run-command: add stdin callback for parallelization

If a user of the run_processes_parallel() API wants to pipe a large
amount of information to the stdin of each parallel command, that
data could exceed the pipe buffer of the process's stdin and can be
too big to store in-memory via strbuf & friends or to slurp to a file.

Generally this is solved by repeatedly writing to child_process.in
between calls to start_command() and finish_command(). For a specific
pre-existing example of this, see transport.c:run_pre_push_hook().

This adds a generic callback API to run_processes_parallel() to do
exactly that in a unified manner, similar to the existing callback APIs,
which can then be used by hooks.h to convert the remaining hooks to the
new, simpler parallel interface.

Signed-off-by: Emily Shaffer <emilyshaffer@google.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Adrian Ratiu <adrian.ratiu@collabora.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>

1 files changed, 80 insertions, 7 deletions

diff --git a/run-command.c b/run-command.c
index 3989673569..aaf0e4ecee 100644
--- a/run-command.c
+++ b/run-command.c
@@ -1490,6 +1490,16 @@ static int child_is_working(const struct parallel_child *pp_child)
 	return pp_child->state == GIT_CP_WORKING;
 }
 
+static int child_is_ready_for_cleanup(const struct parallel_child *pp_child)
+{
+	return child_is_working(pp_child) && !pp_child->process.in;
+}
+
+static int child_is_receiving_input(const struct parallel_child *pp_child)
+{
+	return child_is_working(pp_child) && pp_child->process.in > 0;
+}
+
 struct parallel_processes {
 	size_t nr_processes;
 
@@ -1659,6 +1669,44 @@ static int pp_start_one(struct parallel_processes *pp,
 	return 0;
 }
 
+static void pp_buffer_stdin(struct parallel_processes *pp,
+			    const struct run_process_parallel_opts *opts)
+{
+	/* Buffer stdin for each pipe. */
+	for (size_t i = 0; i < opts->processes; i++) {
+		struct child_process *proc = &pp->children[i].process;
+		int ret;
+
+		if (!child_is_receiving_input(&pp->children[i]))
+			continue;
+
+		/*
+		 * child input is provided via path_to_stdin when the feed_pipe cb is
+		 * missing, so we just signal an EOF.
+		 */
+		if (!opts->feed_pipe) {
+			close(proc->in);
+			proc->in = 0;
+			continue;
+		}
+
+		/**
+		 * Feed the pipe:
+		 *   ret < 0 means error
+		 *   ret == 0 means there is more data to be fed
+		 *   ret > 0 means feeding finished
+		 */
+		ret = opts->feed_pipe(proc->in, opts->data, pp->children[i].data);
+		if (ret < 0)
+			die_errno("feed_pipe");
+
+		if (ret) {
+			close(proc->in);
+			proc->in = 0;
+		}
+	}
+}
+
 static void pp_buffer_stderr(struct parallel_processes *pp,
 			     const struct run_process_parallel_opts *opts,
 			     int output_timeout)
@@ -1729,6 +1777,7 @@ static int pp_collect_finished(struct parallel_processes *pp,
 		pp->children[i].state = GIT_CP_FREE;
 		if (pp->pfd)
 			pp->pfd[i].fd = -1;
+		pp->children[i].process.in = 0;
 		child_process_init(&pp->children[i].process);
 
 		if (opts->ungroup) {
@@ -1763,6 +1812,27 @@ static int pp_collect_finished(struct parallel_processes *pp,
 	return result;
 }
 
+static void pp_handle_child_IO(struct parallel_processes *pp,
+				const struct run_process_parallel_opts *opts,
+				int output_timeout)
+{
+	/*
+	 * First push input, if any (it might no-op), to child tasks to avoid them blocking
+	 * after input. This also prevents deadlocks when ungrouping below, if a child blocks
+	 * while the parent also waits for them to finish.
+	 */
+	pp_buffer_stdin(pp, opts);
+
+	if (opts->ungroup) {
+		for (size_t i = 0; i < opts->processes; i++)
+			if (child_is_ready_for_cleanup(&pp->children[i]))
+				pp->children[i].state = GIT_CP_WAIT_CLEANUP;
+	} else {
+		pp_buffer_stderr(pp, opts, output_timeout);
+		pp_output(pp);
+	}
+}
+
 void run_processes_parallel(const struct run_process_parallel_opts *opts)
 {
 	int i, code;
@@ -1782,6 +1852,13 @@ void run_processes_parallel(const struct run_process_parallel_opts *opts)
 					   "max:%"PRIuMAX,
 					   (uintmax_t)opts->processes);
 
+	/*
+	 * Child tasks might receive input via stdin, terminating early (or not), so
+	 * ignore the default SIGPIPE which gets handled by each feed_pipe_fn which
+	 * actually writes the data to children stdin fds.
+	 */
+	sigchain_push(SIGPIPE, SIG_IGN);
+
 	pp_init(&pp, opts, &pp_sig);
 	while (1) {
 		for (i = 0;
@@ -1799,13 +1876,7 @@ void run_processes_parallel(const struct run_process_parallel_opts *opts)
 		}
 		if (!pp.nr_processes)
 			break;
-		if (opts->ungroup) {
-			for (size_t i = 0; i < opts->processes; i++)
-				pp.children[i].state = GIT_CP_WAIT_CLEANUP;
-		} else {
-			pp_buffer_stderr(&pp, opts, output_timeout);
-			pp_output(&pp);
-		}
+		pp_handle_child_IO(&pp, opts, output_timeout);
 		code = pp_collect_finished(&pp, opts);
 		if (code) {
 			pp.shutdown = 1;
@@ -1816,6 +1887,8 @@ void run_processes_parallel(const struct run_process_parallel_opts *opts)
 
 	pp_cleanup(&pp, opts);
 
+	sigchain_pop(SIGPIPE);
+
 	if (do_trace2)
 		trace2_region_leave(tr2_category, tr2_label, NULL);
 }