RE: Re: Re: Using (stripped down ei.h) with generic TCP instead of ports

For this particular effort, I am using Beckhoff’s TwinCAT 3 framework for PC-based industrial control.  In addition to the normal industrial programming standards, they provide a C/C++ implementation that provides *hard* real-time capability through a modified subset of the standard libraries implemented in the kernel mode of the underlying (Windows 10 in this case) OS.  Within this environment I am constrained by the APIs offered to communicate with the outside world / programs in user mode.  I have to use proprietary libraries to get socket functionality etc. and thus I can not use the erl_interface without modification (and the port driver and NIF approaches are not applicable in this scenario).

I have been learning about Erlang and am very intrigued by the language.  I would like to try out some ideas implemented in Erlang but some of the code will have hard real-time constraints as well as other industrial communication protocol requirements.  Thus, I will have to split the code across the BEAM and the TwinCAT3 environments.  Since I am at the beginning, I have the luxury of defining the architecture/interfaces and I thought passing Erlang (binary) terms between the two environments would be a nice approach to allow somewhat native communication between hard real-time C/C++ processes while also avoiding writing an arbitrary serialized protocol from scratch….   Perhaps I am being naïve.  Also, after digging some more, I have discovered that a lot of the functions in ei/erl_interface that I was excited about have been deprecated in OTP 22 and are scheduled for (have been already?) removal.

Thanks for your help and patience,
Brett


From: Albin Stigö <[email protected]>
Sent: Sunday, September 20, 2020 10:58 AM
To: Brett Hemes <[email protected]>
Cc: Max Lapshin <[email protected]>; Erlang Questions <[email protected]>
Subject: [EXTERNAL] Re: Re: Using (stripped down ei.h) with generic TCP instead of ports

Which kernel are you using?

--Albin

On Sat, Sep 19, 2020, 21:29 Brett Hemes <[email protected]<mailto:[email protected]>> wrote:
Could you perhaps elaborate on your reply a bit?  The RT development environment (running in kernel mode) offers only a memory mapped interface besides the aforementioned UDP/TCP option for communication between the RT modules and user mode applications.  I believe “exposing the interface” would require me to both make an interface from scratch and then wrap it as a port driver or NIF to get it into the BEAM.  Aside from some arbitrary handshaking interface via registers I could imagine passing strings via fixed length character buffers but I will still need some form of encoding/decoding… which is how I ended up looking at the ei library.

Also, most of what I have read regarding Erlang seems to favor ports over drivers and NIFs as they better align with the Erlang philosophy but I am too new to all this to really know when each is preferred.

Thanks much,
Brett

From: Albin Stigö <[email protected]<mailto:[email protected]>>
Sent: Thursday, September 17, 2020 9:32 AM
To: Max Lapshin <[email protected]<mailto:[email protected]>>
Cc: Brett Hemes <[email protected]<mailto:[email protected]>>; Erlang Questions <[email protected]<mailto:[email protected]>>
Subject: [EXTERNAL] Re: Using (stripped down ei.h) with generic TCP instead of ports

I would expose the kernel code through a char or block device driver. There are already many facilities in most kernels for effective communication with userpace. a
After all, that's the whole point of a kernel...

--Albin
On Thu, Sep 17, 2020, 16:00 Max Lapshin <[email protected]<mailto:[email protected]>> wrote:
Your idea is ok.  Just strip everything you don't need from ei and it will work for you.

As for me, I like enif api more, but Sverker has pointed that I'm misuing it inside drivers and it may break once =)