Serial-ftdi Issues

[Anton Lundin]

On 29 August, 2017 - John Van Ostrand wrote:

> I'm trying to get the Cochran computers working with the FTDI library and
> serial_ftdi so I can read them on an Android biuld. I'm having two issues that
> I have to ask for advice about.
> 
> 1. In serial_ftdi_read() a failed read exponentially backs off when there's a
> short read. This is causing Cochran reads to timeout about 95% of the time.
> It's as if reading too late causes lost data. Having a fixed backoff of 10ms
> works. Are there any issues if I alter this backoff tactic? It could be
> shortened to 1ms if 10ms is too long.
> 
> diff --git a/core/serial_ftdi.c b/core/serial_ftdi.c
> index f8cba95..bc20ce1 100644
> --- a/core/serial_ftdi.c
> +++ b/core/serial_ftdi.c
> @@ -384,7 +384,7 @@ static dc_status_t serial_ftdi_read (dc_custom_io_t *io,
> void *data, size_t size
>         if (timeout == -1)
>                 timeout = 10000;
>  
> -       int backoff = 1;
> +       int backoff = 10;
>         int slept = 0;
>         unsigned int nbytes = 0;
>         while (nbytes < size) {
> @@ -402,12 +402,11 @@ static dc_status_t serial_ftdi_read (dc_custom_io_t *io,
> void *data, size_t size
>                         }
>                         serial_ftdi_sleep (device, backoff);
>                         slept += backoff;
> -                       backoff *= 2;
>                         if (backoff + slept > timeout)
>                                 backoff = timeout - slept;
>                 } else {
>                         // Reset backoff to 1 on success.
> -                       backoff = 1;
> +                       backoff = 10;
>                 }
>  
>                 nbytes += n;


This is probably related to the issue Jocke sees on android:
https://github.com/Subsurface-divelog/subsurface/pull/421

I don't mind another back off algorithm, just update the comment that
says "Exponential backoff" to.


This came about when Venkatesh converted this async api to a sync api to
work with the api libdivecomputer expects.

> 
> 2. It doesn't make any sense to me but setting serial break on then off is the
> trigger to the Cochran to wake up, so I need the ability to set breaks.
> Currently the code doesn't support setting breaks and is explained with the
> comment "// Not implemented in libftdi yet. Research it further."
> 
> So I looked at it and it's supported with the following libftdi-1.3 function:
> 
>       int ftdi_set_line_property2(struct ftdi_context *ftdi, enum
> ftdi_bits_type bits,
>                                 enum ftdi_stopbits_type sbit, enum
> ftdi_parity_type parity,
>                                 enum ftdi_break_type break_type);
> 
> To make this work seamlessly with libdivecomputer's serial function I need to
> emulate the dc_serial_set_break function. A prototype serial_ftdi_set_break
> function is there and I can change it to use ftdi_set_line_property2, which
> means I have to get the parameters bits, sbit, and parity from somewhere.
> 
> There isn't a function that will read this from the device.
> 
> I could create some global vars in serial_ftdi.c, initialize them with 8,1,N
> values, update them as needed. To support multiple devices concurrently it
> would have to be an array indexed by device but it's not clear what index would
> work best.
> 
> I was thinking of a global variable like:
> 
>    struct serial_parm {
>       struct ftdi_context *ftdi_ctx;
>       enum ftdi_bits_type bits;
>       enum ftdi_stopbits_type sbit;
>       enum ftdi_parity_type parity;
>    } *serial_parms = NULL;
> 
> Then realloc() to add (on serial_ftdi_open) or remove (on serial_ftdi_close).
> When serial_ftdi_configure is called I can search the array and update the
> values so I can use them in serial_ftdi_set_break.
> 
> Does this work or is there a better idea?

Don't do it with a global variable. There's already a ftdi_serial_t
which contains such information, so use that structure and add whatever
more needed to be in it.

Each function gets that structure as dc_custom_io_t userdata argument.


//Anton


-- 
Anton Lundin	+46702-161604