4.8.1 planning

[Linus Torvalds]

On Thu, Jul 12, 2018 at 3:16 PM Anton Lundin <glance at acc.umu.se> wrote:
>
> I just dug up my Sport and put a battery in it and it fails to to
> download, Both on linux and on Android. I tested it via rfcomm and that
> worked just fine.

So this actually would mesh with the delay being the issue.

If you use /dev/rfcomm and just open it as a regular serial device,
then you use the src/serial_posix.c code, and dc_iostream_sleep() will
call dc_serial_sleep().

Which does the nanosleep that it always used to do, and that we never
used to override.

The new libdivecomputer "iostream" model made everybody have to
implement their own sleep implementation, which was obviously not a
great idea. Not even libdivecomputer got it right (ie the socket sleep
bug on OS X).

So the "works with rfcomm, doesn't work with direct bluetooth socket"
would be explained by that sleep difference.

And it also would explain why it just randomly works for some people
regardless, since it's all just random timing anyway.

But it would be great if somebody could test that patch. None of the
dive computers I have care (and most of them don't have a front end
that does any delays in the first place).

              Linus