread function
nesting.
* GIT_TR2_EVENT is a new structured format. It writes event data as a
series of JSON records.
Calls to trace2 functions log to any of the 3 output targets enabled
without the need to call different trace_printf* or trace_performance*
routines.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A recent update accidentally squelched an error message when the
run_command API failed to run a missing command, which has been
corrected.
* jc/run-command-report-exec-failure-fix:
run-command: report exec failure
In 321fd823 ("run-command: mark path lookup errors with ENOENT",
2018-10-24), we rewrote the logic to execute a command by looking
in the directories on $PATH; as a side effect, a request to run a
command that is not found on $PATH is noticed even before a child
process is forked to execute it.
We however stopped to report an exec failure in such a case by
mistake. Add a logic to report the error unless silent-exec-failure
is requested, to match the original code.
Reported-by: John Passaro <john.a.passaro@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The codebase has been cleaned up to reduce "#ifndef NO_PTHREADS".
* nd/pthreads:
Clean up pthread_create() error handling
read-cache.c: initialize copy_len to shut up gcc 8
read-cache.c: reduce branching based on HAVE_THREADS
read-cache.c: remove #ifdef NO_PTHREADS
pack-objects: remove #ifdef NO_PTHREADS
preload-index.c: remove #ifdef NO_PTHREADS
grep: clean up num_threads handling
grep: remove #ifdef NO_PTHREADS
attr.c: remove #ifdef NO_PTHREADS
name-hash.c: remove #ifdef NO_PTHREADS
index-pack: remove #ifdef NO_PTHREADS
send-pack.c: move async's #ifdef NO_PTHREADS back to run-command.c
run-command.h: include thread-utils.h instead of pthread.h
thread-utils: macros to unconditionally compile pthreads API
Normally pthread_create() rarely fails. But with new pthreads wrapper,
pthread_create() will return ENOSYS on a system without thread support.
Threaded code _is_ protected by HAVE_THREADS and pthread_create()
should never run in the first place. But the situation could change in
the future and bugs may sneak in. Make sure that all pthread_create()
reports the error cause.
While at there, mark these strings for translation if they aren't.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
On systems that do not support multithread, start_async() is
implemented with fork(). This implementation details unfortunately
leak out at least in send-pack.c [1].
To keep the code base clean of NO_PTHREADS, move the this #ifdef back
to run-command.c. The new wrapper function async_with_fork() at least
helps suggest that this special "close()" is related to async in fork
mode.
[1] 09c9957cf7 (send-pack: avoid deadlock when pack-object dies early
- 2011-04-25)
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since commit e3a434468f (run-command: use the
async-signal-safe execv instead of execvp, 2017-04-19),
prepare_cmd() does its own PATH lookup for any commands we
run (on non-Windows platforms).
However, its logic does not match the old execvp call when
we fail to find a matching entry in the PATH. Instead of
feeding the name directly to execv, execvp would consider
that an ENOENT error. By continuing and passing the name
directly to execv, we effectively behave as if "." was
included at the end of the PATH. This can have confusing and
even dangerous results.
The fix itself is pretty straight-forward. There's a new
test in t0061 to cover this explicitly, and I've also added
a duplicate of the ENOENT test to ensure that we return the
correct errno for this case.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In d8193743e0 (usage.c: add BUG() function, 2017-05-12), a new macro
was introduced to use for reporting bugs instead of die(). It was then
subsequently used to convert one single caller in 588a538ae5
(setup_git_env: convert die("BUG") to BUG(), 2017-05-12).
The cover letter of the patch series containing this patch
(cf 20170513032414.mfrwabt4hovujde2@sigill.intra.peff.net) is not
terribly clear why only one call site was converted, or what the plan
is for other, similar calls to die() to report bugs.
Let's just convert all remaining ones in one fell swoop.
This trick was performed by this invocation:
sed -i 's/die("BUG: /BUG("/g' $(git grep -l 'die("BUG' \*.c)
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The slightly misleading name die_bug() of the function intended to
report a bug is actually called always, and only reports a bug if the
passed-in parameter `err` is non-zero.
It uses die_errno() to report the bug, to helpfully include the error
message corresponding to `err`.
However, as these messages indicate bugs, we really should use BUG().
And as BUG() is a macro to be able to report the exact file and line
number, we need to convert die_bug() to a macro instead of only
replacing the die_errno() by a call to BUG().
While at it, use a name more indicative of the purpose: CHECK_BUG().
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is more consistent with the project style. The majority of Git's
source files use dashes in preference to underscores in their file names.
Signed-off-by: Stefan Beller <sbeller@google.com>
Patch generated with Coccinelle and contrib/coccinelle/strbuf.cocci.
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a command sets a new env variable GIT_DIR=.git, we need more context
to know where that '.git' is related to.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Occasionally submodule code could execute new commands with GIT_DIR set
to some submodule. GIT_TRACE prints just the command line which makes it
hard to tell that it's not really executed on this repository.
Print the env delta (compared to parent environment) in this case.
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We normally print full command line, including the program and its
argument. When git_cmd is set, we have a special code path to run the
right "git" program and child_process.argv[0] will not contain the
program name anymore. As a result, we print just the command
arguments.
I thought it was a regression when the code was refactored and git_cmd
added, but apparently it's not. git_cmd mode was introduced before
tracing was added in 8852f5d704 (run_command(): respect GIT_TRACE -
2008-07-07) so it's more like an oversight in 8852f5d704.
Fix it, print the program name "git" in git_cmd mode. It's nice to have
now. But it will be more important later when we start to print env
variables too, in shell syntax. The lack of a program name would look
confusing then.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is the same as the old code that uses trace_argv_printf() in
run-command.c. This function will be improved in later patches to
print more information from struct child_process.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When an hook is present but the file is not set as executable then git will
ignore the hook.
For now this is silent which can be confusing.
This commit adds this warning to improve the situation:
hint: The 'pre-commit' hook was ignored because it's not set as executable.
hint: You can disable this warning with `git config advice.ignoredHook false`
To allow the old use-case of enabling/disabling hooks via the executable flag a
new setting is introduced: advice.ignoredHook.
Signed-off-by: Damien Marié <damien@dam.io>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use the macro ALLOC_ARRAY to allocate an array. This is shorter and
easier, as it automatically infers the size of elements.
Patch generated with Coccinelle and contrib/coccinelle/array.cocci.
Signeg-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
By declaring the task_cb parameter of type `void **`, the signature of
the get_next_task method suggests that the "task-specific cookie" can be
defined in that method, and the signatures of the start_failure and of
the task_finished methods declare that parameter of type `void *`,
suggesting that those methods are mere users of said cookie.
That convention makes a total lot of sense, because the tasks are pretty
much dead when one of the latter two methods is called: there would be
little use to reset that cookie at that point because nobody would be
able to see the change afterwards.
However, this is not what the code actually does. For all three methods,
it passes the *address* of pp->children[i].data.
As reasoned above, this behavior makes no sense. So let's change the
implementation to adhere to the convention suggested by the signatures.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Acked-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In some situations run-command will incorrectly try (and fail) to
execute a directory instead of an executable file. This was observed by
having a directory called "ssh" in $PATH before the real ssh and trying
to use ssh protoccol, reslting in the following:
$ git ls-remote ssh://url
fatal: cannot exec 'ssh': Permission denied
It ends up being worse and run-command will even try to execute a
non-executable file if it preceeds the executable version of a file on
the PATH. For example, if PATH=~/bin1:~/bin2:~/bin3 and there exists a
directory 'git-hello' in 'bin1', a non-executable file 'git-hello' in
bin2 and an executable file 'git-hello' (which prints "Hello World!") in
bin3 the following will occur:
$ git hello
fatal: cannot exec 'git-hello': Permission denied
This is due to only checking 'access()' when locating an executable in
PATH, which doesn't distinguish between files and directories. Instead
use 'is_executable()' which check that the path is to a regular,
executable file. Now run-command won't try to execute the directory or
non-executable file 'git-hello':
$ git hello
Hello World!
which matches what execvp(3) would have done when asked to execute
git-hello with such a $PATH.
Reported-by: Brian Hatfield <bhatfield@google.com>
Signed-off-by: Brandon Williams <bmwill@google.com>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Move the logic for 'is_executable()' from help.c to run_command.c and
expose it so that callers from outside help.c can access the function.
This is to enable run-command to be able to query if a file is
executable in a future patch.
Signed-off-by: Brandon Williams <bmwill@google.com>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signal handlers of the parent firing in the forked child may
have unintended side effects. Rather than auditing every signal
handler we have and will ever have, block signals while forking
and restore default signal handlers in the child before execve.
Restoring default signal handlers is required because
execve does not unblock signals, it only restores default
signal handlers. So we must restore them with sigprocmask
before execve, leaving a window when signal handlers
we control can fire in the child. Continue ignoring
ignored signals, but reset the rest to defaults.
Similarly, disable pthread cancellation to future-proof our code
in case we start using cancellation; as cancellation is
implemented with signals in glibc.
Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
All non-Async-Signal-Safe functions (e.g. malloc and die) were removed
between 'fork' and 'exec' in start_command in order to avoid potential
deadlocking when forking while multiple threads are running. This
deadlocking is possible when a thread (other than the one forking) has
acquired a lock and didn't get around to releasing it before the fork.
This leaves the lock in a locked state in the resulting process with no
hope of it ever being released.
Add a note describing this potential pitfall before the call to 'fork()'
so people working in this section of the code know to only use
Async-Signal-Safe functions in the child process.
Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
All of our standard error handling paths have the potential to
call malloc or take stdio locks; so we must avoid them inside
the forked child.
Instead, the child only writes an 8 byte struct atomically to
the parent through the notification pipe to propagate an error.
All user-visible error reporting happens from the parent;
even avoiding functions like atexit(3) and exit(3).
Helped-by: Eric Wong <e@80x24.org>
Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In order to avoid allocation between 'fork()' and 'exec()' prepare the
environment to be used in the child process prior to forking.
Switch to using 'execve()' so that the construct child environment can
used in the exec'd process.
Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Convert the function used to exec from 'execvp()' to 'execv()' as the (p)
variant of exec isn't async-signal-safe and has the potential to call malloc
during the path resolution it performs. Instead we simply do the path
resolution ourselves during the preparation stage prior to forking. There also
don't exist any portable (p) variants which also take in an environment to use
in the exec'd process. This allows easy migration to using 'execve()' in a
future patch.
Also, as noted in [1], in the event of an ENOEXEC the (p) variants of
exec will attempt to execute the command by interpreting it with the
'sh' utility. To maintain this functionality, if 'execv()' fails with
ENOEXEC, start_command will atempt to execute the command by
interpreting it with 'sh'.
[1] http://pubs.opengroup.org/onlinepubs/009695399/functions/exec.html
Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
According to [1] we need to only call async-signal-safe operations between fork
and exec. Using malloc to build the argv array isn't async-signal-safe.
In order to avoid allocation between 'fork()' and 'exec()' prepare the
argv array used in the exec call prior to forking the process.
[1] http://pubs.opengroup.org/onlinepubs/009695399/functions/fork.html
Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Callers of the run-command API may mark a child as
"clean_on_exit"; it gets added to a list and killed when the
main process dies. Since commit 46df6906f
(execv_dashed_external: wait for child on signal death,
2017-01-06), we respect an extra "wait_after_clean" flag,
which we expect to find in the child_process struct.
When Git is built with NO_PTHREADS, we start "struct
async" processes by forking rather than spawning a thread.
The resulting processes get added to the cleanup list but
they don't have a child_process struct, and the cleanup
function ends up dereferencing NULL.
We should notice this case and assume that the processes do
not need to be waited for (i.e., the same behavior they had
before 46df6906f).
Reported-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Jeff King <peff@peff.net>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Names of the various hook scripts must be spelled exactly, but on
Windows, an .exe binary must be named with .exe suffix; notice
$GIT_DIR/hooks/<hookname>.exe as a valid <hookname> hook.
* js/mingw-hooks-with-exe-suffix:
mingw: allow hooks to be .exe files
Executable files in Windows need to have the extension '.exe', otherwise
they do not work. Extend the hooks to not just look at the hard coded
names, but also at the names extended by the custom STRIP_EXTENSION,
which is defined as '.exe' in Windows.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When you hit ^C to interrupt a git command going to a pager,
this usually leaves the pager running. But when a dashed
external is in use, the pager ends up in a funny state and
quits (but only after eating one more character from the
terminal!). This fixes it.
Explaining the reason will require a little background.
When git runs a pager, it's important for the git process to
hang around and wait for the pager to finish, even though it
has no more data to feed it. This is because git spawns the
pager as a child, and thus the git process is the session
leader on the terminal. After it dies, the pager will finish
its current read from the terminal (eating the one
character), and then get EIO trying to read again.
When you hit ^C, that sends SIGINT to git and to the pager,
and it's a similar situation. The pager ignores it, but the
git process needs to hang around until the pager is done. We
addressed that long ago in a3da882120 (pager: do
wait_for_pager on signal death, 2009-01-22).
But when you have a dashed external (or an alias pointing to
a builtin, which will re-exec git for the builtin), there's
an extra process in the mix. For instance, running:
$ git -c alias.l=log l
will end up with a process tree like:
git (parent)
\
git-log (child)
\
less (pager)
If you hit ^C, SIGINT goes to all of them. The pager ignores
it, and the child git process will end up in wait_for_pager().
But the parent git process will die, and the usual EIO
trouble happens.
So we really want the parent git process to wait_for_pager(),
but of course it doesn't know anything about the pager at
all, since it was started by the child. However, we can
have it wait on the git-log child, which in turn is waiting
on the pager. And that's what this patch does.
There are a few design decisions here worth explaining:
1. The new feature is attached to run-command's
clean_on_exit feature. Partly this is convenience,
since that feature already has a signal handler that
deals with child cleanup.
But it's also a meaningful connection. The main reason
that dashed externals use clean_on_exit is to bind the
two processes together. If somebody kills the parent
with a signal, we propagate that to the child (in this
instance with SIGINT, we do propagate but it doesn't
matter because the original signal went to the whole
process group). Likewise, we do not want the parent
to go away until the child has done so.
In a traditional Unix world, we'd probably accomplish
this binding by just having the parent execve() the
child directly. But since that doesn't work on Windows,
everything goes through run_command's more spawn-like
interface.
2. We do _not_ automatically waitpid() on any
clean_on_exit children. For dashed externals this makes
sense; we know that the parent is doing nothing but
waiting for the child to exit anyway. But with other
children, it's possible that the child, after getting
the signal, could be waiting on the parent to do
something (like closing a descriptor). If we were to
wait on such a child, we'd end up in a deadlock. So
this errs on the side of caution, and lets callers
enable the feature explicitly.
3. When we send children the cleanup signal, we send all
the signals first, before waiting on any children. This
is to avoid the case where one child might be waiting
on another one to exit, causing a deadlock. We inform
all of them that it's time to die before reaping any.
In practice, there is only ever one dashed external run
from a given process, so this doesn't matter much now.
But it future-proofs us if other callers start using
the wait_after_clean mechanism.
There's no automated test here, because it would end up racy
and unportable. But it's easy to reproduce the situation by
running the log command given above and hitting ^C.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Ævar Arnfjörð Bjarmason
Derrick Stolee
Jeff King
Junio C Hamano
Trygve Aaberge
Andras Kucsma
brian m. carlson
René Scharfe
Jeff Hostetler
Nguyễn Thái Ngọc Duy
Johannes Schindelin
Stefan Beller
Damien Marié
Brandon Williams
Eric Wong