2.99 PDB procedure file-openraster-save throws CRITICAL at Gimp.ObjectArray.new(), patch attached (#5984) · Issues · GNOME / GIMP

2.99 PDB procedure file-openraster-save throws CRITICAL at Gimp.ObjectArray.new(), patch attached

Reproduce, simply but without a useful backtrace:

"Export As" format ora

File>New, choose OK
File>Export As.., choose "SelectFileType>Open Raster", choose "Export"

Expect: success

Actual: Console shows

/usr/lib/python3/dist-packages/gi/overrides/GObject.py:266: Warning: g_object_ref: assertion 'G_IS_OBJECT (object)' failed
  _gi._gvalue_set(self, py_value)
/usr/local/lib/x86_64-linux-gnu/gimp/2.99/plug-ins/file-openraster/file-openraster.py:142: Warning: g_object_ref: assertion 'G_IS_OBJECT (object)' failed
  Gimp.get_pdb().run_procedure('file-png-save', [

(gimp-2.99:37): GLib-GObject-CRITICAL **: 10:42:04.955: g_object_ref: assertion 'G_IS_OBJECT (object)' failed

(gimp-2.99:37): GLib-GObject-CRITICAL **: 10:42:04.955: g_object_unref: assertion 'G_IS_OBJECT (object)' failed

(file-png:82): GLib-GObject-CRITICAL **: 10:42:04.985: g_object_ref: assertion 'G_IS_OBJECT (object)' failed

(file-png:82): GLib-GObject-CRITICAL **: 10:42:04.985: g_object_unref: assertion 'G_IS_OBJECT (object)' failed

(file-png:82): GLib-GObject-CRITICAL **: 10:42:04.985: g_object_ref: assertion 'G_IS_OBJECT (object)' failed

(file-png:82): GLib-GObject-CRITICAL **: 10:42:04.985: g_object_ref: assertion 'G_IS_OBJECT (object)' failed

(file-png:82): GLib-GObject-CRITICAL **: 10:42:04.986: g_object_unref: assertion 'G_IS_OBJECT (object)' failed

(file-png:82): GEGL-CRITICAL **: 10:42:05.122: gegl_buffer_get_extent: assertion 'GEGL_IS_BUFFER (buffer)' failed
/usr/local/lib/x86_64-linux-gnu/gimp/2.99/plug-ins/file-png/file-png: fatal error: Segmentation fault
/usr/local/lib/x86_64-linux-gnu/gimp/2.99/plug-ins/file-png/file-png (pid:82): [E]xit, show [S]tack trace or [P]roceed:

I omit the backtrace, its not useful since many errors intervene with the root cause.

To reproduce so that it stops at the first fault:

set env var GIMP_PLUGIN_DEBUG=file-png,fatal-warnings (note openraster.py calls file-png-save, where error is)
as before, try export as ora

Aside: this is one case where my patch #5971 helps debug, since there is a tree of PDB procedure calls,
you don't know which one is at fault ahead of time:

1) install my patch from #5971 (has simple merge conflicts)
2) set env var GIMP_PLUGIN_DEBUG=all,fatal-critical
3) as before, try export as ora

Now the console shows:

Plugin file-openraster.py: GLib-GObject: CRITICAL: g_object_ref: assertion 'G_IS_OBJECT (object)' failed
GLib-GObject (pid:80): [E]xit, show [S]tack trace or [P]roceed:

Enter "s". Snippet of backtrace is:

# Stack traces obtained from PID 80 - Thread 80 #

[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
0x00007f5a3fd22142 in read () from /lib/x86_64-linux-gnu/libc.so.6
  Id   Target Id                                Frame
* 1    Thread 0x7f5a3fa45740 (LWP 80) "python3" 0x00007f5a3fd22142 in read () from /lib/x86_64-linux-gnu/libc.so.6

Thread 1 (Thread 0x7f5a3fa45740 (LWP 80)):
#0  0x00007f5a3fd22142 in read () at /lib/x86_64-linux-gnu/libc.so.6
#1  0x00007f5a3ea822c1 in gimp_stack_trace_print (prog_name=0x7f5a3f195300 "GLib-GObject", stream=0x7f5a3fdfd6a0 <_IO_2_1_stdout_>, trace=0x0) at ../libgimpbase/gimputils.c:1345
        status = 32602
        stack_printed = 0
        gtrace = 0x0
        gimp_pid = "80\000G_IS_OBJECT ("
        buffer = "ct_ref: assertioCRITICAL: g_objep\022n?Z\177\000\000pbK?Z\177\000\000", '\377' <repeats 16 times>, "\000\000\000\000\000\000\000\000\000\000  \000   \000\000\000\000\000\000\377\000\000\000\377\377\377\377\377\377\360\251\220\000\000\000\000\000\360\251\220\000\000\000\000\000\360\251\220", '\000' <repeats 13 times>, "\360\251\220\000\000\000\000\000\360\251\220\000\000\000\000\000\360\251\220", '\000' <repeats 13 times>, "\026\000\000\000\001", '\000' <repeats 12 times>, "S\031?Z\177\000\000\200\311\337?Z\177\000\000\b\000\000\000\000\000\000\000"...
        read_n = 42136848
        sync_fd = {8, 9}
        out_fd = {10, 11}
        fork_pid = 83
        pid = 80
        eintr_count = 0
        tid = 80
#2  0x00007f5a3ea82617 in gimp_stack_trace_query (prog_name=0x7f5a3f195300 "GLib-GObject") at ../libgimpbase/gimputils.c:1515
        buf = "s\n", '\000' <repeats 13 times>
#3  0x00007f5a3d03bdae in gimp_fatal_func (log_domain=0x7f5a3f195300 "GLib-GObject", flags=10, message=0x2819d20 "g_object_ref: assertion 'G_IS_OBJECT (object)' failed", data=0x0) at ../libgimp/gimp-debug.c:397
        sigset = {__val = {0 <repeats 16 times>}}
        level = 0x7f5a3d04ee24 "CRITICAL"
#4  0x00007f5a3f20839c in g_logv () at /lib/x86_64-linux-gnu/libglib-2.0.so.0
#5  0x00007f5a3f208583 in g_log () at /lib/x86_64-linux-gnu/libglib-2.0.so.0
#6  0x00007f5a3f168960 in g_object_ref () at /lib/x86_64-linux-gnu/libgobject-2.0.so.0
#7  0x00007f5a3ea7dc3e in gimp_object_array_new (object_type=Python Exception <class 'gdb.error'> No type named TypeNode.:
, data=0x28da340, length=1, static_data=0) at ../libgimpbase/gimpparamspecs.c:1483
        tmp = 0x28ded90
        i = 0
        array = 0x281dec0
        __func__ = "gimp_object_array_new"
#8  0x00007f5a3ea7dcc6 in gimp_object_array_copy (array=0x281dee0) at ../libgimpbase/gimpparamspecs.c:1510
#9  0x00007f5a3f16175b in g_boxed_copy () at /lib/x86_64-linux-gnu/libgobject-2.0.so.0
#10 0x00007f5a3f161a92 in g_value_set_boxed () at /lib/x86_64-linux-gnu/libgobject-2.0.so.0
#11 0x00007f5a3f30abc7 in  () at /usr/lib/python3/dist-packages/gi/_gi.cpython-38-x86_64-linux-gnu.so
#12 0x00007f5a3f30b1c2 in  () at /usr/lib/python3/dist-packages/gi/_gi.cpython-38-x86_64-linux-gnu.so
#13 0x00000000005f42ea in PyCFunction_Call ()
#14 0x00000000005f46d6 in _PyObject_MakeTpCall ()
#15 0x0000000000570936 in _PyEval_EvalFrameDefault ()
#16 0x00000000005f7146 in _PyFunction_Vectorcall ()
#17 0x000000000056b399 in _PyEval_EvalFrameDefault ()
#18 0x000000000056955a in _PyEval_EvalCodeWithName ()
#19 0x00000000005f7323 in _PyFunction_Vectorcall ()
#20 0x000000000059c654 in  ()
#21 0x00000000005f463f in _PyObject_MakeTpCall ()
#22 0x0000000000570936 in _PyEval_EvalFrameDefault ()
#23 0x000000000056955a in _PyEval_EvalCodeWithName ()
#24 0x00000000005f7323 in _PyFunction_Vectorcall ()
#25 0x000000000056b26e in _PyEval_EvalFrameDefault ()
#26 0x000000000056955a in _PyEval_EvalCodeWithName ()
#27 0x00000000005f7323 in _PyFunction_Vectorcall ()
#28 0x000000000056b26e in _PyEval_EvalFrameDefault ()
#29 0x000000000056955a in _PyEval_EvalCodeWithName ()
#30 0x00000000005f7323 in _PyFunction_Vectorcall ()
#31 0x00000000005f6e27 in PyObject_CallObject ()
#32 0x00007f5a3f2f8e4e in  () at /usr/lib/python3/dist-packages/gi/_gi.cpython-38-x86_64-linux-gnu.so
#33 0x00007f5a3f578e06 in  () at /lib/x86_64-linux-gnu/libffi.so.7
#34 0x00007f5a3f579188 in  () at /lib/x86_64-linux-gnu/libffi.so.7
#35 0x00007f5a3d01be06 in gimp_save_procedure_run (procedure=0x28d3170, args=0x281d8c0) at ../libgimp/gimpsaveprocedure.c:176
        save_proc = 0x28d3170
        remaining = 0x281df00
        return_values = 0x271b378
        run_mode = GIMP_RUN_INTERACTIVE
        image = 0x280fe10
        drawables = 0x285e760
        file = 0x281de80
        n_drawables = 1
        i = 5

Reading the stack trace:

6 throws the CRITICAL

7 ..... "object_type=Python Exception <class 'gdb.error'>" only means that gdb failed while trying to print the stacktrace, not that the openraster.py faulted

I believe the issue starts at about line 149 of openraster.py, trying to create a GimpObjectArray for a list of drawables. (committed recently in dbc19800.)

[
...
GObject.Value(Gimp.ObjectArray, Gimp.ObjectArray.new(Gimp.Drawable, [drawable], 1)),
...
]

This code fixes it (see the attached patch.)

# Create an empty gvalue that is a GimpObjectArray
a_gvalue = GObject.Value(Gimp.ObjectArray.__gtype__)
# Set its value from a list holding one drawable
Gimp.value_set_object_array(a_gvalue, drawable.__gtype__, [drawable])

[
...
a_gvalue,
...
]

I can't explain why the original code fails. Maybe there is a problem in GI annotations. The attached patch might be considered an expedient work-around. Maybe there is another Python construct that works, but several I tried failed.

GObject.Value(Gimp.ObjectArray, Gimp.ObjectArray.new(Gimp.Drawable, [drawable])),
GObject.Value(Gimp.ObjectArray, Gimp.ObjectArray.new(Gimp.Drawable, [drawable], 1)),
GObject.Value(Gimp.ObjectArray, Gimp.ObjectArray.new(drawable.__gtype__, [drawable], 1)),

There could also be an education/documentation issue. It is hard to understand what the gir doc means, for Gimp, language Python. It says:

Gimp.ObjectArray
from gi.repository import Gimp

object_array = Gimp.ObjectArray()

Methods
Gimp.ObjectArray.copy
Gimp.ObjectArray.free
Fields
Gimp.ObjectArray->data
Gimp.ObjectArray->length
Gimp.ObjectArray->object_type
Gimp.ObjectArray->static_data
More Information
Gimp

" object_array = Gimp.ObjectArray()" seems to suggest you could instantiate one, that the class/type name is the constructor?

The new() method is not documented, but in my experience, works sometimes, e.g. for Gimp.Image.new()

Gimp.ObjectArray is a type that you can use where you need to pass a type, and in Python you can instantiate one but it would always be empty? and there is no method to give or take its contents ? and the other methods should rarely be needed? The GI annotation for Gimp.ObjectArray might say "is an opaque type only used to create a GValue using the factory method Gimp.value_set_object_array()" ? Or maybe a Gimp.ObjectArrayFactory would be helpful?

Related: #5312 #5838 (closed) (dbc19800) both trying to fix openraster

Context: self-built recent master using meson build Ubuntu 20.04

0003-file-openraster-save-throws-CRITICAL.patch