comments: plans and questions

1 files changed, 37 insertions, 0 deletions

diff --git a/tools/hi-q.c b/tools/hi-q.c
index 59cad5d..769433a 100644
--- a/tools/hi-q.c
+++ b/tools/hi-q.c
@@ -88,6 +88,43 @@ typedef enum {MSG, ENV} channeler;
 //
 // Note that series and stub use the same exit codes as qq.
 //
+
+/*
+Notation for future use:
+ 0< &kb        1> &scr      2> &scr     prog1  # stand-alone
+ 0< &kb        1> &redpipe  2> &scr     prog2  # upstream end of pipe
+ 0< &redpipe   1> &scr      2> &scr     prog3  # downstream end of pipe
+
+ 0< &msg       1< &envelope 2> &log     qmail-queue
+
+ 0< &kb        1> &scr      2> &scr   7> &up    8< &down  parent
+ 0< &up        1> &down     2> &scr                       childprocess
+
+Simple case:
+ 0< &msg       1> &msg2     2> &log     skrewt
+ 0< &msg2      1> &msg3     2> &log     spamc
+ 0< &msg3      1< &envelope 2> &log     qmail-queue
+
+Fancier "triangular piping" case:
+ 0< &msg       1< &env      2> &log   7> &msg2  8> &env2  skrewt
+ 0< &msg2      1> &msg3     2> &log                       spamc
+ 0< &msg3      1< &env2     2> &log                       qmail-queue
+
+Note that units 7 and 8 are arbitrary and could be renumbered.  In
+contrast, many of the other are fixed by standards and/or traditions.
+
+Some questions:
+
+How hard is it to detect a /dry/ pipe segment, i.e. one where all the
+writing-ends have been closed, and all the bytes have been read?  This
+means the reading ends can be closed, freeing up FD units.
+
+Similarly, now hard is it to detect a /broken/ pipe segment, i.e. one
+where all the reading ends have been closed?  This means the writing
+ends can be closed, freeing up FD units.
+
+*/
+
 typedef enum {series, stub, sa, qq, postspam, fail} moder;
 
 class jobber{