Hopefully fix crash when the same frame is encoded twice.prores-debug

Previously if some frame X was passed to the encoder twice
I think this would happen:

X1 written; X stored to _pending_images
X2 written; _pending_images remains the same
X1 encode finishes; X is discarded from _pending_images
Data for X2 is read by the encoder but was just freed

I think this might have resulted in intermittent crashes, which were
fairly common in the 2.17.x branch with the test
ffmpeg_encoder_prores_regression_1

But I didn't conclusively prove it.

1 files changed, 13 insertions, 3 deletions

diff --git a/src/lib/ffmpeg_file_encoder.cc b/src/lib/ffmpeg_file_encoder.cc
index 1294b9b3c..8045600fe 100644
--- a/src/lib/ffmpeg_file_encoder.cc
+++ b/src/lib/ffmpeg_file_encoder.cc
@@ -411,7 +411,13 @@ FFmpegFileEncoder::video (shared_ptr<PlayerVideo> video, DCPTime time)
 
 	{
 		boost::mutex::scoped_lock lm (_pending_images_mutex);
-		_pending_images[image->data()[0]] = image;
+		auto key = image->data()[0];
+		auto iter = _pending_images.find(key);
+		if (iter != _pending_images.end()) {
+			iter->second.second++;
+		} else {
+			_pending_images[key] = { image, 1 };
+		}
 	}
 
 	for (int i = 0; i < 3; ++i) {
@@ -502,7 +508,11 @@ void
 FFmpegFileEncoder::buffer_free2 (uint8_t* data)
 {
 	boost::mutex::scoped_lock lm (_pending_images_mutex);
-	if (_pending_images.find(data) != _pending_images.end()) {
-		_pending_images.erase (data);
+	auto iter = _pending_images.find(data);
+	if (iter != _pending_images.end()) {
+		iter->second.second--;
+		if (iter->second.second == 0) {
+			_pending_images.erase(data);
+		}
 	}
 }