ISC /dev/kdmouse "FIFO" problem

[Checkpoint Technologies]

I have a small problem with /dev/kdmouse: there appears to be a 4-byte
buffer in between the mouse messages (or whatever generates them) and my
task, which is trying to read them.  Since mouse messages are 3 bytes
long, receiving a single message is unreliable.

It's like this: I'll begin reading from /dev/kdmouse. I press a mouse
button *very* *carefully*, making sure the mouse doesn't also move.
This generates a single message indicating button down, but my program
receives nothing.  When I release the button, I get the button-press
message, but not the button-release message. I press it again, and I get
the button-release message.  Then I release it again, and I get a
button-press *and* a button-release.

My program has tried this several different ways, both a loop reading 1
character 3 times, and a single read for 3 characters.

I know one commercial X server (Xdgis) which could not solve this
problem and resorted to polling the mouse using ioctl(MOUSEIOCREAD).
This is unacceptable.  I also know that ISC'x own X server works just
right. So there's some majik to incant that I don't know, it appears.
I'd like to know. Could some kind soul enlighten me? Thanks all.
-- 
First comes the logo: C H E C K P O I N T  T E C H N O L O G I E S      / /  
                                                ckp at grebyn.com      \\ / /    
Then, the disclaimer:  All expressed opinions are, indeed, opinions. \  / o
Now for the witty part:    I'm pink, therefore, I'm spam!             \/