tab-completion of cached_method in extension class crashes SageMath

[videlec]

Tab completion on the prefix of a cached_method of an extension class not in Sage source code provokes of crash. See https://gitlab.com/videlec/sage_tab_completion_bug to reproduce.

The bug was confirmed reproducibly on

• archlinux sage 9.3.rc2
• online cocalc with sage 9.2

The bug is not present on

• docker sage 8.8 and 9.1.rc2

Workaround

Compile your Cython code with

import Cython.Compiler.Options
Cython.Compiler.Options.embed_pos_in_docstring = True

Tasks

• tab-completion sometimes access attributes #31643: about tab-completion accessing attributes
• fix sage inspection module

Component: misc

Issue created by migration from https://trac.sagemath.org/ticket/31632

[dimpase]

comment:2

This is already in 9.2 (reproduced following the instructions in the gitlab fork)

$ sage
┌────────────────────────────────────────────────────────────────────┐
│ SageMath version 9.2, Release Date: 2020-10-24                     │
│ Create a "Sage Worksheet" file for the notebook interface.         │
│ Enhanced for CoCalc.                                               │
│ Using Python 3.8.5. Type "help()" for help.                        │
└────────────────────────────────────────────────────────────────────┘
sage: sage: import example_cython 
....: sage: a = example_cython.A() 
....:                                                                                                                                                                                 
sage: a.one                                                                                                                                                                           
<built-in method one of example_cython.A object at 0x7ff9bdaceb60>
sage: a.t                                                                                                                                                                             
---------------------------------------------------------------------------
---------------------------------------------------------------------------
KeyError                 Python 3.8.5: /ext/sage/sage-9.2/local/bin/python3
                                                   Fri Apr  9 15:07:38 2021
A problem occurred executing Python code.  Here is the sequence of function
calls leading up to the error, with the most recent (innermost) call last.
/ext/sage/sage-9.2/local/lib/python3.8/site-packages/jedi/cache.py in wrapper(self=<Completion: two>, *args=(), **kwargs={})
    108         try:
--> 109             return dct[key]
        dct = {}
        key = ((), frozenset())
    110         except KeyError:

KeyError: ((), frozenset())

During handling of the above exception, another exception occurred:

---------------------------------------------------------------------------
KeyError                 Python 3.8.5: /ext/sage/sage-9.2/local/bin/python3
                                                   Fri Apr  9 15:07:38 2021
A problem occurred executing Python code.  Here is the sequence of function
calls leading up to the error, with the most recent (innermost) call last.
/ext/sage/sage-9.2/local/lib/python3.8/site-packages/jedi/cache.py in wrapper(self=<CompiledValue: <sage.misc.cachefunc.CachedMethodCallerNoArgs obje..>, *args=(), **kwargs={})
    108         try:
--> 109             return dct[key]
        dct = {}
        key = ((), frozenset())
    110         except KeyError:

KeyError: ((), frozenset())
...

[videlec]

comment:4

Replying to @dimpase:

This is already in 9.2 (reproduced following the instructions in the gitlab fork)

Thanks for testing. I believe that it is a Python3 thing (in particular quite old).

[videlec]

comment:8

On conda (with sage 9.2) I get the error but no crash

Traceback (most recent call last):
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/cache.py", line 110, in wrapper
    return dct[key]
KeyError: ((), frozenset())

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/cache.py", line 110, in wrapper
    return dct[key]
KeyError: (('py__doc__',), frozenset())

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/sage/misc/sageinspect.py", line 2378, in sage_getsourcelines
    return inspect.getsourcelines(obj)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 955, in getsourcelines
    lines, lnum = findsource(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 768, in findsource
    file = getsourcefile(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 684, in getsourcefile
    filename = getfile(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 666, in getfile
    type(object).__name__))
TypeError: module, class, method, function, traceback, frame, or code object was expected, got method_descriptor

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/sage/misc/sageinspect.py", line 2378, in sage_getsourcelines
    return inspect.getsourcelines(obj)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 955, in getsourcelines
    lines, lnum = findsource(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 768, in findsource
    file = getsourcefile(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 684, in getsourcefile
    filename = getfile(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 653, in getfile
    raise TypeError('{!r} is a built-in class'.format(object))
TypeError: <class 'method_descriptor'> is a built-in class

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/IPython/terminal/ptutils.py", line 115, in get_completions
    yield from self._get_completions(body, offset, cursor_position, self.ipy_completer)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/IPython/terminal/ptutils.py", line 131, in _get_completions
    for c in completions:
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/IPython/core/completer.py", line 438, in _deduplicate_completions
    completions = list(completions)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/IPython/core/completer.py", line 1827, in completions
    for c in self._completions(text, offset, _timeout=self.jedi_compute_type_timeout/1000):
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/IPython/core/completer.py", line 1884, in _completions
    signature = _make_signature(jm)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/IPython/core/completer.py", line 993, in _make_signature
    signatures = completion.get_signatures()
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/api/classes.py", line 576, in get_signatures
    for s in self._get_signatures()
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/api/classes.py", line 565, in _get_signatures
    return [sig for name in names for sig in name.infer().get_signatures()]
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/api/classes.py", line 565, in <listcomp>
    return [sig for name in names for sig in name.infer().get_signatures()]
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/base_value.py", line 512, in get_signatures
    return [sig for c in self._set for sig in c.get_signatures()]
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/base_value.py", line 512, in <listcomp>
    return [sig for c in self._set for sig in c.get_signatures()]
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/compiled/value.py", line 138, in get_signatures
    _, return_string = self._parse_function_doc()
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/cache.py", line 112, in wrapper
    result = method(self, *args, **kwargs)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/compiled/value.py", line 146, in _parse_function_doc
    doc = self.py__doc__()
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/compiled/value.py", line 116, in py__doc__
    return self.access_handle.py__doc__()
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/compiled/subprocess/__init__.py", line 386, in _workaround
    return self._cached_results(name, *args, **kwargs)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/cache.py", line 112, in wrapper
    result = method(self, *args, **kwargs)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/compiled/subprocess/__init__.py", line 390, in _cached_results
    return self._subprocess.get_compiled_method_return(self.id, name, *args, **kwargs)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/compiled/subprocess/functions.py", line 27, in get_compiled_method_return
    return getattr(handle.access, attribute)(*args, **kwargs)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/jedi/inference/compiled/access.py", line 189, in py__doc__
    return inspect.getdoc(self._obj) or ''
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 601, in getdoc
    doc = object.__doc__
  File "sage/docs/instancedoc.pyx", line 211, in sage.docs.instancedoc.InstanceDocDescriptor.__get__ (build/cythonized/sage/docs/instancedoc.c:1800)
    return self.instancedoc(obj)
  File "sage/misc/cachefunc.pyx", line 877, in sage.misc.cachefunc.CachedFunction._instancedoc_ (build/cythonized/sage/misc/cachefunc.c:5026)
    sourcelines = sage_getsourcelines(f)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/sage/misc/sageinspect.py", line 2397, in sage_getsourcelines
    return sage_getsourcelines(obj.__class__)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/sage/misc/sageinspect.py", line 2395, in sage_getsourcelines
    return sage_getsourcelines(B)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/sage/misc/sageinspect.py", line 2398, in sage_getsourcelines
    raise err
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/site-packages/sage/misc/sageinspect.py", line 2378, in sage_getsourcelines
    return inspect.getsourcelines(obj)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 955, in getsourcelines
    lines, lnum = findsource(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 768, in findsource
    file = getsourcefile(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 684, in getsourcefile
    filename = getfile(object)
  File "/home/vincent/miniconda3/envs/flatsurvey/lib/python3.7/inspect.py", line 653, in getfile
    raise TypeError('{!r} is a built-in class'.format(object))
TypeError: <class 'object'> is a built-in class

[mwageringel]

comment:11

I have also had problems with cached_method in Cython code, in the past. As far as I recall, it is necessary to define a public __cached_methods attribute for them to work. I have quickly checked it on your example though and it did not seem to resolve the issue, so this may be a different problem.

[kliem]

comment:12

You should in addition inherit from SageObject, when using cached_method (or at least try), but this also did not resolve the issue.

[kliem]

comment:13

Indeed this seems to have appeared at some point recent.

On debian buster I can't reproduce it with sage 9.1, at least not the hard crash. The underlying problem to fetch a.two.__doc__ is way older, but hasn't lead to a hard crash.

This seems to be related #31480 (which was a blocker, because as you noted hard crashes are annoying and it had a super easy fix). This tab completion hard crash also appeared with some ipython upgrade or so that happened recently.

[kliem]

comment:14

I think the mentioned workaround is also a possible solution. You cannot just cythonize a file and then except it to work completely nice with sage. It should be clear that you need to run a special cythonize function that enables the proper defaults for sage (e.g. cython aliases etc).

However, I can't find such a function and I think sage should provide a customized cythonize_for_sage that sets the proper arguments (also we usually assume cdivison=True etc, I don't know if this is respected in your case either).

Or at least provide the cythonize arguments somewhere...

Am I making any sense? Anyway, I think this is out of scope for sage 9.3

[videlec]

comment:15

Replying to @kliem:

I think the mentioned workaround is also a possible solution. You cannot just cythonize a file and then except it to work completely nice with sage. It should be clear that you need to run a special cythonize function that enables the proper defaults for sage (e.g. cython aliases etc).

However, I can't find such a function and I think sage should provide a customized cythonize_for_sage that sets the proper arguments (also we usually assume cdivison=True etc, I don't know if this is respected in your case either).

Or at least provide the cythonize arguments somewhere...

Am I making any sense? Anyway, I think this is out of scope for sage 9.3

I see at least three problems:

• Access to the documentation should not trigger the access the source code. The CachedFunction._instancedoc_ makes a call to sage_getsourcelines if the line in the source code is not available. This is exactly what the workaround in the ticket description provides.

• Accessing the documentation/source should never result in a crash

• Why on earth tab-completion have anything to do with access to the documentation?

I believe there is work to be done on different fronts: the sageinspect module, the CachedFunction._instancedoc_ and understand how the tab-completion works.

[kliem]

comment:16

Ok, yes, I agree with you.

• Sage crashing is very unfortunate behavior and it definitely shouldn't happen because some trivial python error occurs.
• Fix incorrect lazy imports at startup #31480 only fixed the symptoms. The point above applies there as well. But also tab completion should not even trigger loading lazy imports (unless your actually tab completing within the lazy imported thing and then of course it needs to grep the signature via the inspect module and maybe this needs to access source or something like this).

To verify that tab completion is doing too much one can do the following:

Start sage fresh

sage: type(polytopes._object)
<class 'NoneType'>

Hit tab after pol. polytopes isn't even your first guess, but it's on the list. Now repeat:

sage: type(polytopes._object)
<class 'sage.geometry.polyhedron.library.Polytopes'>

So tab completion does not work well with lazy imports at all.

[videlec]

comment:17

Certainly unrelated: On my machine, when I do a copy/paste of a sage prompt line I also get a bunch of white spaces in the buffer. The same happened in your examples from [comment:16] (I removed the spaces manually). This behavior was not there before.

[videlec]

comment:18

Replying to @kliem:

To verify that tab completion is doing too much one can do the following:

Start sage fresh

sage: type(polytopes._object)
<class 'NoneType'>

Hit tab after pol. polytopes isn't even your first guess, but it's on the list. Now repeat:

sage: type(polytopes._object)
<class 'sage.geometry.polyhedron.library.Polytopes'>

So tab completion does not work well with lazy imports at all.

For me it is even stranger: before the tab and and after the first I still get None. Only on the second run it gets lazily imported.

[kliem]

comment:19

Replying to @videlec:

Replying to @kliem:

To verify that tab completion is doing too much one can do the following:

Start sage fresh

sage: type(polytopes._object)
<class 'NoneType'>

Hit tab after pol. polytopes isn't even your first guess, but it's on the list. Now repeat:

sage: type(polytopes._object)
<class 'sage.geometry.polyhedron.library.Polytopes'>

So tab completion does not work well with lazy imports at all.

For me it is even stranger: before the tab and and after the first I still get None. Only on the second run it gets lazily imported.

Yes, you are right. I got this wrong. If you tab complete p twice, then everythings neatly lazy imported that starts with a p is being imported.

[mkoeppe]

comment:20

Note that tab completion runs in a separate thread, which previously caused trouble in #22704

[videlec]

comment:21

Not reproducible with

from IPython import get_ipython

ip = get_ipython()
c = ip.Completer
ip.ex("from sage.all import *")
for _ in range(4):
    print(c.all_completions("pol"))
    print(ip.ex('type(polytopes._object)'))

which prints four times

['polar_plot', 'polygen', 'polygens', 'polygon', 'polygon2d', 'polygon3d', 'polygon_spline', 'polygonal_number', 'polygons3d', 'polylog', 'polymake', 'polytopes']
None

[kliem]

comment:22

Replying to @mkoeppe:

Note that tab completion runs in a separate thread, which previously caused trouble in #22704

Ah, this probably explains, why it takes two times, until you see the effect.

You do this effect immediatly in memory usage. Tab completion of h takes something around 70 MB of memory. How stupid is that.

Also tab completion of l gives:

sage: l// Giac share root-directory:/home/jonathan/Applications/sage/local/share/giac/
// Giac share root-directory:/home/jonathan/Applications/sage/local/share/giac/
Help file /home/jonathan/Applications/sage/local/share/giac/doc/fr/aide_cas not found
Added 0 synonyms

And tab completion twice s has sage crash hard again. The essense of it being

AttributeError: module 'sage.coding.linear_code' has no attribute 'self_orthogonal_binary_codes'

because it is incorrectly lazily imported in src/sage/coding/all.py. Implying three things for me:

• Again, sage should not crash hard because of such a silly thing.
• We should have a doctest somewhere that unrolls all lazy imports, just to make sure that they actually work. Apparently this is not being tested.
• Of course it should be fixed in src/sage/coding/all.py.

[videlec]

comment:23

In the first place, tab completion should not trigger the lazy imports... Why is it doing so?

[kliem]

comment:24

I mentioned this above and agree. I don't know. For the moment (and all I'll do today), I just remarked that this is clearly wrong.

If you autocomplete within a lazy imported thing as in polytopes.cube(int (because you are too lazy to type the argument intervals), you need to fetch the signature, which will probably trigger an import. However, just an import of that one class is justified. Otherwise it should not trigger any import (it only needs the names after all).

[videlec]

comment:25

Concerning lazy imports, I think I got it. Since some recent Python3 versions, objects could be wrapped and in such case the object implements __wrapped__ (this is for example the case with some decorators in functools). Ipython tries to be smart with this with respect to tab completion and hence check for __wrapped__. Doing this on a lazy_import will trigger the import.

[videlec]

comment:26

Nope. That could not explain...

[videlec]

comment:27

With

diff --git a/src/sage/misc/lazy_import.pyx b/src/sage/misc/lazy_import.pyx
index 336567e22c..49abc0b849 100644
--- a/src/sage/misc/lazy_import.pyx
+++ b/src/sage/misc/lazy_import.pyx
@@ -325,6 +325,7 @@ cdef class LazyImport(object):
             sage: my_integer.sqrt is Integer.sqrt
             True
         """
+        print('getattr(attr={}) of self={}'.format(attr, self))
         return getattr(self.get_object(), attr)
 
     # We need to wrap all the slot methods, as they are not forwarded

you can have a look at what is asked to your lazy_import objects.

Here is what I got (up to desynchronization)

getattr(attr=Algebras) of self=<class 'sage.categories.groups.Groups'>
getattr(attr=Finite) of self=<class 'sage.categories.groups.Groups'>
getattr(attr=Finite) of self=<class 'sage.categories.groups.Groups'>
sage: pol<TAB>
getattr(attr=__module__) of self=<built-in function polygon_spline>
getattr(attr=__module__) of self=<built-in function polygon_spline>
getattr(attr=__module__) of self=<built-in function polygon_spline>
getattr(attr=__dict__) of self=<built-in function polygon_spline>
getattr(attr=__wrapped__) of self=<built-in function polygon_spline>
getattr(attr=__signature__) of self=<built-in function polygon_spline>
getattr(attr=_partialmethod) of self=<built-in function polygon_spline>
getattr(attr=__name__) of self=<built-in function polygon_spline>
getattr(attr=__code__) of self=<built-in function polygon_spline>
getattr(attr=__defaults__) of self=<built-in function polygon_spline>
getattr(attr=__kwdefaults__) of self=<built-in function polygon_spline>
getattr(attr=__annotations__) of self=<built-in function polygon_spline>
getattr(attr=__text_signature__) of self=<built-in function polygon_spline>
getattr(attr=__module__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__module__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__module__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__dict__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__wrapped__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__signature__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=_partialmethod) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__name__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__code__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__defaults__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__kwdefaults__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__annotations__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>
getattr(attr=__text_signature__) of self=<sage.geometry.polyhedron.library.Polytopes object at 0x7f21a0b12e20>

The three getattr(attr=XXX) of self=<class 'sage.categories.groups.Groups'> are definitely unrelated but worth looking at since they should not be there! I thought that resolution of lazy import was forbidden on startup.

[videlec]

comment:28

All right, it has to do with extension class. My [comment:25] was looking in the good direction: IPython does call stuff on the object in some cases.

sage: cython("""
....: cdef class A(object):
....:     def __getattr__(self, key):
....:         print('getattr(key={})'.format(key))
....:         raise AttributeError
....: """)
sage: a = A()
sage: my_custom_name = A()
sage: from IPython import get_ipython                                                                                                                                                          
sage: ip = get_ipython()                                                                                                                                                                       
sage: ip.Completer.all_completions("my_custo")                                                                                                                                                 
getattr(key=__module__)
getattr(key=__module__)
['my_custom_name']

[videlec]

comment:29

Tab completion is now #31643