redis-cli: always report server errors on read errors.

Before this commit we may have not consumer buffers when a read error is
encountered. Such buffers may contain errors that are important clues
for the user: for instance a protocol error in the payload we send in
pipe mode will cause the server to abort the connection. If the user
does not get the protocol error, debugging what is happening can be a
nightmare.

This commit fixes issue #3756.

src/redis-cli.c

@@ -6073,6 +6073,7 @@ static void pipeMode(void) {
         /* Handle the readable state: we can read replies from the server. */
         if (mask & AE_READABLE) {
             ssize_t nread;
+            int read_error = 0;
 
             /* Read from socket and feed the hiredis reader. */
             do {
@@ -6080,7 +6081,8 @@ static void pipeMode(void) {
                 if (nread == -1 && errno != EAGAIN && errno != EINTR) {
                     fprintf(stderr, "Error reading from the server: %s\n",
                         strerror(errno));
-                    exit(1);
+                    read_error = 1;
+                    break;
                 }
                 if (nread > 0) {
                     redisReaderFeed(reader,ibuf,nread);
@@ -6113,6 +6115,11 @@ static void pipeMode(void) {
                     freeReplyObject(reply);
                 }
             } while(reply);
+
+            /* Abort on read errors. We abort here because it is important
+             * to consume replies even after a read error: this way we can
+             * show a potential problem to the user. */
+            if (read_error) exit(1);
         }
 
         /* Handle the writable state: we can send protocol to the server. */