port.write fails with non-zero writeTimeout on MacOS 10.13.6

[euantorano]

port.write fails with non-zero writeTimeout on MacOS 10.13.6

[euantorano]

Ah, yes. That makes more sense - good catch on my writeTimeout bug too. Luckily I didn't commit straight to master 😄

I'll do some testing after work tonight. Unfortunately I got a new mac recently with USB-C ports and only one USB-C to USB-A adapter, so tetsing my USB to serial ports is difficult without using two computers.

[euantorano]

Yep, looks like the default writeTimeout is TIMEOUT_INFINITE which is -1. The code then checks if writeTimeout != 0 and if so, uses a timeout. This would fail in the infinite case.

[euantorano]

Created PR #43, would you mind checking if it resolves the issue?

[martin-c]

Thanks for your quick response Euan! Unfortunately, it this doesn't seem to fix the problem. I still get the same error: Invalid argument [OSError]. But what I didn't test yesterday is trying a writeTimeout value other than 1000 which appears in serial.nim and 0. It seems any timeout value from 1 to 999 works, but 1000 and up do not.

By the way, I think your commit introduces a change in the logic of the write function: testing for port.writeTimeout > 0'i32 on line 769 means that the conditional on line 780 if port.writeTimeout < 0: will always be false, right? So now timeout values of -1 and 0 both use posix.write after the else clause, and ptrTimer will never be nil whereas before this commit a timeout value of -1 used select with a nil timeout pointer (block indefinitely), and timeout = 0 used select with a 0uS timeout. I'm not sure if this is what you intended or not.

I'm trying to find why select timeout values of 1,000,000uS cause an error on MacOS.

[martin-c]

So this is what needs to go on lines 777 and 778 I think:

    timer.tv_usec = Suseconds((port.writeTimeout mod 1000) * 1000)
    timer.tv_sec = Time(port.writeTimeout div 1000)

[euantorano]

@martin-c Would you mind testing again? During testing I found a bug with read() when using an infinite timeout which caused read() to be called repeatedly, which I've fixed in #43 along with hopefully fixing the write timeout. I've tested sending from my Mac to a windows machine and from a windows machine to my Mac without any issues now.

There is some odd behaviour with SerialStream.readLine(), but I'll need to look into that another day - I had to hit enter twice on my Windows machine for it to be received on the Mac side...