Gmoccapy - open_file in gmoccapy.pref cause crash

[zz912]

RIP instalation
branch master

If I add this to gmoccapy.pref:
open_file = /home/zdenek/linuxcnc/linuxcnc-zz912/configs/nc_files/3D_Chips.ngc
it will crash Gmoccapy

[Gmoccapy][ERROR]  Found an error!
The following information may be useful in troubleshooting:

Traceback (most recent call last):
  File "/home/zdenek/linuxcnc/linuxcnc-zz912/bin/gmoccapy", line 6136, in <module>
    inifile = sys.argv[2]
              ~~~~~~~~^^^
IndexError: list index out of range
 (gmoccapy:62)

[Sigma1912]

Same here.

Debugging like this:

With 'open_file = /home/user/git/linuxcnc-master/configs/nc_files/arcspiral.ngc'

2-sys.argv:  ['']

[Sigma1912]

Storing 'sys.arcv' seems to fix the startup issue and loads the file:

    store_sys_argv= sys.argv
    # instantiate gmoccapy
    app = gmoccapy(sys.argv)

    # get the INI path
    inifile = store_sys_argv[2]

    # find the POST_GUI HAL file path
    LOG.debug("inifile = {0} :".format(store_sys_argv[2]))
    postgui_halfile = app.get_ini_info.get_postgui_halfile()

However there is another error when trying to open the 'Calibration' tool on the settings page:

[Sigma1912]

the second error could also be fixed by using a local copy of argv:

I haven't really found out why the 'sys.argv' list is suddenly empty when an 'open_file' is set.
So the above 'fixes' seem rather amateurish.

[Sigma1912]

Maybe @hansu has a more educated fix?

[Sigma1912]

Not sure what to make of this but it seems that after

'        self.widgets["hbx_upper"].show()':

sys.argv is empty:

[Sigma1912]

Git bisect points to this
[d71c67c] Fix Py_UnbufferedStdioFlag deprecation warning. Fix python initialization using config. Fix releasing python config after use.

[BsAtHome]

The question is what self.widgets["hbx_upper"].show() actually does. Does the .show() method or its descendants scan the sys.argv array? Maybe to do sizing/full-screen magic?

The problem seems unrelated to what I said in #3511 (comment) because the example that shows the bug does not involve the PyConfig afaics. This may actually be a bug caused somewhere else that only got exposed by the #3265 PyConfig patch.

What you can test is whether it is a problem in the .show() method or something else. What happens when you reassign sys.argv after the first call to .show() in line 362 and then what does it say after the .show() call in line 364? Is it again emptied or stable? If it is empty again, then you can be rather sure that there is a problem in .show() or a memory corruption bug that exposes the problem in .show(). If it is a simple memory bug, then you may want to do some juggling to copy the data and reassigning the sys.argv object to be another object you just created.

[Sigma1912]

Does the .show() method or its descendants scan the sys.argv array? Maybe to do sizing/full-screen magic?

I have no idea. It seems to be related to something in the "hbx_upper" container as commenting that line will make the problem go away (but of course part of the gui is not visible):

This may actually be a bug caused somewhere else that only got exposed by the #3265 PyConfig patch.

That may well be the case. It might also be a timing issue (race condition) as @c-morley apparently does not see this error on his system: #3511 (comment)

To me it seemed easier to just store the arguments instead of chasing this down the rabbit hole hence the approach in the fix in PR #3511

[Sigma1912]

It seems that this has something to do with gremlin preview. If I remove the gremlin widget from gmoccapy.glade and all gremin related code in gmoccapy.py the gui starts and the gcode file is apparently loaded. The gcode is not visible but I can start the program:

[Sigma1912]

I guess it could also be the now missing gcode viewer ie the 'EMC_SourceView' widget.

[Sigma1912]

If I undo the deleting of the gremlin preview in gmoccapy.glade (ie the 'HAL_Gremlin' widget) the error reappears:

[Gmoccapy][ERROR]  Found an error!
The following information may be useful in troubleshooting:

Traceback (most recent call last):
  File "/home/user/git/linuxcnc-master/bin/gmoccapy", line 6139, in <module>
    inifile = sys.argv[2]
              ~~~~~~~~^^^
IndexError: list index out of range

Which would confirm that it is indeed the gremlin widget that is causing this issue.
https://github.com/LinuxCNC/linuxcnc/blob/master/lib/python/gladevcp/hal_gremlin.py

[BsAtHome]

Until so far, the problem is tracked to happen when the preview is loaded from the call GlCanonDraw::load_preview() in the call to gcode.parse() (

linuxcnc/lib/python/rs274/glcanon.py

Line 2013 in 9bfdb69

result, seq = gcode.parse(f, canon, *args)

).

That call will invoke code from the gcode.so library and is actually a call to parse_file() in gcodemodule.cc (

linuxcnc/src/emc/rs274ngc/gcodemodule.cc

Line 855 in 9bfdb69

static PyObject *parse_file(PyObject * /*self*/, PyObject *args) {

).

There the interpreter will be instantiated and the old one will be discarded is it exists. Now, the problem can be in the way this interacts with PyConfig, which is afaik not supposed to be called multiple times. The code already has a python interpreter running afaik, which is most likely already initialized. So, the instantiation of the Interp:

linuxcnc/src/emc/rs274ngc/gcodemodule.cc

Lines 873 to 880 in 9bfdb69

	if(pinterp) {
	delete pinterp;
	pinterp = 0;
	}
	if(interpname && *interpname)
	pinterp = interp_from_shlib(interpname);
	if(!pinterp)
	pinterp = new Interp;

may be causing the whole python interpreter to be zapped and reinitialized and taking the sys.argv array with it. That probably indicates that the use of PyConfig in the Interp class is wrong or must take the current python interpreter's config and only modify it.

[c-morley]

In qt5_graphics.py uses virtually the same code as gremlin.
I see the same behavior, sys.argv is stripped.