old-cross-binutils/gdb/python/py-utils.c
Tom Tromey f870a310ee gdb
2010-03-05  Corinna Vinschen  <vinschen@redhat.com>
	    Tom Tromey  <tromey@redhat.com>

	* utils.c (host_char_to_target): Add 'gdbarch' argument.
	(parse_escape): Likewise.
	* python/py-utils.c (unicode_to_target_string): Update.
	(unicode_to_target_python_string): Update.
	(target_string_to_unicode): Update.
	* printcmd.c (printf_command): Update.
	* p-exp.y (yylex): Update.
	* objc-exp.y (yylex): Update.
	* mi/mi-parse.c: Include charset.h.
	(mi_parse_escape): New function.
	(mi_parse_argv): Use it.
	* jv-exp.y (yylex): Update.
	* i386-cygwin-tdep.c (i386_cygwin_auto_wide_charset): New
	function.
	(i386_cygwin_init_abi): Call set_gdbarch_auto_wide_charset.
	* gdbarch.sh (auto_charset, auto_wide_charset): New.
	* gdbarch.c: Rebuild.
	* gdbarch.h: Rebuild.
	* defs.h (parse_escape): Update.
	* cli/cli-setshow.c: Include arch-utils.h.
	(do_setshow_command): Update.
	* cli/cli-cmds.c (echo_command): Update.
	* charset.h (target_charset, target_wide_charset): Update.
	* charset.c: Include arch-utils.h.
	(target_charset_name): Default to "auto".
	(target_wide_charset_name): Likewise.
	(show_target_charset_name): Handle "auto".
	(show_target_wide_charset_name): Likewise.
	(be_le_arch): New global.
	(set_be_le_names): Add 'gdbarch' argument.
	(validate): Likewise.  Don't call set_be_le_names.
	(set_charset_sfunc, set_host_charset_sfunc)
	(set_target_charset_sfunc, set_target_wide_charset_sfunc):
	Update.
	(target_charset): Add 'gdbarch' argument.
	(target_wide_charset): Likewise.  Remove 'byte_order' argument.
	(auto_target_charset_name): New global.
	(default_auto_charset, default_auto_wide_charset): New functions.
	(_initialize_charset): Set auto_target_charset_name.  Allow "auto"
	for target charsets.  Copy result of nl_langinfo.  Use GetACP if
	USE_WIN32API.
	* c-lang.c (charset_for_string_type): Add 'gdbarch' argument,
	remove 'byte_order' argument.  Update.
	(classify_type): Likewise.
	(c_emit_char): Update.
	(c_printchar): Update.
	(c_printstr): Update.
	(c_get_string): Update.
	(evaluate_subexp_c): Update.
	* arch-utils.h (default_auto_charset, default_auto_wide_charset):
	Declare.
	* python/python.c (gdbpy_target_charset): New function.
	(gdbpy_target_wide_charset): Likewise.
	(GdbMethods): Update.
	* NEWS: Update.
gdb/doc
	* gdb.texinfo (Basic Python): Document target_charset and
	target_wide_charset.
gdb/testsuite
	* gdb.python/py-prettyprint.py (pp_nullstr.to_string): Use
	gdb.target_charset.
	(pp_ns.to_string): Likewise.
2010-03-05 20:18:19 +00:00

223 lines
6.2 KiB
C

/* General utility routines for GDB/Python.
Copyright (C) 2008, 2009, 2010 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/>. */
#include "defs.h"
#include "charset.h"
#include "python-internal.h"
/* This is a cleanup function which decrements the refcount on a
Python object. */
static void
py_decref (void *p)
{
PyObject *py = p;
/* Note that we need the extra braces in this 'if' to avoid a
warning from gcc. */
if (py)
{
Py_DECREF (py);
}
}
/* Return a new cleanup which will decrement the Python object's
refcount when run. */
struct cleanup *
make_cleanup_py_decref (PyObject *py)
{
return make_cleanup (py_decref, (void *) py);
}
/* Converts a Python 8-bit string to a unicode string object. Assumes the
8-bit string is in the host charset. If an error occurs during conversion,
returns NULL with a python exception set.
As an added bonus, the functions accepts a unicode string and returns it
right away, so callers don't need to check which kind of string they've
got.
If the given object is not one of the mentioned string types, NULL is
returned, with the TypeError python exception set. */
PyObject *
python_string_to_unicode (PyObject *obj)
{
PyObject *unicode_str;
/* If obj is already a unicode string, just return it.
I wish life was always that simple... */
if (PyUnicode_Check (obj))
{
unicode_str = obj;
Py_INCREF (obj);
}
else if (PyString_Check (obj))
unicode_str = PyUnicode_FromEncodedObject (obj, host_charset (), NULL);
else
{
PyErr_SetString (PyExc_TypeError,
_("Expected a string or unicode object."));
unicode_str = NULL;
}
return unicode_str;
}
/* Returns a newly allocated string with the contents of the given unicode
string object converted to CHARSET. If an error occurs during the
conversion, NULL will be returned and a python exception will be set.
The caller is responsible for xfree'ing the string. */
static char *
unicode_to_encoded_string (PyObject *unicode_str, const char *charset)
{
char *result;
PyObject *string;
/* Translate string to named charset. */
string = PyUnicode_AsEncodedString (unicode_str, charset, NULL);
if (string == NULL)
return NULL;
result = xstrdup (PyString_AsString (string));
Py_DECREF (string);
return result;
}
/* Returns a PyObject with the contents of the given unicode string
object converted to a named charset. If an error occurs during
the conversion, NULL will be returned and a python exception will
be set. */
static PyObject *
unicode_to_encoded_python_string (PyObject *unicode_str, const char *charset)
{
PyObject *string;
/* Translate string to named charset. */
string = PyUnicode_AsEncodedString (unicode_str, charset, NULL);
if (string == NULL)
return NULL;
return string;
}
/* Returns a newly allocated string with the contents of the given unicode
string object converted to the target's charset. If an error occurs during
the conversion, NULL will be returned and a python exception will be set.
The caller is responsible for xfree'ing the string. */
char *
unicode_to_target_string (PyObject *unicode_str)
{
return unicode_to_encoded_string (unicode_str,
target_charset (python_gdbarch));
}
/* Returns a PyObject with the contents of the given unicode string
object converted to the target's charset. If an error occurs
during the conversion, NULL will be returned and a python exception
will be set. */
PyObject *
unicode_to_target_python_string (PyObject *unicode_str)
{
return unicode_to_encoded_python_string (unicode_str,
target_charset (python_gdbarch));
}
/* Converts a python string (8-bit or unicode) to a target string in
the target's charset. Returns NULL on error, with a python exception set.
The caller is responsible for xfree'ing the string. */
char *
python_string_to_target_string (PyObject *obj)
{
PyObject *str;
char *result;
str = python_string_to_unicode (obj);
if (str == NULL)
return NULL;
result = unicode_to_target_string (str);
Py_DECREF (str);
return result;
}
/* Converts a python string (8-bit or unicode) to a target string in the
target's charset. Returns NULL on error, with a python exception
set. */
PyObject *
python_string_to_target_python_string (PyObject *obj)
{
PyObject *str;
PyObject *result;
str = python_string_to_unicode (obj);
if (str == NULL)
return NULL;
result = unicode_to_target_python_string (str);
Py_DECREF (str);
return result;
}
/* Converts a python string (8-bit or unicode) to a target string in
the host's charset. Returns NULL on error, with a python exception set.
The caller is responsible for xfree'ing the string. */
char *
python_string_to_host_string (PyObject *obj)
{
PyObject *str;
char *result;
str = python_string_to_unicode (obj);
if (str == NULL)
return NULL;
result = unicode_to_encoded_string (str, host_charset ());
Py_DECREF (str);
return result;
}
/* Converts a target string of LENGTH bytes in the target's charset to a
Python Unicode string. If LENGTH is -1, convert until a null byte is found.
Returns NULL on error, with a python exception set. */
PyObject *
target_string_to_unicode (const gdb_byte *str, int length)
{
if (length == -1)
length = strlen (str);
return PyUnicode_Decode (str, length, target_charset (python_gdbarch), NULL);
}
/* Return true if OBJ is a Python string or unicode object, false
otherwise. */
int
gdbpy_is_string (PyObject *obj)
{
return PyString_Check (obj) || PyUnicode_Check (obj);
}