Fix errors not printed

[hdiethelm]

This fixes #3948

The commit ddf60cd inhibits printing twice by always using old_handler() to print from motion and inhibits console print for motion->milltask messages. TBD if this is fine like this. Otherwise, I would just accept that errors are printed twice. Better than not at all.

Some options I tried and then discarded:

• Add a flag to emcmot_error_t that shows if milltask is up and running and only invoke old_handler() if milltask is not running

  • When milltask is not yet running and then starts, the messages are still printed twice
  • When milltask dies between emcmotErrorPutfv() and the polling of the error buffer, nothing is shown

• Inhinibit print to console by flag in milltask

  • This would need a flag in the dbuf struct which tells the milltask if the message was printed also with old_handler(). This dbuf looks complicated and it is bugprone if I change something here. Something like this would be needed while inhibitConsole is passed trough the dbuf:

  // read the emcmot error
    if (0 != usrmotReadEmcmotError(&inhibitConsole, errorString)) {
        // no error, so ignore
    } else {
        // an error to report
        // disable stdout print due to this error is from motion
        // and already printed to stdout
        if(inhibitConsole){
            RCS_PRINT_DESTINATION_TYPE prev_dest = get_rcs_print_destination();
            set_rcs_print_destination(RCS_PRINT_TO_NULL);
        }
        emcOperatorError("%s", errorString);
        if(inhibitConsole){
            set_rcs_print_destination(prev_dest);
        }
    }

[hdiethelm]

Meanwhile, I found a solution that doesn't print twice on the console. But the message in the console is not translated any more. See last commit ddf60cd

@BsAtHome Which version do you think is better?

In the below images, I first trigger an error using G2 I1000 F2000 which is printed with reportError(). Then I trigger loadrt pushmsg msgs="e|error|error" to show an rtapi error.

Before last commit:

After last commit:

Edit: There was a bug visible in the image. Fixed in force push.

[BsAtHome]

Meanwhile, I found a solution that doesn't print twice on the console. But the message in the console is not translated any more.

Where in the chain is the translation performed?

[hdiethelm]

Meanwhile, I found a solution that doesn't print twice on the console. But the message in the console is not translated any more.

Where in the chain is the translation performed?

I have no clue how this works. rcs_print() -> rcs_fputs() just does fputs(_str, stdout); and then it is translated. Might be some linker trick that overlays an stdout filter?

[hdiethelm]

Meanwhile, I found a solution that doesn't print twice on the console. But the message in the console is not translated any more.

Where in the chain is the translation performed?

I have no clue how this works. rcs_print() -> rcs_fputs() just does fputs(_str, stdout); and then it is translated. Might be some linker trick that overlays an stdout filter?

Ah now, that one was well hidden:

Before this line it is untranslated, after it is translated:

linuxcnc/src/emc/motion/usrmotintf.cc

Line 218 in 3cd702c

result = snprintdbuf(e, EMCMOT_ERROR_LEN, &di);

The translation is here:

linuxcnc/src/emc/motion/stashf_wrap.h

Line 29 in 3cd702c

if(*fmt) fmt = gettext(fmt);

And if %s is used, it is here:

linuxcnc/src/emc/motion/stashf_wrap.h

Line 85 in 3cd702c

if(*s) s = gettext(s);

snprintdbuf() is a good name for translating text... ;-)

[hdiethelm]

Tests with RTAI show one change:
Motion errors are now not shown anymore in terminal but in the kernel messages like all other real time errors.
The error is still shown in the GUI.

If desired, there is probably a way to enable the terminal errors for RTAI, but they are anyway shown in the GUI.

[andypugh]

Is this still draft, or good to go?