Re: [gforth] problem with gforth7 and serial access

[kibler]

I just completed some testing that got me nowhere.
I tried both 7.0 and 7.9 both with same results.

Basically my setup is thus - I have SiLabs board connected
to /dev/ttyS0 running at 38400 Baud, and sending a non-stop
full page of text that includes the cpu temperature and mem
dump of a web page.

When using minicom, I get the full page of output, refreshing
at about a one second rate. No problems with minicom, works
as expected.

Run gforth inside the contrib directory after changing the
terminal program to do 38400 instead of the 9600 - nothing.
Try and run the string and get:

  ok
COM1 B38400 c" 8N1" terminal
:8: Interpreting a compile-only word
COM1 B38400 >>>c"<<< 8N1" terminal
Backtrace:
$7FBA30F8ECF8 throw

Do "term" and get only a blank screen, F1 doesn't
do anything. Should be seeing the same text I see in minicom,
but nothing - no errors, no help, nothing.

Got any ideas what might be going on?
Will look at the R8C code next for ideas.
Have two serial ports, so will next try "term" talking to one,
and minicom talking on other to see if that works.
Bill.


On 03/22/12 08:36, Bernd Paysan wrote:
> Am Donnerstag, 22. März 2012, 08:30:52 schrieb kibler:
> > I am currently looking at the entire contrib section and it seems that
> > the c-lib calls are in 7.0, and as such then we might only have to load
> > the entire contrib set to get working serial on 7.0 versions. I will try
> > that later today, as that seems like a do-able compromise. We would
> > just need to ship all of those files with our own code - an option we
> > already use.
> >
> > Thanks Bernd for the suggested arch code - I will look at that as well,
> > but my minor checking says that the select code is broken, and thus
> > I would expect that to not work - assuming my testing is valid - not
> > always a good assumption to make.
> The terminal.fs code in arch/r8c is not architecture-dependent, it is just an
> application for the R8C board, using the serial interface of a PC.  And it
> does not use the select code of key_avail.
>
> I don't know why the select code should not work on serial lines, this must be
> a bug in Linux.