gcc/
[official-gcc.git] / gcc / gdbhooks.py
blob0af8ecd0056e619563ffd20bc7d7ca1b5df8f16e
1 # Python hooks for gdb for debugging GCC
2 # Copyright (C) 2013-2014 Free Software Foundation, Inc.
4 # Contributed by David Malcolm <dmalcolm@redhat.com>
6 # This file is part of GCC.
8 # GCC is free software; you can redistribute it and/or modify it under
9 # the terms of the GNU General Public License as published by the Free
10 # Software Foundation; either version 3, or (at your option) any later
11 # version.
13 # GCC is distributed in the hope that it will be useful, but WITHOUT
14 # ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 # FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 # for more details.
18 # You should have received a copy of the GNU General Public License
19 # along with GCC; see the file COPYING3. If not see
20 # <http://www.gnu.org/licenses/>.
22 """
23 Enabling the debugging hooks
24 ----------------------------
25 gcc/configure (from configure.ac) generates a .gdbinit within the "gcc"
26 subdirectory of the build directory, and when run by gdb, this imports
27 gcc/gdbhooks.py from the source directory, injecting useful Python code
28 into gdb.
30 You may see a message from gdb of the form:
31 "path-to-build/gcc/.gdbinit" auto-loading has been declined by your `auto-load safe-path'
32 as a protection against untrustworthy python scripts. See
33 http://sourceware.org/gdb/onlinedocs/gdb/Auto_002dloading-safe-path.html
35 The fix is to mark the paths of the build/gcc directory as trustworthy.
36 An easy way to do so is by adding the following to your ~/.gdbinit script:
37 add-auto-load-safe-path /absolute/path/to/build/gcc
38 for the build directories for your various checkouts of gcc.
40 If it's working, you should see the message:
41 Successfully loaded GDB hooks for GCC
42 as gdb starts up.
44 During development, I've been manually invoking the code in this way, as a
45 precanned way of printing a variety of different kinds of value:
47 gdb \
48 -ex "break expand_gimple_stmt" \
49 -ex "run" \
50 -ex "bt" \
51 --args \
52 ./cc1 foo.c -O3
54 Examples of output using the pretty-printers
55 --------------------------------------------
56 Pointer values are generally shown in the form:
57 <type address extra_info>
59 For example, an opt_pass* might appear as:
60 (gdb) p pass
61 $2 = <opt_pass* 0x188b600 "expand"(170)>
63 The name of the pass is given ("expand"), together with the
64 static_pass_number.
66 Note that you can dereference the pointer in the normal way:
67 (gdb) p *pass
68 $4 = {type = RTL_PASS, name = 0x120a312 "expand",
69 [etc, ...snipped...]
71 and you can suppress pretty-printers using /r (for "raw"):
72 (gdb) p /r pass
73 $3 = (opt_pass *) 0x188b600
75 Basic blocks are shown with their index in parentheses, apart from the
76 CFG's entry and exit blocks, which are given as "ENTRY" and "EXIT":
77 (gdb) p bb
78 $9 = <basic_block 0x7ffff041f1a0 (2)>
79 (gdb) p cfun->cfg->x_entry_block_ptr
80 $10 = <basic_block 0x7ffff041f0d0 (ENTRY)>
81 (gdb) p cfun->cfg->x_exit_block_ptr
82 $11 = <basic_block 0x7ffff041f138 (EXIT)>
84 CFG edges are shown with the src and dest blocks given in parentheses:
85 (gdb) p e
86 $1 = <edge 0x7ffff043f118 (ENTRY -> 6)>
88 Tree nodes are printed using Python code that emulates print_node_brief,
89 running in gdb, rather than in the inferior:
90 (gdb) p cfun->decl
91 $1 = <function_decl 0x7ffff0420b00 foo>
92 For usability, the type is printed first (e.g. "function_decl"), rather
93 than just "tree".
95 RTL expressions use a kludge: they are pretty-printed by injecting
96 calls into print-rtl.c into the inferior:
97 Value returned is $1 = (note 9 8 10 [bb 3] NOTE_INSN_BASIC_BLOCK)
98 (gdb) p $1
99 $2 = (note 9 8 10 [bb 3] NOTE_INSN_BASIC_BLOCK)
100 (gdb) p /r $1
101 $3 = (rtx_def *) 0x7ffff043e140
102 This won't work for coredumps, and probably in other circumstances, but
103 it's a quick way of getting lots of debuggability quickly.
105 Callgraph nodes are printed with the name of the function decl, if
106 available:
107 (gdb) frame 5
108 #5 0x00000000006c288a in expand_function (node=<cgraph_node* 0x7ffff0312720 "foo">) at ../../src/gcc/cgraphunit.c:1594
109 1594 execute_pass_list (g->get_passes ()->all_passes);
110 (gdb) p node
111 $1 = <cgraph_node* 0x7ffff0312720 "foo">
113 vec<> pointers are printed as the address followed by the elements in
114 braces. Here's a length 2 vec:
115 (gdb) p bb->preds
116 $18 = 0x7ffff0428b68 = {<edge 0x7ffff044d380 (3 -> 5)>, <edge 0x7ffff044d3b8 (4 -> 5)>}
118 and here's a length 1 vec:
119 (gdb) p bb->succs
120 $19 = 0x7ffff0428bb8 = {<edge 0x7ffff044d3f0 (5 -> EXIT)>}
122 You cannot yet use array notation [] to access the elements within the
123 vector: attempting to do so instead gives you the vec itself (for vec[0]),
124 or a (probably) invalid cast to vec<> for the memory after the vec (for
125 vec[1] onwards).
127 Instead (for now) you must access m_vecdata:
128 (gdb) p bb->preds->m_vecdata[0]
129 $20 = <edge 0x7ffff044d380 (3 -> 5)>
130 (gdb) p bb->preds->m_vecdata[1]
131 $21 = <edge 0x7ffff044d3b8 (4 -> 5)>
133 import os.path
134 import re
136 import gdb
137 import gdb.printing
138 import gdb.types
140 # Convert "enum tree_code" (tree.def and tree.h) to a dict:
141 tree_code_dict = gdb.types.make_enum_dict(gdb.lookup_type('enum tree_code'))
143 # ...and look up specific values for use later:
144 IDENTIFIER_NODE = tree_code_dict['IDENTIFIER_NODE']
145 TYPE_DECL = tree_code_dict['TYPE_DECL']
147 # Similarly for "enum tree_code_class" (tree.h):
148 tree_code_class_dict = gdb.types.make_enum_dict(gdb.lookup_type('enum tree_code_class'))
149 tcc_type = tree_code_class_dict['tcc_type']
150 tcc_declaration = tree_code_class_dict['tcc_declaration']
152 class Tree:
154 Wrapper around a gdb.Value for a tree, with various methods
155 corresponding to macros in gcc/tree.h
157 def __init__(self, gdbval):
158 self.gdbval = gdbval
160 def is_nonnull(self):
161 return long(self.gdbval)
163 def TREE_CODE(self):
165 Get gdb.Value corresponding to TREE_CODE (self)
166 as per:
167 #define TREE_CODE(NODE) ((enum tree_code) (NODE)->base.code)
169 return self.gdbval['base']['code']
171 def DECL_NAME(self):
173 Get Tree instance corresponding to DECL_NAME (self)
175 return Tree(self.gdbval['decl_minimal']['name'])
177 def TYPE_NAME(self):
179 Get Tree instance corresponding to result of TYPE_NAME (self)
181 return Tree(self.gdbval['type_common']['name'])
183 def IDENTIFIER_POINTER(self):
185 Get str correspoinding to result of IDENTIFIER_NODE (self)
187 return self.gdbval['identifier']['id']['str'].string()
189 class TreePrinter:
190 "Prints a tree"
192 def __init__ (self, gdbval):
193 self.gdbval = gdbval
194 self.node = Tree(gdbval)
196 def to_string (self):
197 # like gcc/print-tree.c:print_node_brief
198 # #define TREE_CODE(NODE) ((enum tree_code) (NODE)->base.code)
199 # tree_code_name[(int) TREE_CODE (node)])
200 if long(self.gdbval) == 0:
201 return '<tree 0x0>'
203 val_TREE_CODE = self.node.TREE_CODE()
205 # extern const enum tree_code_class tree_code_type[];
206 # #define TREE_CODE_CLASS(CODE) tree_code_type[(int) (CODE)]
208 val_tree_code_type = gdb.parse_and_eval('tree_code_type')
209 val_tclass = val_tree_code_type[val_TREE_CODE]
211 val_tree_code_name = gdb.parse_and_eval('tree_code_name')
212 val_code_name = val_tree_code_name[long(val_TREE_CODE)]
213 #print val_code_name.string()
215 result = '<%s 0x%x' % (val_code_name.string(), long(self.gdbval))
216 if long(val_tclass) == tcc_declaration:
217 tree_DECL_NAME = self.node.DECL_NAME()
218 if tree_DECL_NAME.is_nonnull():
219 result += ' %s' % tree_DECL_NAME.IDENTIFIER_POINTER()
220 else:
221 pass # TODO: labels etc
222 elif long(val_tclass) == tcc_type:
223 tree_TYPE_NAME = Tree(self.gdbval['type_common']['name'])
224 if tree_TYPE_NAME.is_nonnull():
225 if tree_TYPE_NAME.TREE_CODE() == IDENTIFIER_NODE:
226 result += ' %s' % tree_TYPE_NAME.IDENTIFIER_POINTER()
227 elif tree_TYPE_NAME.TREE_CODE() == TYPE_DECL:
228 if tree_TYPE_NAME.DECL_NAME().is_nonnull():
229 result += ' %s' % tree_TYPE_NAME.DECL_NAME().IDENTIFIER_POINTER()
230 if self.node.TREE_CODE() == IDENTIFIER_NODE:
231 result += ' %s' % self.node.IDENTIFIER_POINTER()
232 # etc
233 result += '>'
234 return result
236 ######################################################################
237 # Callgraph pretty-printers
238 ######################################################################
240 class CGraphNodePrinter:
241 def __init__(self, gdbval):
242 self.gdbval = gdbval
244 def to_string (self):
245 result = '<cgraph_node* 0x%x' % long(self.gdbval)
246 if long(self.gdbval):
247 # symtab_node::name calls lang_hooks.decl_printable_name
248 # default implementation (lhd_decl_printable_name) is:
249 # return IDENTIFIER_POINTER (DECL_NAME (decl));
250 tree_decl = Tree(self.gdbval['decl'])
251 result += ' "%s"' % tree_decl.DECL_NAME().IDENTIFIER_POINTER()
252 result += '>'
253 return result
255 ######################################################################
257 class GimplePrinter:
258 def __init__(self, gdbval):
259 self.gdbval = gdbval
261 def to_string (self):
262 if long(self.gdbval) == 0:
263 return '<gimple 0x0>'
264 val_gimple_code = self.gdbval['code']
265 val_gimple_code_name = gdb.parse_and_eval('gimple_code_name')
266 val_code_name = val_gimple_code_name[long(val_gimple_code)]
267 result = '<%s 0x%x' % (val_code_name.string(),
268 long(self.gdbval))
269 result += '>'
270 return result
272 ######################################################################
273 # CFG pretty-printers
274 ######################################################################
276 def bb_index_to_str(index):
277 if index == 0:
278 return 'ENTRY'
279 elif index == 1:
280 return 'EXIT'
281 else:
282 return '%i' % index
284 class BasicBlockPrinter:
285 def __init__(self, gdbval):
286 self.gdbval = gdbval
288 def to_string (self):
289 result = '<basic_block 0x%x' % long(self.gdbval)
290 if long(self.gdbval):
291 result += ' (%s)' % bb_index_to_str(long(self.gdbval['index']))
292 result += '>'
293 return result
295 class CfgEdgePrinter:
296 def __init__(self, gdbval):
297 self.gdbval = gdbval
299 def to_string (self):
300 result = '<edge 0x%x' % long(self.gdbval)
301 if long(self.gdbval):
302 src = bb_index_to_str(long(self.gdbval['src']['index']))
303 dest = bb_index_to_str(long(self.gdbval['dest']['index']))
304 result += ' (%s -> %s)' % (src, dest)
305 result += '>'
306 return result
308 ######################################################################
310 class Rtx:
311 def __init__(self, gdbval):
312 self.gdbval = gdbval
314 def GET_CODE(self):
315 return self.gdbval['code']
317 def GET_RTX_LENGTH(code):
318 val_rtx_length = gdb.parse_and_eval('rtx_length')
319 return long(val_rtx_length[code])
321 def GET_RTX_NAME(code):
322 val_rtx_name = gdb.parse_and_eval('rtx_name')
323 return val_rtx_name[code].string()
325 def GET_RTX_FORMAT(code):
326 val_rtx_format = gdb.parse_and_eval('rtx_format')
327 return val_rtx_format[code].string()
329 class RtxPrinter:
330 def __init__(self, gdbval):
331 self.gdbval = gdbval
332 self.rtx = Rtx(gdbval)
334 def to_string (self):
336 For now, a cheap kludge: invoke the inferior's print
337 function to get a string to use the user, and return an empty
338 string for gdb
340 # We use print_inline_rtx to avoid a trailing newline
341 gdb.execute('call print_inline_rtx (stderr, (const_rtx) %s, 0)'
342 % long(self.gdbval))
343 return ''
345 # or by hand; based on gcc/print-rtl.c:print_rtx
346 result = ('<rtx_def 0x%x'
347 % (long(self.gdbval)))
348 code = self.rtx.GET_CODE()
349 result += ' (%s' % GET_RTX_NAME(code)
350 format_ = GET_RTX_FORMAT(code)
351 for i in range(GET_RTX_LENGTH(code)):
352 print format_[i]
353 result += ')>'
354 return result
356 ######################################################################
358 class PassPrinter:
359 def __init__(self, gdbval):
360 self.gdbval = gdbval
362 def to_string (self):
363 result = '<opt_pass* 0x%x' % long(self.gdbval)
364 if long(self.gdbval):
365 result += (' "%s"(%i)'
366 % (self.gdbval['name'].string(),
367 long(self.gdbval['static_pass_number'])))
368 result += '>'
369 return result
371 ######################################################################
373 class VecPrinter:
374 # -ex "up" -ex "p bb->preds"
375 def __init__(self, gdbval):
376 self.gdbval = gdbval
378 def display_hint (self):
379 return 'array'
381 def to_string (self):
382 # A trivial implementation; prettyprinting the contents is done
383 # by gdb calling the "children" method below.
384 return '0x%x' % long(self.gdbval)
386 def children (self):
387 if long(self.gdbval) == 0:
388 return
389 m_vecpfx = self.gdbval['m_vecpfx']
390 m_num = m_vecpfx['m_num']
391 m_vecdata = self.gdbval['m_vecdata']
392 for i in range(m_num):
393 yield ('[%d]' % i, m_vecdata[i])
395 ######################################################################
397 # TODO:
398 # * hashtab
399 # * location_t
401 class GdbSubprinter(gdb.printing.SubPrettyPrinter):
402 def __init__(self, name, class_):
403 super(GdbSubprinter, self).__init__(name)
404 self.class_ = class_
406 def handles_type(self, str_type):
407 raise NotImplementedError
409 class GdbSubprinterTypeList(GdbSubprinter):
411 A GdbSubprinter that handles a specific set of types
413 def __init__(self, str_types, name, class_):
414 super(GdbSubprinterTypeList, self).__init__(name, class_)
415 self.str_types = frozenset(str_types)
417 def handles_type(self, str_type):
418 return str_type in self.str_types
420 class GdbSubprinterRegex(GdbSubprinter):
422 A GdbSubprinter that handles types that match a regex
424 def __init__(self, regex, name, class_):
425 super(GdbSubprinterRegex, self).__init__(name, class_)
426 self.regex = re.compile(regex)
428 def handles_type(self, str_type):
429 return self.regex.match(str_type)
431 class GdbPrettyPrinters(gdb.printing.PrettyPrinter):
432 def __init__(self, name):
433 super(GdbPrettyPrinters, self).__init__(name, [])
435 def add_printer_for_types(self, name, class_, types):
436 self.subprinters.append(GdbSubprinterTypeList(name, class_, types))
438 def add_printer_for_regex(self, name, class_, regex):
439 self.subprinters.append(GdbSubprinterRegex(name, class_, regex))
441 def __call__(self, gdbval):
442 type_ = gdbval.type.unqualified()
443 str_type = str(type_)
444 for printer in self.subprinters:
445 if printer.enabled and printer.handles_type(str_type):
446 return printer.class_(gdbval)
448 # Couldn't find a pretty printer (or it was disabled):
449 return None
452 def build_pretty_printer():
453 pp = GdbPrettyPrinters('gcc')
454 pp.add_printer_for_types(['tree'],
455 'tree', TreePrinter)
456 pp.add_printer_for_types(['cgraph_node *'],
457 'cgraph_node', CGraphNodePrinter)
458 pp.add_printer_for_types(['gimple', 'gimple_statement_base *'],
459 'gimple',
460 GimplePrinter)
461 pp.add_printer_for_types(['basic_block', 'basic_block_def *'],
462 'basic_block',
463 BasicBlockPrinter)
464 pp.add_printer_for_types(['edge', 'edge_def *'],
465 'edge',
466 CfgEdgePrinter)
467 pp.add_printer_for_types(['rtx_def *'], 'rtx_def', RtxPrinter)
468 pp.add_printer_for_types(['opt_pass *'], 'opt_pass', PassPrinter)
470 pp.add_printer_for_regex(r'vec<(\S+), (\S+), (\S+)> \*',
471 'vec',
472 VecPrinter)
474 return pp
476 gdb.printing.register_pretty_printer(
477 gdb.current_objfile(),
478 build_pretty_printer())
480 def find_gcc_source_dir():
481 # Use location of global "g" to locate the source tree
482 sym_g = gdb.lookup_global_symbol('g')
483 path = sym_g.symtab.filename # e.g. '../../src/gcc/context.h'
484 srcdir = os.path.split(path)[0] # e.g. '../../src/gcc'
485 return srcdir
487 class PassNames:
488 """Parse passes.def, gathering a list of pass class names"""
489 def __init__(self):
490 srcdir = find_gcc_source_dir()
491 self.names = []
492 with open(os.path.join(srcdir, 'passes.def')) as f:
493 for line in f:
494 m = re.match('\s*NEXT_PASS \((.+)\);', line)
495 if m:
496 self.names.append(m.group(1))
498 class BreakOnPass(gdb.Command):
500 A custom command for putting breakpoints on the execute hook of passes.
501 This is largely a workaround for issues with tab-completion in gdb when
502 setting breakpoints on methods on classes within anonymous namespaces.
504 Example of use: putting a breakpoint on "final"
505 (gdb) break-on-pass
506 Press <TAB>; it autocompletes to "pass_":
507 (gdb) break-on-pass pass_
508 Press <TAB>:
509 Display all 219 possibilities? (y or n)
510 Press "n"; then type "f":
511 (gdb) break-on-pass pass_f
512 Press <TAB> to autocomplete to pass classnames beginning with "pass_f":
513 pass_fast_rtl_dce pass_fold_builtins
514 pass_feedback_split_functions pass_forwprop
515 pass_final pass_fre
516 pass_fixup_cfg pass_free_cfg
517 Type "in<TAB>" to complete to "pass_final":
518 (gdb) break-on-pass pass_final
519 ...and hit <RETURN>:
520 Breakpoint 6 at 0x8396ba: file ../../src/gcc/final.c, line 4526.
521 ...and we have a breakpoint set; continue execution:
522 (gdb) cont
523 Continuing.
524 Breakpoint 6, (anonymous namespace)::pass_final::execute (this=0x17fb990) at ../../src/gcc/final.c:4526
525 4526 virtual unsigned int execute (function *) { return rest_of_handle_final (); }
527 def __init__(self):
528 gdb.Command.__init__(self, 'break-on-pass', gdb.COMMAND_BREAKPOINTS)
529 self.pass_names = None
531 def complete(self, text, word):
532 # Lazily load pass names:
533 if not self.pass_names:
534 self.pass_names = PassNames()
536 return [name
537 for name in sorted(self.pass_names.names)
538 if name.startswith(text)]
540 def invoke(self, arg, from_tty):
541 sym = '(anonymous namespace)::%s::execute' % arg
542 breakpoint = gdb.Breakpoint(sym)
544 BreakOnPass()
546 print('Successfully loaded GDB hooks for GCC')