Update ideas.txt.

1 files changed, 13 insertions, 0 deletions

diff --git a/TODO b/TODO
index 57917d9..8ef14fc 100644
--- a/TODO
+++ b/TODO
@@ -12,6 +12,19 @@ FujiChat features, we're almost at parity!
 
 Other stuff:
 
+- If an VBI-driven keyboard buffer turns out to be impossible or
+  too much of a PITA to contemplate, another idea to avoid dropping
+  keystrokes: after every keypress, wait something like 1/4 or 1/3
+  second for another keypress, before checking for incoming net
+  traffic. As long as the user's typing 4 (or 3) keys per second,
+  he won't lose keystrokes. Probably have to put a limit on it,
+  a pathological case would be holding down a key long enough to
+  fill the edit box with 240 of the same character. That would
+  cause a 60 sec delay in net traffic, which might actually cause
+  us to time out... need some data on just how fast a fast typist
+  really types, and on average, how often they pause (to think or
+  read back what they just wrote, etc). Also look into detecting
+  key repeats (SRTIMER).
 - Write a cgetc() replacement that doesn't call the OS K: "get one byte"
   routine. I was avoiding it because it will need a 192-byte table
   (keycode -> atascii lookup)... but I'm spending more than 192 bytes