Incorrect _COMMTIMEOUTS struct in serial_windows.go

[dmnef]

The correct field order should be like this:

type _COMMTIMEOUTS struct {
    ReadIntervalTimeout         uint32
    ReadTotalTimeoutConstant    uint32
    ReadTotalTimeoutMultiplier  uint32
    WriteTotalTimeoutConstant   uint32
    WriteTotalTimeoutMultiplier uint32
}

See comments there: http://msdn.microsoft.com/en-us/library/windows/desktop/aa363190%28v=vs.85%29.aspx

[lnmx]

Hi @dmnef, thanks for the report.

Both COMMTIMEOUTS and SERIAL_TIMEOUTS show the same field order, with the Multipliers before the Constants. The comments on COMMTIMEOUTS did not make it any clearer.

I would like to run a few tests on my Windows system to see if the fields seem to be swapped, and if so, I will make the change.

[dmnef]

Ok, my story:

I tried to communicate with a bootloader residing in an MCU connected via standard FTDI-based USB-to-serial converter.
Communication was simple: my program would send a 12-byte command and wait for an 5-byte answer. I've set a timeout by defining ReadTotalTimeoutConstant to desired value (5000 milliseconds in my case) and zeroing the other fields. I've also set 1 millisecond timeout for VCP.
Because there was only one time-consuming command, most of the time MCU answered almost instantly (about 50us) and the interval between communication sessions was expected to be the same as VCP's timeout -- 1ms.
I have already tried the same setup using Python and pyserial lib and everything worked as expected.
But writing almost the same program in Go i've run into a strange performance problem: the interval was unstable and about 16ms (strange -- because VCP's default timeout value is also 16ms).

Problem was solved by swapping mentioned fields.

[lnmx]

Hi @dmnef, I added examples/timeout to test this out. With a USB-to-Serial adapter (nothing connected to the serial end), I set the timeouts and attempt to read some bytes. The Read() call should block for read_constant+(read_length*read_multiplier) before returning 0 bytes.

For read_constant=500, read_length=10 and read_multiplier=100, I get:

read = 0, time = 1500 ms, err = <nil>

...as expected. If the fields were backwards, I expect the timeout to happen at 5100ms.

Could you try this example in your environment and see if you get different results?

[dmnef]

I tried. Got exactly same results. Sorry for misleading.

Although the problem i've mentioned still exists.