Note problems with lossless msg-handling alternatives
authorSiraaj Khandkar <siraaj@khandkar.net>
Thu, 26 Mar 2020 21:45:16 +0000 (17:45 -0400)
committerSiraaj Khandkar <siraaj@khandkar.net>
Thu, 26 Mar 2020 21:45:16 +0000 (17:45 -0400)
x5/khatus.c

index 35509d4..5928dfb 100644 (file)
@@ -291,6 +291,19 @@ slots_read(Config *cfg, struct timespec *ti, char *buf)
                                 * This is an acceptable trade-off because we
                                 * are a stateless reporter of a _most-recent_
                                 * status, not a stateful accumulator.
+                                *
+                                * ### LOSSLESS ALTERNATIVES ###
+                                * - Read each pipe until EOF before reading
+                                *   another.
+                                *   PROBLEM: a fast writer can trap us in the
+                                *   read loop.
+                                *
+                                * - Read each pipe until EOM, but close only
+                                *   at EOF.
+                                *   PROBLEM: a fast writer can fill the pipe
+                                *   faster than we can read it and we end-up
+                                *   displaying stale data.
+                                *
                                 */
                                case END_OF_MESSAGE:
                                case END_OF_FILE:
This page took 0.018371 seconds and 4 git commands to generate.