master gremlin: Initialize preview interpreter with active modal gcodes

[Sigma1912]

fixes #4121 and improves preview for gcode with incomplete preambles.

• Creates a helper file in the Gremlin folder that is copied to lib/python on make.
• Adds a test creating the initcode according to the current modal gcodes and checking interpreter output on loading a gcode file lacking a preamble.
• gremlin.py and `test-ui.py' import and use the inicodes helper function
• qt5-graphics.py used by QtDragon can be easily modified to use the same helper function in a separate PR.

No regression on #2753
No regression on #3855

Affects gmoccapy and any other gui using gremlin.

[grandixximo]

s.gcodes is scaled by 10 (G21=210, G38.2=382). In Py3 active_gcodes[i]/10 is float division: 170 -> 'G17.0' (invalid), 382 -> 'G38.2' (ok). Whole codes get a stray .0 and won't parse.

hal_glib.py does it right:

code = "G%d" % (i//10) if i%10==0 else "G%d.%d" % (i//10, i%10)

Suggest same split, and a test that initcode round-trips through the interpreter.

[c-morley]

As qr_graphics uses very similar cide, can you add to or create a pull request for it too? I'm away working and dont want to get them out of step.

[Sigma1912]

@grandixximo

Whole codes get a stray .0 and won't parse.

It's an easy enough change but having a trailing .0 parses fine for me.
Examples:

• G7.0 does the same as G7
• G0.0 X1 does the same as G0 X1

I'll have a look at adding a test. Do you have any pointers on how I might test the preview?

@c-morley
Sure, I'll have a look as soon as this is merged.

[c-morley]

Your the best thanks ;)

[grandixximo]

You're right, I checked read_g in interp_read.cc ( value_read*10, floor, fraction <0.001 accepted ), so G17.0 rounds to 170 and parses identically to G17. My bad, it's purely cosmetic. Only reason to split would be cleaner debug output, fully optional.

For testing the preview headless: gcode.parse(filename, canon, *initcodes) is the core call ( see GLCanon.load_preview in lib/python/rs274/glcanon.py ). You can drive it without a GUI using a canon subclassed from Translated in lib/python/rs274/interpret.py that just records the emitted straight_feed / arc_feed segments. Build the same unitcode/initcode strings your patch generates, feed a g-code file with an incomplete preamble ( e.g. a G2/G3 arc relying on a modal plane, or motion relying on modal G90/G91 ), and assert the recorded segments match the expected path. That can drop straight into tests/ as a non-GUI sim test so it guards against regressions on #4121, #2753 and #3855.

[Sigma1912]

@grandixximo
Trying to run gcode.parse(filename, canon, *initcodes) using the PrintCanonclass but I'm struggling.

Running my rump python code (see below) I get this error message :

Starting 
File exists
Traceback (most recent call last):
  File "/home/user/git/mylinuxcnc-master/tests/interp_initcode/./test-ui.py", line 50, in <module>
    result, seq = gcode.parse(filename, PrintCanon, unitcode, initcode)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: interp_error > 0 but no Python exception set
Shutting down and cleaning up LinuxCNC...

After some print debugging I found that apparently ìnterp_error is 18 after this line:

linuxcnc/src/emc/rs274ngc/gcodemodule.cc

Line 894 in 106519f

pinterp->init();

Any idea what that might mean?

This is the current rump code:

#!/usr/bin/env python3
import gcode
import sys
import os

class PrintCanon:
    def set_g5x_offset(self, *args):
        print("set_g5x_offset", args)

    def set_g92_offset(self, *args):
        print("set_g92_offset", args)

    def next_line(self, state):
        print("next_line", state.sequence_number)
        self.state = state

    def set_plane(self, plane):
        print("set plane", plane)

    def set_feed_rate(self, arg):
        print("set feed rate", arg)

    def comment(self, arg):
        print("#", arg)

    def straight_traverse(self, *args):
        print("straight_traverse %.4g %.4g %.4g  %.4g %.4g %.4g" % args)

    def straight_feed(self, *args):
        print("straight_feed %.4g %.4g %.4g  %.4g %.4g %.4g" % args)

    def dwell(self, arg):
        if arg < .1:
            print("dwell %f ms" % (1000 * arg))
        else:
            print("dwell %f seconds" % arg)

    def arc_feed(self, *args):
        print("arc_feed %.4g %.4g  %.4g %.4g %.4g  %.4g  %.4g %.4g %.4g" % args)


sys.stderr.write("Starting \n")
filename = "test.ngc"
unitcode = "G21"
initcode = "G7"
if os.path.exists(filename):
    print("File exists")
else:
    print("File does not exist")
result, seq = gcode.parse(filename, PrintCanon, unitcode, initcode)

fail = True
if fail:
    sys.exit(1)

sys.stderr.write('Success\n')
sys.exit(0)

[grandixximo]

error 18 isn't a single g-code error, it's the interp_error counter in gcodemodule.cc. It gets bumped every time the interpreter asks the canon a question and the canon has no method to answer ( see the interp_error++ lines in gcodemodule.cc: get_block_delete, get_tool, get_axis_mask, get_external_angular_units, get_external_length_units ). Your bare PrintCanon ( copied from lib/python/rs274/interpret.py ) implements only the motion-output callbacks, not these query callbacks, so it racks up errors during setup. Those callbacks normally live in StatMixin, which PrintCanon doesn't inherit.

For a headless test that needs no running LinuxCNC, just stub them with constants:

class PreviewCanon(PrintCanon):
    def get_external_length_units(self):  return 1.0   # mm; use 1/25.4 for inch
    def get_external_angular_units(self): return 1.0
    def get_axis_mask(self):              return 0b111111111  # all 9 axes
    def get_block_delete(self):           return 0
    def change_tool(self, pocket):        pass
    def get_tool(self, pocket):
        # toolno, x,y,z,a,b,c,u,v,w, diam, frontangle, backangle, orientation
        return (pocket, 0.,0.,0.,0.,0.,0.,0.,0.,0., 0., 0., 0., 0)
    def next_line(self, state):
        self.state = state

canon = PreviewCanon()
result, seq = gcode.parse(filename, canon, "G21", "G7")
if result > gcode.MIN_ERROR:
    print(gcode.strerror(result))

If you'd rather mirror what gremlin actually does ( real tool table, real units ), subclass StatMixin with a live linuxcnc.stat() instead of the stubs, but for a sim test in tests/ the constants above keep it self-contained.

[Sigma1912]

Thanks for the pointers.

I've got something that seems to be working now.

[Sigma1912]

@grandixximo
Added the test. Is that what you had in mind?

[grandixximo]

Add a .gitignore for result, stderr, test.var*, temp_log like tests/startup-state/.gitignore.

[Sigma1912]

Add a .gitignore for result, stderr, test.var*, temp_log like tests/startup-state/.gitignore.

Done

[grandixximo]

The .gitignore went into the wrong dir: it landed in tests/abort/feed-rate/.gitignore and overwrote that file's sim.var* line. Please revert feed-rate to sim.var* and add a new tests/interp_initcode/.gitignore:

result
stderr
test.var*
temp_log

Two other things still open:

1. Duplicated init-code logic. create_unitcode_and_initcode() in test-ui.py is a verbatim copy of the loop in gremlin.py's load(), so the test guards its own copy, not the real code. Factor it into one helper both import.

2. finally: sys.exit(0) defeats the error path. It overrides the SystemExit from except, so the script always exits 0 and the sys.exit(1) calls are dead. Also print(e) goes to temp_log, not stderr. Drop the sys.exit(0) from finally and send the error to stderr.

[Sigma1912]

Thanks for your valuable input (and patience).

The .gitignore went into the wrong dir

Apologies, that was a bit sloppy on my part.

1. Duplicated init-code logic. create_unitcode_and_initcode() in test-ui.py is a verbatim copy of the loop in gremlin.py's load(), so the test guards its own copy, not the real code. Factor it into one helper both import.

• Created a helper file in the Gremlin folder that is copied to lib/python on make.
• gremlin.py and `test-ui.py' import and use the inicodes helper function
• qt5-graphics.py used by QtDragon can be easily modified to use the same helper function in a separate PR.

2. finally: sys.exit(0) defeats the error path. It overrides the SystemExit from except, so the script always exits 0 and the sys.exit(1) calls are dead. Also print(e) goes to temp_log, not stderr. Drop the sys.exit(0) from finally and send the error to stderr.

Done

Tested:

• Gremlin preview using Gmoccapy
• New tests/interp_initcode/

[grandixximo]

All three points look good now, thanks. One tiny thing on the Submakefile: the added rule targets emc/usr_intf/preview_helpers/%, but the file lives in emc/usr_intf/gremlin/, so that rule never matches. The existing ../lib/python/%: emc/usr_intf/gremlin/% rule right above it already copies preview_helpers.py. The new rule can be dropped ( and there's a stray trailing space after preview_helpers.py in the PYTARGETS += line ).

[Sigma1912]

The new rule can be dropped ( and there's a stray trailing space after preview_helpers.py in the PYTARGETS += line ).

Done. Thanks again for the review!

[grandixximo]

Great, all looks good now. Thanks for the quick turnaround and the thorough testing. LGTM.