Allow configuring of coloured console output

[FreddieAkeroyd]

The STREAM_DEBUG_COLOR environment variable can be set to: yes - always generate ANSI colour escape codes on debug/error output no - use plain text output for debug/error auto - (default) output ANSI codes if terminal supports it

[FreddieAkeroyd]

Should I read STREAM_DEBUG_COLOR on every call to ansiEscape()?

[dirk-zimoch]

Hi Freddie, Thanks for your work. Looks good. Reading the environment variable every time may imply some severe overhead. But I have not tested that. At least the glibc implementation uses a linear search over all environment variables each time. (https://code.woboq.org/userspace/glibc/stdlib/getenv.c.html). I would only do that each time the output file changes (-> streamSetLogfile). Maybe that function should be relocated to StreamError.cc?

[FreddieAkeroyd]

I think the output file referred to by streamSetLogfile only gets sent non-coloured output? As far as I could tell coloured output was only being sent to stdout/stderr, but maybe there is an indirect mechanism?

As well as the environment variable there is also checking if output is to console or file:

• would users need to change STREAM_DEBUG_COLOR on a running IOC? Currently they need to set it early in st.cmd and it will never be re-read
• how often should isatty(fileno(epicsGetStdout())) be checked? And is epicsGetStdout() the best thing to use? I am wondering if it might get confused by e.g. temporary redirection in a cmd file and checking stdout directly might be a better indication of an interactive/non-interactive ioc?

[dirk-zimoch]

You may be right. It is so long ago that I last looked at the code. In that case, I do not expect stout to change at run-time thus checking isatty(stdout) once should be sufficient. And it is stdout, not epicsGetStdout() where the output is printed to (I think). I would rather keep EPICS out of the generic code like StreamError.

If the user only ever sets STREAM_DEBUG_COLOR in the startup script before calling iocInit, reading the variable once should be fine. But for changing it at run-time it may be better to have a global variable accessible from iocsh, e.g var streamDebugColor. That would only cost execution time when changing it but not when reading it. In that case an environment variable may be redundant. the iocsh variable can be a boolean initialized with isatty(stdout).

[FreddieAkeroyd]

Yes, an iocsh meachanism sounds a good route. I think IOC variables can only be int/double type so we can either use some integer values (0=auto,1=yes, 2=no) or make streamDebugColor a function instead and use streamDebugColor("auto") in ioc

[dirk-zimoch]

We don't need auto when we initialize with isatty(stdout), because that is auto. The user can then set it to 0 to 1. Setting it to auto at run-time makes not much sense.

[FreddieAkeroyd]

Yes, you are right - we don't re-check stdout so auto as an option doesn't make much sense. I'd been thinking along the lines of e.g. temporarily setting "y" or "n" in a cmd and then going back to the auto determined option afterwards, but you probably either know exactly what you want to go back to or should save a copy of the variable if you really care about the previous settings. So i'll go for integers 0=n,1=y with initial value determined at program startup.

As for variable name I think streamDebugColored rather than streamDebugColor in case somebody thinks there is the option to set the specific colour with the command.

[dirk-zimoch]

:+1:

[FreddieAkeroyd]

So I believe that is working, in that I can toggle the variable and see the control characters appear. Unfortunately something is up with my computer and I have to run "legacy consoles" which don't support the escapes so I can't tell if they would color the terminal, but if the coloring works for you on linux that is probably good enough. I'll see if i can get a colleague to build the branch and try it.

[dirk-zimoch]

Some compiler errors on Linux...

[dirk-zimoch]

With the above fix, it works on Linux. I can switch color on and off at run time. Also starting the ioc with stdout redirected to a file correctly initializes streamDebugColored off.