Re: [PATCH v2 3/5] Create MI commands using python.

[Simon Marchi <simark@simark.ca>]

On 2019-05-14 7:24 a.m., Jan Vrany wrote:
> From: Didier Nadeau <didier.nadeau@gmail.com>
> 
> This commit allows an user to create custom MI commands using Python
> similarly to what is possible for Python CLI commands.
> 
> A new subclass of mi_command is defined for Python MI commands,
> mi_command_py. A new file, py-micmd.c contains the logic for Python
> MI commands.
> 
> gdb/ChangeLog
> 
>     * Makefile.in (SUBDIR_PYTHON_SRCS): Add py-micmd.c.
>     * mi/mi-cmds.c (insert_mi_cmd_entry): Remove static.
>     (mi_cmd_table): Remove static.
>     * mi/mi-cmds.h (insert_mi_cmd_entry): New declaration.
>     (mi_cmd_table): New declaration.
>     * python/py-micmd.c: New file
>     (parse_mi_result): New function.
>     (micmdpy_init): New function.
>     (gdbpy_initialize_micommands): New function.
>     (mi_command_py): New class.
>     * python/py-micmd.h: New file
>     (micmdpy_object): New struct.
>     (micmdpy_object): New typedef.
>     (mi_command_py): New class.
>     * python/python-internal.h
>     (gdbpy_initialize_micommands): New declaration.
>     * python/python.c
>     (_initialize_python): New call to gdbpy_initialize_micommands.
>     (finalize_python): Finalize Python MI commands.
> ---
>  gdb/ChangeLog                |  23 +++
>  gdb/Makefile.in              |   1 +
>  gdb/mi/mi-cmds.c             |   7 +-
>  gdb/mi/mi-cmds.h             |   9 ++
>  gdb/python/py-micmd.c        | 300 +++++++++++++++++++++++++++++++++++
>  gdb/python/py-micmd.h        |  51 ++++++
>  gdb/python/python-internal.h |   2 +
>  gdb/python/python.c          |  13 +-
>  8 files changed, 401 insertions(+), 5 deletions(-)
>  create mode 100644 gdb/python/py-micmd.c
>  create mode 100644 gdb/python/py-micmd.h
> 
> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
> index 03eb42555e..4cbe97002b 100644
> --- a/gdb/ChangeLog
> +++ b/gdb/ChangeLog
> @@ -1,3 +1,26 @@
> +2019-05-02  Didier Nadeau  <didier.nadeau@gmail.com>
> +            Jan Vrany  <jan.vrany@fit.cvut.cz>
> +
> +	* Makefile.in (SUBDIR_PYTHON_SRCS): Add py-micmd.c.
> +	* mi/mi-cmds.c (insert_mi_cmd_entry): Remove static.
> +	(mi_cmd_table): Remove static.
> +	* mi/mi-cmds.h (insert_mi_cmd_entry): New declaration.
> +	(mi_cmd_table): New declaration.
> +	* python/py-micmd.c: New file
> +	(parse_mi_result): New function.
> +	(micmdpy_init): New function.
> +	(gdbpy_initialize_micommands): New function.
> +	(mi_command_py): New class.
> +	* python/py-micmd.h: New file
> +	(micmdpy_object): New struct.
> +	(micmdpy_object): New typedef.
> +	(mi_command_py): New class.
> +	* python/python-internal.h
> +	(gdbpy_initialize_micommands): New declaration.
> +	* python/python.c
> +	(_initialize_python): New call to gdbpy_initialize_micommands.
> +	(finalize_python): Finalize Python MI commands.
> +
>  2019-05-02  Didier Nadeau  <didier.nadeau@gmail.com>
>  	    Jan Vrany <jan.vrany@fit.cvut.cz>
>  
> diff --git a/gdb/Makefile.in b/gdb/Makefile.in
> index 0f49578360..28866962bc 100644
> --- a/gdb/Makefile.in
> +++ b/gdb/Makefile.in
> @@ -382,6 +382,7 @@ SUBDIR_PYTHON_SRCS = \
>  	python/py-instruction.c \
>  	python/py-lazy-string.c \
>  	python/py-linetable.c \
> +	python/py-micmd.c \
>  	python/py-newobjfileevent.c \
>  	python/py-objfile.c \
>  	python/py-param.c \
> diff --git a/gdb/mi/mi-cmds.c b/gdb/mi/mi-cmds.c
> index 44f3813fdc..7e1d2c3f2d 100644
> --- a/gdb/mi/mi-cmds.c
> +++ b/gdb/mi/mi-cmds.c
> @@ -28,12 +28,11 @@
>  
>  /* MI command table (built at run time). */
>  
> -static std::map<std::string, mi_cmd_up> mi_cmd_table;
> +std::map<std::string, mi_cmd_up> mi_cmd_table;
>  
> -/* Insert a new mi-command into the command table.  Return true if
> -   insertion was successful.  */
> +/* See mi-cmds.h.  */
>  
> -static bool
> +bool
>  insert_mi_cmd_entry (mi_cmd_up command)
>  {
>    gdb_assert (command != NULL);
> diff --git a/gdb/mi/mi-cmds.h b/gdb/mi/mi-cmds.h
> index fc432a16b9..a2757bae20 100644
> --- a/gdb/mi/mi-cmds.h
> +++ b/gdb/mi/mi-cmds.h
> @@ -22,6 +22,8 @@
>  #ifndef MI_MI_CMDS_H
>  #define MI_MI_CMDS_H
>  
> +#include <map>
> +
>  enum print_values {
>     PRINT_NO_VALUES,
>     PRINT_ALL_VALUES,
> @@ -190,9 +192,16 @@ typedef std::unique_ptr<mi_command> mi_cmd_up;
>  
>  extern mi_command *mi_cmd_lookup (const char *command);
>  
> +extern std::map<std::string, mi_cmd_up> mi_cmd_table;
> +
>  /* Debug flag */
>  extern int mi_debug_p;
>  
>  extern void mi_execute_command (const char *cmd, int from_tty);
>  
> +/* Insert a new mi-command into the command table.  Return true if
> +   insertion was successful.  */
> +
> +extern bool insert_mi_cmd_entry (mi_cmd_up command);
> +
>  #endif /* MI_MI_CMDS_H */
> diff --git a/gdb/python/py-micmd.c b/gdb/python/py-micmd.c
> new file mode 100644
> index 0000000000..c8d429bb4b
> --- /dev/null
> +++ b/gdb/python/py-micmd.c
> @@ -0,0 +1,300 @@
> +/* MI Command Set for GDB, the GNU debugger.
> +
> +   Copyright (C) 2019 Free Software Foundation, Inc.
> +
> +   This file is part of GDB.
> +
> +   This program is free software; you can redistribute it and/or modify
> +   it under the terms of the GNU General Public License as published by
> +   the Free Software Foundation; either version 3 of the License, or
> +   (at your option) any later version.
> +
> +   This program is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +   GNU General Public License for more details.
> +
> +   You should have received a copy of the GNU General Public License
> +   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
> +
> +/* gdb MI commands implemented in Python  */
> +
> +#include "defs.h"
> +#include "python-internal.h"
> +#include "python/py-micmd.h"
> +#include "arch-utils.h"
> +#include "charset.h"
> +#include "language.h"
> +
> +#include <string>
> +
> +static PyObject *invoke_cst;
> +
> +extern PyTypeObject
> +  micmdpy_object_type CPYCHECKER_TYPE_OBJECT_FOR_TYPEDEF ("micmdpy_object");
> +
> +/* If the command invoked returns a list, this function parses it and create an
> +   appropriate MI out output.
> +
> +   The returned values must be Python string, and can be contained within Python
> +   lists and dictionaries. It is possible to have a multiple levels of lists
> +   and/or dictionaries.  */
> +
> +static void
> +parse_mi_result (PyObject *result, const char *field_name)
> +{
> +  struct ui_out *uiout = current_uiout;
> +
> +  if (PyString_Check (result))
> +    {
> +      goto generic;

Hmm is this goto really necessary?  Can't you just remove this if, and execution will
go directly in the else?  Or is a string considered a sequence by PySequence_Check?

In any case, there's gotta be a better way to organize this that doesn't require the goto.

> +    }
> +  else if (PyDict_Check (result))
> +    {
> +      PyObject *key, *value;
> +      Py_ssize_t pos = 0;
> +      ui_out_emit_tuple tuple_emitter (uiout, field_name);
> +      while (PyDict_Next (result, &pos, &key, &value))
> +	{
> +	  gdb::unique_xmalloc_ptr<char> key_string (gdbpy_obj_to_string (key));

When parsing a dict, I think it would be good to enforce that keys are strings, I
don't really see the use case to allow arbitrary objects there.  In the MI, keys
will be strings, and I don't really see the use case of letting the user do

return {
  MyObject(): "foo"
}

where the str(MyObject()) will be used as the key.  By validating that keys are
strings, I think we can help the user catch errors earlier in development.

> +	  parse_mi_result (value, key_string.get ());
> +	}
> +    }
> +  else if (PySequence_Check (result))
> +    {
> +      PyObject *item;
> +      Py_ssize_t i = 0;

Declare these variables in the most specific scope possible (the for loop header for i,
in the for loop for item).

> +      ui_out_emit_list list_emitter (uiout, field_name);
> +      for (i = 0; i < PySequence_Size (result); ++i)
> +	{
> +	  item = PySequence_GetItem (result, i);

PySequence_GetItem returns a new reference, which must be dropped.

Also, you could probably use PySequence_ITEM which will be a bit faster.  It should be
safe, since we know result is a PySequence and we don't use negative indices.

> +	  parse_mi_result (item, NULL);
> +	}
> +    }
> +  else if (PyIter_Check (result))
> +    {
> +      gdbpy_ref<> item;
> +      ui_out_emit_list list_emitter (uiout, field_name);
> +      while (item.reset (PyIter_Next (result)), item != nullptr)
> +	{
> +	  parse_mi_result (item.get (), NULL);
> +	}

Remove extra curly braces.

> +    }
> +  else
> +    {
> +    generic:
> +      gdb::unique_xmalloc_ptr<char> string (gdbpy_obj_to_string (result));
> +      uiout->field_string (field_name, string.get ());
> +    }
> +}
> +
> +/* Object initializer; sets up gdb-side structures for MI command.
> +
> +   Use: __init__(NAME).
> +
> +   NAME is the name of the MI command to register.  It must start with a dash
> +   as a traditional MI commands do.  */

"as traditional MI commands", drop the "a"

> +
> +static int
> +micmdpy_init (PyObject *self, PyObject *args, PyObject *kw)
> +{
> +  const char *name;
> +
> +  if (!PyArg_ParseTuple (args, "s", &name))
> +    return -1;
> +
> +  /* Validate command name */
> +  const int name_len = strlen (name);
> +  if (name_len < 2)

This should probably name_len == 0, to match the error message.

> +    {
> +      PyErr_Format (PyExc_RuntimeError, _("MI command name is empty"));
> +      return -1;
> +    }
> +  else if ((name[0] != '-') || !isalnum (name[1]))
> +    {
> +      PyErr_Format (PyExc_RuntimeError,
> +		    _("MI command name does not start with '-'"
> +		      " followed by a letter or digit"));
> +      return -1;
> +    }
Could all these error messages be just calls to error(), handled by the catch below?

We would then raise some gdb.error exceptions instead of RuntimeError, maybe it's desirable?

> +  else
> +    for (int i = 2; i < name_len; i++)
> +      {
> +	if (!isalnum (name[i]) && name[i] != '-')
> +	  {
> +	    PyErr_Format (PyExc_RuntimeError,
> +			  _("MI command name contains invalid character: %c"),
> +			  name[i]);
> +	    return -1;
> +	  }
> +      }
> +
> +  gdbpy_ref<> self_ref = gdbpy_ref<>::new_reference (self);
> +  try
> +    {
> +      mi_command *micommand = new mi_command_py (name + 1, NULL, self_ref);
> +
> +      bool result = insert_mi_cmd_entry (std::move (mi_cmd_up (micommand)));

Make micommand a mi_cmd_up directly:

  mi_cmd_up micommand (new mi_command_py (...));

> +
> +      if (!result)
> +	{
> +	  PyErr_Format (
> +	    PyExc_RuntimeError,
> +	    _("Unable to insert command. The name might already be in use."));
> +	  return -1;
> +	}
> +    }
> +  catch (const gdb_exception &except)
> +    {
> +      GDB_PY_SET_HANDLE_EXCEPTION (except);
> +    }
> +
> +  return 0;
> +}
> +
> +mi_command_py::mi_command_py (const char *name, int *suppress_notification,
> +			      gdbpy_ref<> object)
> +  : mi_command (name, suppress_notification), pyobj (object)
> +{
> +}
> +
> +void
> +mi_command_py::invoke (struct mi_parse *parse)
> +{
> +  std::unique_ptr<scoped_restore_tmpl<int>> restore
> +    = do_suppress_notification ();
> +
> +  mi_parse_argv (parse->args, parse);
> +
> +  if (parse->argv == NULL)
> +    error (_("Problem parsing arguments: %s %s"), parse->command, parse->args);
> +
> +  PyObject *obj = this->pyobj.get ();
> +  int i;
> +
> +  gdbpy_enter enter_py (get_current_arch (), current_language);
> +
> +  if (!obj)
> +    error (_("-%s: invalid invocation of Python micommand object."),
> +	   name ().c_str ());

Can this condition happen because of a mistake by the user, or it would only be because
of a logic error in GDB?  If it's the latter, it should be a gdb_assert (obj != nullptr);

> +
> +  if (!PyObject_HasAttr (obj, invoke_cst))
> +    {
> +      error (_("-%s: python command object missing 'invoke' method."),
> +	     name ().c_str ());
> +    }

I would make that check (presence of invoke method) when the command is registered.  It would
help users catch mistakes earlier.   Here, it could just be a gdb_assert to confirm it exists.

> +
> +  gdbpy_ref<> argobj (PyList_New (parse->argc));
> +  if (argobj == nullptr)
> +    {
> +      gdbpy_print_stack ();
> +      error (_("-%s: failed to create the python arguments list."),
> +	     name ().c_str ());
> +    }

It would be nice to be consistent in how we write "Python" in our messages to the user.
I suggest "Python", with the capital letter, as it's how the language is named.

> +
> +  for (i = 0; i < parse->argc; ++i)

Declare i in the for loop header.

> +    {
> +      /* Since PyList_SetItem steals the reference, we don't use
> +       * gdbpy_ref<> to hold on arg string. */
> +      PyObject *str = PyUnicode_Decode (parse->argv[i], strlen (parse->argv[i]),
> +					host_charset (), NULL);
> +      if (PyList_SetItem (argobj.get (), i, str) != 0)

Actually, I would suggest using gdbpy_ref<>, but then releasing it:

  PyList_SetItem (..., str.release ());

It illustrates better the transfer of ownership, and there's no window where
the reference is unmanaged.

Just wondering, because I know that unicode handling is very different between
Python 2 and 3: did you test with both Python versions?

> +	{
> +	  error (_("-%s: failed to create the python arguments list."),
> +		 name ().c_str ());
> +	}
> +    }
> +
> +  gdb_assert (PyErr_Occurred () == NULL);
> +  gdbpy_ref<> result (
> +    PyObject_CallMethodObjArgs (obj, invoke_cst, argobj.get (), NULL));
> +  if (PyErr_Occurred () != NULL)
> +    {
> +      gdbpy_err_fetch ex;
> +      gdb::unique_xmalloc_ptr<char> ex_msg (ex.to_string ());
> +
> +      if (ex_msg == NULL || *ex_msg == '\0')
> +	error (_("-%s: failed to execute command"), name ().c_str ());
> +      else
> +	error (_("-%s: %s"), name ().c_str (), ex_msg.get ());
> +    }
> +  else if (result != nullptr)
> +    {
> +      if (Py_None != result)
> +	parse_mi_result (result.get (), "result");
> +    }
> +  else
> +    {
> +      error (
> +	_("-%s: command invoke() method returned NULL but no python exception "
> +	  "is set"),
> +	name ().c_str ());

I am curious, did you actually hit this case, where no Python exception was
raised and invoke returned NULL?  I thought the two were pretty coupled
(if return value is NULL, it means there's an exception raise and vice versa).

> +    }
> +}
> +
> +void mi_command_py::finalize ()
> +{
> +  this->pyobj.reset (nullptr);
> +}
> +
> +/* Initialize the MI command object.  */
> +
> +int
> +gdbpy_initialize_micommands ()
> +{
> +  micmdpy_object_type.tp_new = PyType_GenericNew;
> +  if (PyType_Ready (&micmdpy_object_type) < 0)
> +    return -1;
> +
> +  if (gdb_pymodule_addobject (gdb_module, "MICommand",
> +			      (PyObject *) &micmdpy_object_type)
> +      < 0)
> +    return -1;
> +
> +  invoke_cst = PyString_FromString ("invoke");
> +  if (invoke_cst == NULL)
> +    return -1;
> +
> +  return 0;
> +}
> +
> +static PyMethodDef micmdpy_object_methods[] = {{0}};
> +
> +PyTypeObject micmdpy_object_type = {
> +  PyVarObject_HEAD_INIT (NULL, 0) "gdb.MICommand", /*tp_name */
> +  sizeof (micmdpy_object),			   /*tp_basicsize */
> +  0,						   /*tp_itemsize */
> +  0,						   /*tp_dealloc */
> +  0,						   /*tp_print */
> +  0,						   /*tp_getattr */
> +  0,						   /*tp_setattr */
> +  0,						   /*tp_compare */
> +  0,						   /*tp_repr */
> +  0,						   /*tp_as_number */
> +  0,						   /*tp_as_sequence */
> +  0,						   /*tp_as_mapping */
> +  0,						   /*tp_hash */
> +  0,						   /*tp_call */
> +  0,						   /*tp_str */
> +  0,						   /*tp_getattro */
> +  0,						   /*tp_setattro */
> +  0,						   /*tp_as_buffer */
> +  Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE,	/*tp_flags */
> +  "GDB mi-command object",			   /* tp_doc */
> +  0,						   /* tp_traverse */
> +  0,						   /* tp_clear */
> +  0,						   /* tp_richcompare */
> +  0,						   /* tp_weaklistoffset */
> +  0,						   /* tp_iter */
> +  0,						   /* tp_iternext */
> +  micmdpy_object_methods,			   /* tp_methods */
> +  0,						   /* tp_members */
> +  0,						   /* tp_getset */
> +  0,						   /* tp_base */
> +  0,						   /* tp_dict */
> +  0,						   /* tp_descr_get */
> +  0,						   /* tp_descr_set */
> +  0,						   /* tp_dictoffset */
> +  micmdpy_init,					   /* tp_init */
> +  0,						   /* tp_alloc */
> +};
> diff --git a/gdb/python/py-micmd.h b/gdb/python/py-micmd.h
> new file mode 100644
> index 0000000000..418de41a10
> --- /dev/null
> +++ b/gdb/python/py-micmd.h
> @@ -0,0 +1,51 @@
> +/* MI Command Set for GDB, the GNU debugger.
> +
> +   Copyright (C) 2019 Free Software Foundation, Inc.
> +
> +   This file is part of GDB.
> +
> +   This program is free software; you can redistribute it and/or modify
> +   it under the terms of the GNU General Public License as published by
> +   the Free Software Foundation; either version 3 of the License, or
> +   (at your option) any later version.
> +
> +   This program is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +   GNU General Public License for more details.
> +
> +   You should have received a copy of the GNU General Public License
> +   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
> +
> +#ifndef PY_MICMDS_H
> +#define PY_MICMDS_H
> +
> +#include "mi/mi-cmds.h"
> +#include "mi/mi-parse.h"
> +#include "python-internal.h"
> +#include "python/py-ref.h"
> +
> +struct micmdpy_object
> +{
> +  PyObject_HEAD
> +};
> +
> +typedef struct micmdpy_object micmdpy_object;
> +
> +/* MI command implemented on top of a Python command.  */

I find this description a bit misleading, what's a Python command?
The only thing "Python command" can refer to is the gdb.Command class, use
to implement CLI commands in Python, but these Python MI commands are not
related.  I would suggest instead:

/* MI command implemented in Python.  */

> +class mi_command_py : public mi_command
> +{
> +public:
> +  mi_command_py (const char *name, int *suppress_notification,
> +		 gdbpy_ref<> object);

If the suppress_notification parameter is not used for Python commands, you can remove it.

Could you document the constructor?  The OBJECT parameter in particular is a bit opaque.

> +  void invoke (struct mi_parse *parse) override;

This should be private, like in the previous patch.

> +
> +  /* This is called just before shutting down a Python interpreter
> +     to release python object implementing the command */
> +  void finalize ();
> +
> +private:
> +  gdbpy_ref<> pyobj;
> +};
> +
> +#endif
> diff --git a/gdb/python/python-internal.h b/gdb/python/python-internal.h
> index 449926ca87..26606b4b36 100644
> --- a/gdb/python/python-internal.h
> +++ b/gdb/python/python-internal.h
> @@ -539,6 +539,8 @@ int gdbpy_initialize_xmethods (void)
>    CPYCHECKER_NEGATIVE_RESULT_SETS_EXCEPTION;
>  int gdbpy_initialize_unwind (void)
>    CPYCHECKER_NEGATIVE_RESULT_SETS_EXCEPTION;
> +int gdbpy_initialize_micommands (void)
> +  CPYCHECKER_NEGATIVE_RESULT_SETS_EXCEPTION;
>  
>  /* A wrapper for PyErr_Fetch that handles reference counting for the
>     caller.  */
> diff --git a/gdb/python/python.c b/gdb/python/python.c
> index 4dad8ec10d..ee89c603a5 100644
> --- a/gdb/python/python.c
> +++ b/gdb/python/python.c
> @@ -36,6 +36,8 @@
>  #include <ctype.h>
>  #include "location.h"
>  #include "ser-event.h"
> +#include "mi/mi-cmds.h"
> +#include "py-micmd.h"
>  
>  /* Declared constants and enum for python stack printing.  */
>  static const char python_excp_none[] = "none";
> @@ -1564,6 +1566,14 @@ finalize_python (void *ignore)
>    python_gdbarch = target_gdbarch ();
>    python_language = current_language;
>  
> +  for (auto const& name_and_cmd : mi_cmd_table)

Use:

  for (const auto &name_and_cmd : mi_cmd_table)

> +    {
> +      mi_command    *cmd = name_and_cmd.second.get ();
> +      mi_command_py *cmd_py = dynamic_cast<mi_command_py*> (cmd);

Don't align variables like this, use regular formatting.

> +      if (cmd_py != nullptr)
> +        cmd_py->finalize ();
> +    }
> +
>    Py_Finalize ();
>  
>    restore_active_ext_lang (previous_active);
> @@ -1698,7 +1708,8 @@ do_start_initialization ()
>        || gdbpy_initialize_event () < 0
>        || gdbpy_initialize_arch () < 0
>        || gdbpy_initialize_xmethods () < 0
> -      || gdbpy_initialize_unwind () < 0)
> +      || gdbpy_initialize_unwind () < 0
> +      || gdbpy_initialize_micommands () < 0)
>      return false;
>  
>  #define GDB_PY_DEFINE_EVENT_TYPE(name, py_name, doc, base)	\
> 

Simon