1 /* DWARF index writing support for GDB.
3 Copyright (C) 1994-2021 Free Software Foundation, Inc.
5 This file is part of GDB.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>. */
22 #include "dwarf2/index-write.h"
25 #include "cli/cli-decode.h"
26 #include "gdbsupport/byte-vector.h"
27 #include "gdbsupport/filestuff.h"
28 #include "gdbsupport/gdb_unlinker.h"
29 #include "gdbsupport/pathstuff.h"
30 #include "gdbsupport/scoped_fd.h"
31 #include "complaints.h"
32 #include "dwarf2/index-common.h"
34 #include "dwarf2/read.h"
35 #include "dwarf2/dwz.h"
36 #include "gdb/gdb-index.h"
44 #include <forward_list>
46 #include <unordered_map>
47 #include <unordered_set>
49 /* Ensure only legit values are used. */
50 #define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \
52 gdb_assert ((unsigned int) (value) <= 1); \
53 GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \
56 /* Ensure only legit values are used. */
57 #define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \
59 gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \
60 && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \
61 GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \
64 /* Ensure we don't use more than the allotted number of bits for the CU. */
65 #define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \
67 gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \
68 GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \
71 /* The "save gdb-index" command. */
73 /* Write SIZE bytes from the buffer pointed to by DATA to FILE, with
77 file_write (FILE *file
, const void *data
, size_t size
)
79 if (fwrite (data
, 1, size
, file
) != size
)
80 error (_("couldn't data write to file"));
83 /* Write the contents of VEC to FILE, with error checking. */
85 template<typename Elem
, typename Alloc
>
87 file_write (FILE *file
, const std::vector
<Elem
, Alloc
> &vec
)
90 file_write (file
, vec
.data (), vec
.size () * sizeof (vec
[0]));
93 /* In-memory buffer to prepare data to be written later to a file. */
97 /* Copy DATA to the end of the buffer. */
99 void append_data (const T
&data
)
101 std::copy (reinterpret_cast<const gdb_byte
*> (&data
),
102 reinterpret_cast<const gdb_byte
*> (&data
+ 1),
103 grow (sizeof (data
)));
106 /* Copy CSTR (a zero-terminated string) to the end of buffer. The
107 terminating zero is appended too. */
108 void append_cstr0 (const char *cstr
)
110 const size_t size
= strlen (cstr
) + 1;
111 std::copy (cstr
, cstr
+ size
, grow (size
));
114 /* Store INPUT as ULEB128 to the end of buffer. */
115 void append_unsigned_leb128 (ULONGEST input
)
119 gdb_byte output
= input
& 0x7f;
123 append_data (output
);
129 /* Accept a host-format integer in VAL and append it to the buffer
130 as a target-format integer which is LEN bytes long. */
131 void append_uint (size_t len
, bfd_endian byte_order
, ULONGEST val
)
133 ::store_unsigned_integer (grow (len
), len
, byte_order
, val
);
136 /* Return the size of the buffer. */
139 return m_vec
.size ();
142 /* Return true iff the buffer is empty. */
145 return m_vec
.empty ();
148 /* Write the buffer to FILE. */
149 void file_write (FILE *file
) const
151 ::file_write (file
, m_vec
);
155 /* Grow SIZE bytes at the end of the buffer. Returns a pointer to
156 the start of the new block. */
157 gdb_byte
*grow (size_t size
)
159 m_vec
.resize (m_vec
.size () + size
);
160 return &*(m_vec
.end () - size
);
163 gdb::byte_vector m_vec
;
166 /* An entry in the symbol table. */
167 struct symtab_index_entry
169 /* The name of the symbol. */
171 /* The offset of the name in the constant pool. */
172 offset_type index_offset
;
173 /* A sorted vector of the indices of all the CUs that hold an object
175 std::vector
<offset_type
> cu_indices
;
178 /* The symbol table. This is a power-of-2-sized hash table. */
186 offset_type n_elements
= 0;
187 std::vector
<symtab_index_entry
> data
;
189 /* Temporary storage for Ada names. */
190 auto_obstack m_string_obstack
;
193 /* Find a slot in SYMTAB for the symbol NAME. Returns a reference to
196 Function is used only during write_hash_table so no index format backward
197 compatibility is needed. */
199 static symtab_index_entry
&
200 find_slot (struct mapped_symtab
*symtab
, const char *name
)
202 offset_type index
, step
, hash
= mapped_index_string_hash (INT_MAX
, name
);
204 index
= hash
& (symtab
->data
.size () - 1);
205 step
= ((hash
* 17) & (symtab
->data
.size () - 1)) | 1;
209 if (symtab
->data
[index
].name
== NULL
210 || strcmp (name
, symtab
->data
[index
].name
) == 0)
211 return symtab
->data
[index
];
212 index
= (index
+ step
) & (symtab
->data
.size () - 1);
216 /* Expand SYMTAB's hash table. */
219 hash_expand (struct mapped_symtab
*symtab
)
221 auto old_entries
= std::move (symtab
->data
);
223 symtab
->data
.clear ();
224 symtab
->data
.resize (old_entries
.size () * 2);
226 for (auto &it
: old_entries
)
229 auto &ref
= find_slot (symtab
, it
.name
);
230 ref
= std::move (it
);
234 /* Add an entry to SYMTAB. NAME is the name of the symbol.
235 CU_INDEX is the index of the CU in which the symbol appears.
236 IS_STATIC is one if the symbol is static, otherwise zero (global). */
239 add_index_entry (struct mapped_symtab
*symtab
, const char *name
,
240 int is_static
, gdb_index_symbol_kind kind
,
241 offset_type cu_index
)
243 offset_type cu_index_and_attrs
;
245 ++symtab
->n_elements
;
246 if (4 * symtab
->n_elements
/ 3 >= symtab
->data
.size ())
247 hash_expand (symtab
);
249 symtab_index_entry
&slot
= find_slot (symtab
, name
);
250 if (slot
.name
== NULL
)
253 /* index_offset is set later. */
256 cu_index_and_attrs
= 0;
257 DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs
, cu_index
);
258 DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs
, is_static
);
259 DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs
, kind
);
261 /* We don't want to record an index value twice as we want to avoid the
263 We process all global symbols and then all static symbols
264 (which would allow us to avoid the duplication by only having to check
265 the last entry pushed), but a symbol could have multiple kinds in one CU.
266 To keep things simple we don't worry about the duplication here and
267 sort and uniquify the list after we've processed all symbols. */
268 slot
.cu_indices
.push_back (cu_index_and_attrs
);
271 /* Sort and remove duplicates of all symbols' cu_indices lists. */
274 uniquify_cu_indices (struct mapped_symtab
*symtab
)
276 for (auto &entry
: symtab
->data
)
278 if (entry
.name
!= NULL
&& !entry
.cu_indices
.empty ())
280 auto &cu_indices
= entry
.cu_indices
;
281 std::sort (cu_indices
.begin (), cu_indices
.end ());
282 auto from
= std::unique (cu_indices
.begin (), cu_indices
.end ());
283 cu_indices
.erase (from
, cu_indices
.end ());
288 /* A form of 'const char *' suitable for container keys. Only the
289 pointer is stored. The strings themselves are compared, not the
294 c_str_view (const char *cstr
)
298 bool operator== (const c_str_view
&other
) const
300 return strcmp (m_cstr
, other
.m_cstr
) == 0;
303 /* Return the underlying C string. Note, the returned string is
304 only a reference with lifetime of this object. */
305 const char *c_str () const
311 friend class c_str_view_hasher
;
312 const char *const m_cstr
;
315 /* A std::unordered_map::hasher for c_str_view that uses the right
316 hash function for strings in a mapped index. */
317 class c_str_view_hasher
320 size_t operator () (const c_str_view
&x
) const
322 return mapped_index_string_hash (INT_MAX
, x
.m_cstr
);
326 /* A std::unordered_map::hasher for std::vector<>. */
331 size_t operator () (const std::vector
<T
> &key
) const
333 return iterative_hash (key
.data (),
334 sizeof (key
.front ()) * key
.size (), 0);
338 /* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with
339 constant pool entries going into the data buffer CPOOL. */
342 write_hash_table (mapped_symtab
*symtab
, data_buf
&output
, data_buf
&cpool
)
345 /* Elements are sorted vectors of the indices of all the CUs that
346 hold an object of this name. */
347 std::unordered_map
<std::vector
<offset_type
>, offset_type
,
348 vector_hasher
<offset_type
>>
351 /* We add all the index vectors to the constant pool first, to
352 ensure alignment is ok. */
353 for (symtab_index_entry
&entry
: symtab
->data
)
355 if (entry
.name
== NULL
)
357 gdb_assert (entry
.index_offset
== 0);
359 /* Finding before inserting is faster than always trying to
360 insert, because inserting always allocates a node, does the
361 lookup, and then destroys the new node if another node
362 already had the same key. C++17 try_emplace will avoid
365 = symbol_hash_table
.find (entry
.cu_indices
);
366 if (found
!= symbol_hash_table
.end ())
368 entry
.index_offset
= found
->second
;
372 symbol_hash_table
.emplace (entry
.cu_indices
, cpool
.size ());
373 entry
.index_offset
= cpool
.size ();
374 cpool
.append_data (MAYBE_SWAP (entry
.cu_indices
.size ()));
375 for (const auto index
: entry
.cu_indices
)
376 cpool
.append_data (MAYBE_SWAP (index
));
380 /* Now write out the hash table. */
381 std::unordered_map
<c_str_view
, offset_type
, c_str_view_hasher
> str_table
;
382 for (const auto &entry
: symtab
->data
)
384 offset_type str_off
, vec_off
;
386 if (entry
.name
!= NULL
)
388 const auto insertpair
= str_table
.emplace (entry
.name
, cpool
.size ());
389 if (insertpair
.second
)
390 cpool
.append_cstr0 (entry
.name
);
391 str_off
= insertpair
.first
->second
;
392 vec_off
= entry
.index_offset
;
396 /* While 0 is a valid constant pool index, it is not valid
397 to have 0 for both offsets. */
402 output
.append_data (MAYBE_SWAP (str_off
));
403 output
.append_data (MAYBE_SWAP (vec_off
));
407 typedef std::unordered_map
<partial_symtab
*, unsigned int> psym_index_map
;
409 /* Helper struct for building the address table. */
410 struct addrmap_index_data
412 addrmap_index_data (data_buf
&addr_vec_
, psym_index_map
&cu_index_htab_
)
413 : addr_vec (addr_vec_
), cu_index_htab (cu_index_htab_
)
416 struct objfile
*objfile
;
418 psym_index_map
&cu_index_htab
;
420 /* Non-zero if the previous_* fields are valid.
421 We can't write an entry until we see the next entry (since it is only then
422 that we know the end of the entry). */
424 /* Index of the CU in the table of all CUs in the index file. */
425 unsigned int previous_cu_index
;
426 /* Start address of the CU. */
427 CORE_ADDR previous_cu_start
;
430 /* Write an address entry to ADDR_VEC. */
433 add_address_entry (struct objfile
*objfile
, data_buf
&addr_vec
,
434 CORE_ADDR start
, CORE_ADDR end
, unsigned int cu_index
)
436 addr_vec
.append_uint (8, BFD_ENDIAN_LITTLE
, start
);
437 addr_vec
.append_uint (8, BFD_ENDIAN_LITTLE
, end
);
438 addr_vec
.append_data (MAYBE_SWAP (cu_index
));
441 /* Worker function for traversing an addrmap to build the address table. */
444 add_address_entry_worker (void *datap
, CORE_ADDR start_addr
, void *obj
)
446 struct addrmap_index_data
*data
= (struct addrmap_index_data
*) datap
;
447 partial_symtab
*pst
= (partial_symtab
*) obj
;
449 if (data
->previous_valid
)
450 add_address_entry (data
->objfile
, data
->addr_vec
,
451 data
->previous_cu_start
, start_addr
,
452 data
->previous_cu_index
);
454 data
->previous_cu_start
= start_addr
;
457 const auto it
= data
->cu_index_htab
.find (pst
);
458 gdb_assert (it
!= data
->cu_index_htab
.cend ());
459 data
->previous_cu_index
= it
->second
;
460 data
->previous_valid
= 1;
463 data
->previous_valid
= 0;
468 /* Write OBJFILE's address map to ADDR_VEC.
469 CU_INDEX_HTAB is used to map addrmap entries to their CU indices
470 in the index file. */
473 write_address_map (struct objfile
*objfile
, data_buf
&addr_vec
,
474 psym_index_map
&cu_index_htab
)
476 struct addrmap_index_data
addrmap_index_data (addr_vec
, cu_index_htab
);
478 /* When writing the address table, we have to cope with the fact that
479 the addrmap iterator only provides the start of a region; we have to
480 wait until the next invocation to get the start of the next region. */
482 addrmap_index_data
.objfile
= objfile
;
483 addrmap_index_data
.previous_valid
= 0;
485 addrmap_foreach (objfile
->partial_symtabs
->psymtabs_addrmap
,
486 add_address_entry_worker
, &addrmap_index_data
);
488 /* It's highly unlikely the last entry (end address = 0xff...ff)
489 is valid, but we should still handle it.
490 The end address is recorded as the start of the next region, but that
491 doesn't work here. To cope we pass 0xff...ff, this is a rare situation
493 if (addrmap_index_data
.previous_valid
)
494 add_address_entry (objfile
, addr_vec
,
495 addrmap_index_data
.previous_cu_start
, (CORE_ADDR
) -1,
496 addrmap_index_data
.previous_cu_index
);
499 /* Return the symbol kind of PSYM. */
501 static gdb_index_symbol_kind
502 symbol_kind (struct partial_symbol
*psym
)
504 domain_enum domain
= psym
->domain
;
505 enum address_class aclass
= psym
->aclass
;
513 return GDB_INDEX_SYMBOL_KIND_FUNCTION
;
515 return GDB_INDEX_SYMBOL_KIND_TYPE
;
517 case LOC_CONST_BYTES
:
518 case LOC_OPTIMIZED_OUT
:
520 return GDB_INDEX_SYMBOL_KIND_VARIABLE
;
522 /* Note: It's currently impossible to recognize psyms as enum values
523 short of reading the type info. For now punt. */
524 return GDB_INDEX_SYMBOL_KIND_VARIABLE
;
526 /* There are other LOC_FOO values that one might want to classify
527 as variables, but dwarf2read.c doesn't currently use them. */
528 return GDB_INDEX_SYMBOL_KIND_OTHER
;
531 return GDB_INDEX_SYMBOL_KIND_TYPE
;
533 return GDB_INDEX_SYMBOL_KIND_OTHER
;
537 /* Add a list of partial symbols to SYMTAB. */
540 write_psymbols (struct mapped_symtab
*symtab
,
541 std::unordered_set
<partial_symbol
*> &psyms_seen
,
542 const std::vector
<partial_symbol
*> &symbols
,
543 offset_type cu_index
,
546 for (partial_symbol
*psym
: symbols
)
548 const char *name
= psym
->ginfo
.search_name ();
550 if (psym
->ginfo
.language () == language_ada
)
552 /* We want to ensure that the Ada main function's name appears
553 verbatim in the index. However, this name will be of the
554 form "_ada_mumble", and will be rewritten by ada_decode.
555 So, recognize it specially here and add it to the index by
557 if (strcmp (main_name (), name
) == 0)
559 gdb_index_symbol_kind kind
= symbol_kind (psym
);
561 add_index_entry (symtab
, name
, is_static
, kind
, cu_index
);
564 /* In order for the index to work when read back into gdb, it
565 has to supply a funny form of the name: it should be the
566 encoded name, with any suffixes stripped. Using the
567 ordinary encoded name will not work properly with the
568 searching logic in find_name_components_bounds; nor will
569 using the decoded name. Furthermore, an Ada "verbatim"
570 name (of the form "<MumBle>") must be entered without the
571 angle brackets. Note that the current index is unusual,
572 see PR symtab/24820 for details. */
573 std::string decoded
= ada_decode (name
);
574 if (decoded
[0] == '<')
575 name
= (char *) obstack_copy0 (&symtab
->m_string_obstack
,
576 decoded
.c_str () + 1,
577 decoded
.length () - 2);
579 name
= obstack_strdup (&symtab
->m_string_obstack
,
580 ada_encode (decoded
.c_str ()));
583 /* Only add a given psymbol once. */
584 if (psyms_seen
.insert (psym
).second
)
586 gdb_index_symbol_kind kind
= symbol_kind (psym
);
588 add_index_entry (symtab
, name
, is_static
, kind
, cu_index
);
593 /* A helper struct used when iterating over debug_types. */
594 struct signatured_type_index_data
596 signatured_type_index_data (data_buf
&types_list_
,
597 std::unordered_set
<partial_symbol
*> &psyms_seen_
)
598 : types_list (types_list_
), psyms_seen (psyms_seen_
)
601 struct objfile
*objfile
;
602 struct mapped_symtab
*symtab
;
603 data_buf
&types_list
;
604 std::unordered_set
<partial_symbol
*> &psyms_seen
;
608 /* A helper function that writes a single signatured_type to an
612 write_one_signatured_type (void **slot
, void *d
)
614 struct signatured_type_index_data
*info
615 = (struct signatured_type_index_data
*) d
;
616 struct signatured_type
*entry
= (struct signatured_type
*) *slot
;
617 partial_symtab
*psymtab
= entry
->per_cu
.v
.psymtab
;
619 write_psymbols (info
->symtab
, info
->psyms_seen
,
620 psymtab
->global_psymbols
, info
->cu_index
,
622 write_psymbols (info
->symtab
, info
->psyms_seen
,
623 psymtab
->static_psymbols
, info
->cu_index
,
626 info
->types_list
.append_uint (8, BFD_ENDIAN_LITTLE
,
627 to_underlying (entry
->per_cu
.sect_off
));
628 info
->types_list
.append_uint (8, BFD_ENDIAN_LITTLE
,
629 to_underlying (entry
->type_offset_in_tu
));
630 info
->types_list
.append_uint (8, BFD_ENDIAN_LITTLE
, entry
->signature
);
637 /* Recurse into all "included" dependencies and count their symbols as
638 if they appeared in this psymtab. */
641 recursively_count_psymbols (partial_symtab
*psymtab
,
644 for (int i
= 0; i
< psymtab
->number_of_dependencies
; ++i
)
645 if (psymtab
->dependencies
[i
]->user
!= NULL
)
646 recursively_count_psymbols (psymtab
->dependencies
[i
],
649 psyms_seen
+= psymtab
->global_psymbols
.size ();
650 psyms_seen
+= psymtab
->static_psymbols
.size ();
653 /* Recurse into all "included" dependencies and write their symbols as
654 if they appeared in this psymtab. */
657 recursively_write_psymbols (struct objfile
*objfile
,
658 partial_symtab
*psymtab
,
659 struct mapped_symtab
*symtab
,
660 std::unordered_set
<partial_symbol
*> &psyms_seen
,
661 offset_type cu_index
)
665 for (i
= 0; i
< psymtab
->number_of_dependencies
; ++i
)
666 if (psymtab
->dependencies
[i
]->user
!= NULL
)
667 recursively_write_psymbols (objfile
,
668 psymtab
->dependencies
[i
],
669 symtab
, psyms_seen
, cu_index
);
671 write_psymbols (symtab
, psyms_seen
,
672 psymtab
->global_psymbols
, cu_index
,
674 write_psymbols (symtab
, psyms_seen
,
675 psymtab
->static_psymbols
, cu_index
,
679 /* DWARF-5 .debug_names builder. */
683 debug_names (dwarf2_per_objfile
*per_objfile
, bool is_dwarf64
,
684 bfd_endian dwarf5_byte_order
)
685 : m_dwarf5_byte_order (dwarf5_byte_order
),
686 m_dwarf32 (dwarf5_byte_order
),
687 m_dwarf64 (dwarf5_byte_order
),
689 ? static_cast<dwarf
&> (m_dwarf64
)
690 : static_cast<dwarf
&> (m_dwarf32
)),
691 m_name_table_string_offs (m_dwarf
.name_table_string_offs
),
692 m_name_table_entry_offs (m_dwarf
.name_table_entry_offs
),
693 m_debugstrlookup (per_objfile
)
696 int dwarf5_offset_size () const
698 const bool dwarf5_is_dwarf64
= &m_dwarf
== &m_dwarf64
;
699 return dwarf5_is_dwarf64
? 8 : 4;
702 /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */
703 enum class unit_kind
{ cu
, tu
};
705 /* Insert one symbol. */
706 void insert (const partial_symbol
*psym
, int cu_index
, bool is_static
,
709 const int dwarf_tag
= psymbol_tag (psym
);
712 const char *name
= psym
->ginfo
.search_name ();
714 if (psym
->ginfo
.language () == language_ada
)
716 /* We want to ensure that the Ada main function's name appears
717 verbatim in the index. However, this name will be of the
718 form "_ada_mumble", and will be rewritten by ada_decode.
719 So, recognize it specially here and add it to the index by
721 if (strcmp (main_name (), name
) == 0)
723 const auto insertpair
724 = m_name_to_value_set
.emplace (c_str_view (name
),
725 std::set
<symbol_value
> ());
726 std::set
<symbol_value
> &value_set
= insertpair
.first
->second
;
727 value_set
.emplace (symbol_value (dwarf_tag
, cu_index
, is_static
,
731 /* In order for the index to work when read back into gdb, it
732 has to supply a funny form of the name: it should be the
733 encoded name, with any suffixes stripped. Using the
734 ordinary encoded name will not work properly with the
735 searching logic in find_name_components_bounds; nor will
736 using the decoded name. Furthermore, an Ada "verbatim"
737 name (of the form "<MumBle>") must be entered without the
738 angle brackets. Note that the current index is unusual,
739 see PR symtab/24820 for details. */
740 std::string decoded
= ada_decode (name
);
741 if (decoded
[0] == '<')
742 name
= (char *) obstack_copy0 (&m_string_obstack
,
743 decoded
.c_str () + 1,
744 decoded
.length () - 2);
746 name
= obstack_strdup (&m_string_obstack
,
747 ada_encode (decoded
.c_str ()));
750 const auto insertpair
751 = m_name_to_value_set
.emplace (c_str_view (name
),
752 std::set
<symbol_value
> ());
753 std::set
<symbol_value
> &value_set
= insertpair
.first
->second
;
754 value_set
.emplace (symbol_value (dwarf_tag
, cu_index
, is_static
, kind
));
757 /* Build all the tables. All symbols must be already inserted.
758 This function does not call file_write, caller has to do it
762 /* Verify the build method has not be called twice. */
763 gdb_assert (m_abbrev_table
.empty ());
764 const size_t name_count
= m_name_to_value_set
.size ();
765 m_bucket_table
.resize
766 (std::pow (2, std::ceil (std::log2 (name_count
* 4 / 3))));
767 m_hash_table
.reserve (name_count
);
768 m_name_table_string_offs
.reserve (name_count
);
769 m_name_table_entry_offs
.reserve (name_count
);
771 /* Map each hash of symbol to its name and value. */
775 decltype (m_name_to_value_set
)::const_iterator it
;
777 std::vector
<std::forward_list
<hash_it_pair
>> bucket_hash
;
778 bucket_hash
.resize (m_bucket_table
.size ());
779 for (decltype (m_name_to_value_set
)::const_iterator it
780 = m_name_to_value_set
.cbegin ();
781 it
!= m_name_to_value_set
.cend ();
784 const char *const name
= it
->first
.c_str ();
785 const uint32_t hash
= dwarf5_djb_hash (name
);
786 hash_it_pair hashitpair
;
787 hashitpair
.hash
= hash
;
789 auto &slot
= bucket_hash
[hash
% bucket_hash
.size()];
790 slot
.push_front (std::move (hashitpair
));
792 for (size_t bucket_ix
= 0; bucket_ix
< bucket_hash
.size (); ++bucket_ix
)
794 const std::forward_list
<hash_it_pair
> &hashitlist
795 = bucket_hash
[bucket_ix
];
796 if (hashitlist
.empty ())
798 uint32_t &bucket_slot
= m_bucket_table
[bucket_ix
];
799 /* The hashes array is indexed starting at 1. */
800 store_unsigned_integer (reinterpret_cast<gdb_byte
*> (&bucket_slot
),
801 sizeof (bucket_slot
), m_dwarf5_byte_order
,
802 m_hash_table
.size () + 1);
803 for (const hash_it_pair
&hashitpair
: hashitlist
)
805 m_hash_table
.push_back (0);
806 store_unsigned_integer (reinterpret_cast<gdb_byte
*>
807 (&m_hash_table
.back ()),
808 sizeof (m_hash_table
.back ()),
809 m_dwarf5_byte_order
, hashitpair
.hash
);
810 const c_str_view
&name
= hashitpair
.it
->first
;
811 const std::set
<symbol_value
> &value_set
= hashitpair
.it
->second
;
812 m_name_table_string_offs
.push_back_reorder
813 (m_debugstrlookup
.lookup (name
.c_str ()));
814 m_name_table_entry_offs
.push_back_reorder (m_entry_pool
.size ());
815 gdb_assert (!value_set
.empty ());
816 for (const symbol_value
&value
: value_set
)
818 int &idx
= m_indexkey_to_idx
[index_key (value
.dwarf_tag
,
824 m_abbrev_table
.append_unsigned_leb128 (idx
);
825 m_abbrev_table
.append_unsigned_leb128 (value
.dwarf_tag
);
826 m_abbrev_table
.append_unsigned_leb128
827 (value
.kind
== unit_kind::cu
? DW_IDX_compile_unit
829 m_abbrev_table
.append_unsigned_leb128 (DW_FORM_udata
);
830 m_abbrev_table
.append_unsigned_leb128 (value
.is_static
831 ? DW_IDX_GNU_internal
832 : DW_IDX_GNU_external
);
833 m_abbrev_table
.append_unsigned_leb128 (DW_FORM_flag_present
);
835 /* Terminate attributes list. */
836 m_abbrev_table
.append_unsigned_leb128 (0);
837 m_abbrev_table
.append_unsigned_leb128 (0);
840 m_entry_pool
.append_unsigned_leb128 (idx
);
841 m_entry_pool
.append_unsigned_leb128 (value
.cu_index
);
844 /* Terminate the list of CUs. */
845 m_entry_pool
.append_unsigned_leb128 (0);
848 gdb_assert (m_hash_table
.size () == name_count
);
850 /* Terminate tags list. */
851 m_abbrev_table
.append_unsigned_leb128 (0);
854 /* Return .debug_names bucket count. This must be called only after
855 calling the build method. */
856 uint32_t bucket_count () const
858 /* Verify the build method has been already called. */
859 gdb_assert (!m_abbrev_table
.empty ());
860 const uint32_t retval
= m_bucket_table
.size ();
862 /* Check for overflow. */
863 gdb_assert (retval
== m_bucket_table
.size ());
867 /* Return .debug_names names count. This must be called only after
868 calling the build method. */
869 uint32_t name_count () const
871 /* Verify the build method has been already called. */
872 gdb_assert (!m_abbrev_table
.empty ());
873 const uint32_t retval
= m_hash_table
.size ();
875 /* Check for overflow. */
876 gdb_assert (retval
== m_hash_table
.size ());
880 /* Return number of bytes of .debug_names abbreviation table. This
881 must be called only after calling the build method. */
882 uint32_t abbrev_table_bytes () const
884 gdb_assert (!m_abbrev_table
.empty ());
885 return m_abbrev_table
.size ();
888 /* Recurse into all "included" dependencies and store their symbols
889 as if they appeared in this psymtab. */
890 void recursively_write_psymbols
891 (struct objfile
*objfile
,
892 partial_symtab
*psymtab
,
893 std::unordered_set
<partial_symbol
*> &psyms_seen
,
896 for (int i
= 0; i
< psymtab
->number_of_dependencies
; ++i
)
897 if (psymtab
->dependencies
[i
]->user
!= NULL
)
898 recursively_write_psymbols
899 (objfile
, psymtab
->dependencies
[i
], psyms_seen
, cu_index
);
901 write_psymbols (psyms_seen
, psymtab
->global_psymbols
,
902 cu_index
, false, unit_kind::cu
);
903 write_psymbols (psyms_seen
, psymtab
->static_psymbols
,
904 cu_index
, true, unit_kind::cu
);
907 /* Return number of bytes the .debug_names section will have. This
908 must be called only after calling the build method. */
909 size_t bytes () const
911 /* Verify the build method has been already called. */
912 gdb_assert (!m_abbrev_table
.empty ());
913 size_t expected_bytes
= 0;
914 expected_bytes
+= m_bucket_table
.size () * sizeof (m_bucket_table
[0]);
915 expected_bytes
+= m_hash_table
.size () * sizeof (m_hash_table
[0]);
916 expected_bytes
+= m_name_table_string_offs
.bytes ();
917 expected_bytes
+= m_name_table_entry_offs
.bytes ();
918 expected_bytes
+= m_abbrev_table
.size ();
919 expected_bytes
+= m_entry_pool
.size ();
920 return expected_bytes
;
923 /* Write .debug_names to FILE_NAMES and .debug_str addition to
924 FILE_STR. This must be called only after calling the build
926 void file_write (FILE *file_names
, FILE *file_str
) const
928 /* Verify the build method has been already called. */
929 gdb_assert (!m_abbrev_table
.empty ());
930 ::file_write (file_names
, m_bucket_table
);
931 ::file_write (file_names
, m_hash_table
);
932 m_name_table_string_offs
.file_write (file_names
);
933 m_name_table_entry_offs
.file_write (file_names
);
934 m_abbrev_table
.file_write (file_names
);
935 m_entry_pool
.file_write (file_names
);
936 m_debugstrlookup
.file_write (file_str
);
939 /* A helper user data for write_one_signatured_type. */
940 class write_one_signatured_type_data
943 write_one_signatured_type_data (debug_names
&nametable_
,
944 signatured_type_index_data
&&info_
)
945 : nametable (nametable_
), info (std::move (info_
))
947 debug_names
&nametable
;
948 struct signatured_type_index_data info
;
951 /* A helper function to pass write_one_signatured_type to
952 htab_traverse_noresize. */
954 write_one_signatured_type (void **slot
, void *d
)
956 write_one_signatured_type_data
*data
= (write_one_signatured_type_data
*) d
;
957 struct signatured_type_index_data
*info
= &data
->info
;
958 struct signatured_type
*entry
= (struct signatured_type
*) *slot
;
960 data
->nametable
.write_one_signatured_type (entry
, info
);
967 /* Storage for symbol names mapping them to their .debug_str section
969 class debug_str_lookup
973 /* Object constructor to be called for current DWARF2_PER_OBJFILE.
974 All .debug_str section strings are automatically stored. */
975 debug_str_lookup (dwarf2_per_objfile
*per_objfile
)
976 : m_abfd (per_objfile
->objfile
->obfd
),
977 m_per_objfile (per_objfile
)
979 per_objfile
->per_bfd
->str
.read (per_objfile
->objfile
);
980 if (per_objfile
->per_bfd
->str
.buffer
== NULL
)
982 for (const gdb_byte
*data
= per_objfile
->per_bfd
->str
.buffer
;
983 data
< (per_objfile
->per_bfd
->str
.buffer
984 + per_objfile
->per_bfd
->str
.size
);)
986 const char *const s
= reinterpret_cast<const char *> (data
);
987 const auto insertpair
988 = m_str_table
.emplace (c_str_view (s
),
989 data
- per_objfile
->per_bfd
->str
.buffer
);
990 if (!insertpair
.second
)
991 complaint (_("Duplicate string \"%s\" in "
992 ".debug_str section [in module %s]"),
993 s
, bfd_get_filename (m_abfd
));
994 data
+= strlen (s
) + 1;
998 /* Return offset of symbol name S in the .debug_str section. Add
999 such symbol to the section's end if it does not exist there
1001 size_t lookup (const char *s
)
1003 const auto it
= m_str_table
.find (c_str_view (s
));
1004 if (it
!= m_str_table
.end ())
1006 const size_t offset
= (m_per_objfile
->per_bfd
->str
.size
1007 + m_str_add_buf
.size ());
1008 m_str_table
.emplace (c_str_view (s
), offset
);
1009 m_str_add_buf
.append_cstr0 (s
);
1013 /* Append the end of the .debug_str section to FILE. */
1014 void file_write (FILE *file
) const
1016 m_str_add_buf
.file_write (file
);
1020 std::unordered_map
<c_str_view
, size_t, c_str_view_hasher
> m_str_table
;
1022 dwarf2_per_objfile
*m_per_objfile
;
1024 /* Data to add at the end of .debug_str for new needed symbol names. */
1025 data_buf m_str_add_buf
;
1028 /* Container to map used DWARF tags to their .debug_names abbreviation
1033 index_key (int dwarf_tag_
, bool is_static_
, unit_kind kind_
)
1034 : dwarf_tag (dwarf_tag_
), is_static (is_static_
), kind (kind_
)
1039 operator== (const index_key
&other
) const
1041 return (dwarf_tag
== other
.dwarf_tag
&& is_static
== other
.is_static
1042 && kind
== other
.kind
);
1045 const int dwarf_tag
;
1046 const bool is_static
;
1047 const unit_kind kind
;
1050 /* Provide std::unordered_map::hasher for index_key. */
1051 class index_key_hasher
1055 operator () (const index_key
&key
) const
1057 return (std::hash
<int>() (key
.dwarf_tag
) << 1) | key
.is_static
;
1061 /* Parameters of one symbol entry. */
1065 const int dwarf_tag
, cu_index
;
1066 const bool is_static
;
1067 const unit_kind kind
;
1069 symbol_value (int dwarf_tag_
, int cu_index_
, bool is_static_
,
1071 : dwarf_tag (dwarf_tag_
), cu_index (cu_index_
), is_static (is_static_
),
1076 operator< (const symbol_value
&other
) const
1096 /* Abstract base class to unify DWARF-32 and DWARF-64 name table
1101 const bfd_endian dwarf5_byte_order
;
1103 explicit offset_vec (bfd_endian dwarf5_byte_order_
)
1104 : dwarf5_byte_order (dwarf5_byte_order_
)
1107 /* Call std::vector::reserve for NELEM elements. */
1108 virtual void reserve (size_t nelem
) = 0;
1110 /* Call std::vector::push_back with store_unsigned_integer byte
1111 reordering for ELEM. */
1112 virtual void push_back_reorder (size_t elem
) = 0;
1114 /* Return expected output size in bytes. */
1115 virtual size_t bytes () const = 0;
1117 /* Write name table to FILE. */
1118 virtual void file_write (FILE *file
) const = 0;
1121 /* Template to unify DWARF-32 and DWARF-64 output. */
1122 template<typename OffsetSize
>
1123 class offset_vec_tmpl
: public offset_vec
1126 explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_
)
1127 : offset_vec (dwarf5_byte_order_
)
1130 /* Implement offset_vec::reserve. */
1131 void reserve (size_t nelem
) override
1133 m_vec
.reserve (nelem
);
1136 /* Implement offset_vec::push_back_reorder. */
1137 void push_back_reorder (size_t elem
) override
1139 m_vec
.push_back (elem
);
1140 /* Check for overflow. */
1141 gdb_assert (m_vec
.back () == elem
);
1142 store_unsigned_integer (reinterpret_cast<gdb_byte
*> (&m_vec
.back ()),
1143 sizeof (m_vec
.back ()), dwarf5_byte_order
, elem
);
1146 /* Implement offset_vec::bytes. */
1147 size_t bytes () const override
1149 return m_vec
.size () * sizeof (m_vec
[0]);
1152 /* Implement offset_vec::file_write. */
1153 void file_write (FILE *file
) const override
1155 ::file_write (file
, m_vec
);
1159 std::vector
<OffsetSize
> m_vec
;
1162 /* Base class to unify DWARF-32 and DWARF-64 .debug_names output
1163 respecting name table width. */
1167 offset_vec
&name_table_string_offs
, &name_table_entry_offs
;
1169 dwarf (offset_vec
&name_table_string_offs_
,
1170 offset_vec
&name_table_entry_offs_
)
1171 : name_table_string_offs (name_table_string_offs_
),
1172 name_table_entry_offs (name_table_entry_offs_
)
1177 /* Template to unify DWARF-32 and DWARF-64 .debug_names output
1178 respecting name table width. */
1179 template<typename OffsetSize
>
1180 class dwarf_tmpl
: public dwarf
1183 explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_
)
1184 : dwarf (m_name_table_string_offs
, m_name_table_entry_offs
),
1185 m_name_table_string_offs (dwarf5_byte_order_
),
1186 m_name_table_entry_offs (dwarf5_byte_order_
)
1190 offset_vec_tmpl
<OffsetSize
> m_name_table_string_offs
;
1191 offset_vec_tmpl
<OffsetSize
> m_name_table_entry_offs
;
1194 /* Try to reconstruct original DWARF tag for given partial_symbol.
1195 This function is not DWARF-5 compliant but it is sufficient for
1196 GDB as a DWARF-5 index consumer. */
1197 static int psymbol_tag (const struct partial_symbol
*psym
)
1199 domain_enum domain
= psym
->domain
;
1200 enum address_class aclass
= psym
->aclass
;
1208 return DW_TAG_subprogram
;
1210 return DW_TAG_typedef
;
1212 case LOC_CONST_BYTES
:
1213 case LOC_OPTIMIZED_OUT
:
1215 return DW_TAG_variable
;
1217 /* Note: It's currently impossible to recognize psyms as enum values
1218 short of reading the type info. For now punt. */
1219 return DW_TAG_variable
;
1221 /* There are other LOC_FOO values that one might want to classify
1222 as variables, but dwarf2read.c doesn't currently use them. */
1223 return DW_TAG_variable
;
1226 return DW_TAG_structure_type
;
1228 return DW_TAG_module
;
1234 /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */
1235 void write_psymbols (std::unordered_set
<partial_symbol
*> &psyms_seen
,
1236 const std::vector
<partial_symbol
*> &symbols
,
1237 int cu_index
, bool is_static
, unit_kind kind
)
1239 for (partial_symbol
*psym
: symbols
)
1241 /* Only add a given psymbol once. */
1242 if (psyms_seen
.insert (psym
).second
)
1243 insert (psym
, cu_index
, is_static
, kind
);
1247 /* A helper function that writes a single signatured_type
1248 to a debug_names. */
1250 write_one_signatured_type (struct signatured_type
*entry
,
1251 struct signatured_type_index_data
*info
)
1253 partial_symtab
*psymtab
= entry
->per_cu
.v
.psymtab
;
1255 write_psymbols (info
->psyms_seen
, psymtab
->global_psymbols
,
1256 info
->cu_index
, false, unit_kind::tu
);
1257 write_psymbols (info
->psyms_seen
, psymtab
->static_psymbols
,
1258 info
->cu_index
, true, unit_kind::tu
);
1260 info
->types_list
.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order
,
1261 to_underlying (entry
->per_cu
.sect_off
));
1266 /* Store value of each symbol. */
1267 std::unordered_map
<c_str_view
, std::set
<symbol_value
>, c_str_view_hasher
>
1268 m_name_to_value_set
;
1270 /* Tables of DWARF-5 .debug_names. They are in object file byte
1272 std::vector
<uint32_t> m_bucket_table
;
1273 std::vector
<uint32_t> m_hash_table
;
1275 const bfd_endian m_dwarf5_byte_order
;
1276 dwarf_tmpl
<uint32_t> m_dwarf32
;
1277 dwarf_tmpl
<uint64_t> m_dwarf64
;
1279 offset_vec
&m_name_table_string_offs
, &m_name_table_entry_offs
;
1280 debug_str_lookup m_debugstrlookup
;
1282 /* Map each used .debug_names abbreviation tag parameter to its
1284 std::unordered_map
<index_key
, int, index_key_hasher
> m_indexkey_to_idx
;
1286 /* Next unused .debug_names abbreviation tag for
1287 m_indexkey_to_idx. */
1290 /* .debug_names abbreviation table. */
1291 data_buf m_abbrev_table
;
1293 /* .debug_names entry pool. */
1294 data_buf m_entry_pool
;
1296 /* Temporary storage for Ada names. */
1297 auto_obstack m_string_obstack
;
1300 /* Return iff any of the needed offsets does not fit into 32-bit
1301 .debug_names section. */
1304 check_dwarf64_offsets (dwarf2_per_objfile
*per_objfile
)
1306 for (dwarf2_per_cu_data
*per_cu
: per_objfile
->per_bfd
->all_comp_units
)
1308 if (to_underlying (per_cu
->sect_off
) >= (static_cast<uint64_t> (1) << 32))
1311 for (const signatured_type
*sigtype
: per_objfile
->per_bfd
->all_type_units
)
1313 const dwarf2_per_cu_data
&per_cu
= sigtype
->per_cu
;
1315 if (to_underlying (per_cu
.sect_off
) >= (static_cast<uint64_t> (1) << 32))
1321 /* The psyms_seen set is potentially going to be largish (~40k
1322 elements when indexing a -g3 build of GDB itself). Estimate the
1323 number of elements in order to avoid too many rehashes, which
1324 require rebuilding buckets and thus many trips to
1328 psyms_seen_size (dwarf2_per_objfile
*per_objfile
)
1330 size_t psyms_count
= 0;
1331 for (dwarf2_per_cu_data
*per_cu
: per_objfile
->per_bfd
->all_comp_units
)
1333 partial_symtab
*psymtab
= per_cu
->v
.psymtab
;
1335 if (psymtab
!= NULL
&& psymtab
->user
== NULL
)
1336 recursively_count_psymbols (psymtab
, psyms_count
);
1338 /* Generating an index for gdb itself shows a ratio of
1339 TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */
1340 return psyms_count
/ 4;
1343 /* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek
1344 position is at the end of the file. */
1347 assert_file_size (FILE *file
, size_t expected_size
)
1349 const auto file_size
= ftell (file
);
1350 if (file_size
== -1)
1351 perror_with_name (("ftell"));
1352 gdb_assert (file_size
== expected_size
);
1355 /* Write a gdb index file to OUT_FILE from all the sections passed as
1359 write_gdbindex_1 (FILE *out_file
,
1360 const data_buf
&cu_list
,
1361 const data_buf
&types_cu_list
,
1362 const data_buf
&addr_vec
,
1363 const data_buf
&symtab_vec
,
1364 const data_buf
&constant_pool
)
1367 const offset_type size_of_header
= 6 * sizeof (offset_type
);
1368 offset_type total_len
= size_of_header
;
1370 /* The version number. */
1371 contents
.append_data (MAYBE_SWAP (8));
1373 /* The offset of the CU list from the start of the file. */
1374 contents
.append_data (MAYBE_SWAP (total_len
));
1375 total_len
+= cu_list
.size ();
1377 /* The offset of the types CU list from the start of the file. */
1378 contents
.append_data (MAYBE_SWAP (total_len
));
1379 total_len
+= types_cu_list
.size ();
1381 /* The offset of the address table from the start of the file. */
1382 contents
.append_data (MAYBE_SWAP (total_len
));
1383 total_len
+= addr_vec
.size ();
1385 /* The offset of the symbol table from the start of the file. */
1386 contents
.append_data (MAYBE_SWAP (total_len
));
1387 total_len
+= symtab_vec
.size ();
1389 /* The offset of the constant pool from the start of the file. */
1390 contents
.append_data (MAYBE_SWAP (total_len
));
1391 total_len
+= constant_pool
.size ();
1393 gdb_assert (contents
.size () == size_of_header
);
1395 contents
.file_write (out_file
);
1396 cu_list
.file_write (out_file
);
1397 types_cu_list
.file_write (out_file
);
1398 addr_vec
.file_write (out_file
);
1399 symtab_vec
.file_write (out_file
);
1400 constant_pool
.file_write (out_file
);
1402 assert_file_size (out_file
, total_len
);
1405 /* Write contents of a .gdb_index section for OBJFILE into OUT_FILE.
1406 If OBJFILE has an associated dwz file, write contents of a .gdb_index
1407 section for that dwz file into DWZ_OUT_FILE. If OBJFILE does not have an
1408 associated dwz file, DWZ_OUT_FILE must be NULL. */
1411 write_gdbindex (dwarf2_per_objfile
*per_objfile
, FILE *out_file
,
1414 struct objfile
*objfile
= per_objfile
->objfile
;
1415 mapped_symtab symtab
;
1416 data_buf objfile_cu_list
;
1417 data_buf dwz_cu_list
;
1419 /* While we're scanning CU's create a table that maps a psymtab pointer
1420 (which is what addrmap records) to its index (which is what is recorded
1421 in the index file). This will later be needed to write the address
1423 psym_index_map cu_index_htab
;
1424 cu_index_htab
.reserve (per_objfile
->per_bfd
->all_comp_units
.size ());
1426 /* The CU list is already sorted, so we don't need to do additional
1427 work here. Also, the debug_types entries do not appear in
1428 all_comp_units, but only in their own hash table. */
1430 std::unordered_set
<partial_symbol
*> psyms_seen
1431 (psyms_seen_size (per_objfile
));
1432 for (int i
= 0; i
< per_objfile
->per_bfd
->all_comp_units
.size (); ++i
)
1434 dwarf2_per_cu_data
*per_cu
= per_objfile
->per_bfd
->all_comp_units
[i
];
1435 partial_symtab
*psymtab
= per_cu
->v
.psymtab
;
1437 if (psymtab
!= NULL
)
1439 if (psymtab
->user
== NULL
)
1440 recursively_write_psymbols (objfile
, psymtab
, &symtab
,
1443 const auto insertpair
= cu_index_htab
.emplace (psymtab
, i
);
1444 gdb_assert (insertpair
.second
);
1447 /* The all_comp_units list contains CUs read from the objfile as well as
1448 from the eventual dwz file. We need to place the entry in the
1449 corresponding index. */
1450 data_buf
&cu_list
= per_cu
->is_dwz
? dwz_cu_list
: objfile_cu_list
;
1451 cu_list
.append_uint (8, BFD_ENDIAN_LITTLE
,
1452 to_underlying (per_cu
->sect_off
));
1453 cu_list
.append_uint (8, BFD_ENDIAN_LITTLE
, per_cu
->length
);
1456 /* Dump the address map. */
1458 write_address_map (objfile
, addr_vec
, cu_index_htab
);
1460 /* Write out the .debug_type entries, if any. */
1461 data_buf types_cu_list
;
1462 if (per_objfile
->per_bfd
->signatured_types
)
1464 signatured_type_index_data
sig_data (types_cu_list
,
1467 sig_data
.objfile
= objfile
;
1468 sig_data
.symtab
= &symtab
;
1469 sig_data
.cu_index
= per_objfile
->per_bfd
->all_comp_units
.size ();
1470 htab_traverse_noresize (per_objfile
->per_bfd
->signatured_types
.get (),
1471 write_one_signatured_type
, &sig_data
);
1474 /* Now that we've processed all symbols we can shrink their cu_indices
1476 uniquify_cu_indices (&symtab
);
1478 data_buf symtab_vec
, constant_pool
;
1479 write_hash_table (&symtab
, symtab_vec
, constant_pool
);
1481 write_gdbindex_1(out_file
, objfile_cu_list
, types_cu_list
, addr_vec
,
1482 symtab_vec
, constant_pool
);
1484 if (dwz_out_file
!= NULL
)
1485 write_gdbindex_1 (dwz_out_file
, dwz_cu_list
, {}, {}, {}, {});
1487 gdb_assert (dwz_cu_list
.empty ());
1490 /* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */
1491 static const gdb_byte dwarf5_gdb_augmentation
[] = { 'G', 'D', 'B', 0 };
1493 /* Write a new .debug_names section for OBJFILE into OUT_FILE, write
1494 needed addition to .debug_str section to OUT_FILE_STR. Return how
1495 many bytes were expected to be written into OUT_FILE. */
1498 write_debug_names (dwarf2_per_objfile
*per_objfile
,
1499 FILE *out_file
, FILE *out_file_str
)
1501 const bool dwarf5_is_dwarf64
= check_dwarf64_offsets (per_objfile
);
1502 struct objfile
*objfile
= per_objfile
->objfile
;
1503 const enum bfd_endian dwarf5_byte_order
1504 = gdbarch_byte_order (objfile
->arch ());
1506 /* The CU list is already sorted, so we don't need to do additional
1507 work here. Also, the debug_types entries do not appear in
1508 all_comp_units, but only in their own hash table. */
1510 debug_names
nametable (per_objfile
, dwarf5_is_dwarf64
, dwarf5_byte_order
);
1511 std::unordered_set
<partial_symbol
*>
1512 psyms_seen (psyms_seen_size (per_objfile
));
1513 for (int i
= 0; i
< per_objfile
->per_bfd
->all_comp_units
.size (); ++i
)
1515 const dwarf2_per_cu_data
*per_cu
= per_objfile
->per_bfd
->all_comp_units
[i
];
1516 partial_symtab
*psymtab
= per_cu
->v
.psymtab
;
1518 /* CU of a shared file from 'dwz -m' may be unused by this main
1519 file. It may be referenced from a local scope but in such
1520 case it does not need to be present in .debug_names. */
1521 if (psymtab
== NULL
)
1524 if (psymtab
->user
== NULL
)
1525 nametable
.recursively_write_psymbols (objfile
, psymtab
, psyms_seen
, i
);
1527 cu_list
.append_uint (nametable
.dwarf5_offset_size (), dwarf5_byte_order
,
1528 to_underlying (per_cu
->sect_off
));
1531 /* Write out the .debug_type entries, if any. */
1532 data_buf types_cu_list
;
1533 if (per_objfile
->per_bfd
->signatured_types
)
1535 debug_names::write_one_signatured_type_data
sig_data (nametable
,
1536 signatured_type_index_data (types_cu_list
, psyms_seen
));
1538 sig_data
.info
.objfile
= objfile
;
1539 /* It is used only for gdb_index. */
1540 sig_data
.info
.symtab
= nullptr;
1541 sig_data
.info
.cu_index
= 0;
1542 htab_traverse_noresize (per_objfile
->per_bfd
->signatured_types
.get (),
1543 debug_names::write_one_signatured_type
,
1549 /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */
1551 const offset_type bytes_of_header
1552 = ((dwarf5_is_dwarf64
? 12 : 4)
1554 + sizeof (dwarf5_gdb_augmentation
));
1555 size_t expected_bytes
= 0;
1556 expected_bytes
+= bytes_of_header
;
1557 expected_bytes
+= cu_list
.size ();
1558 expected_bytes
+= types_cu_list
.size ();
1559 expected_bytes
+= nametable
.bytes ();
1562 if (!dwarf5_is_dwarf64
)
1564 const uint64_t size64
= expected_bytes
- 4;
1565 gdb_assert (size64
< 0xfffffff0);
1566 header
.append_uint (4, dwarf5_byte_order
, size64
);
1570 header
.append_uint (4, dwarf5_byte_order
, 0xffffffff);
1571 header
.append_uint (8, dwarf5_byte_order
, expected_bytes
- 12);
1574 /* The version number. */
1575 header
.append_uint (2, dwarf5_byte_order
, 5);
1578 header
.append_uint (2, dwarf5_byte_order
, 0);
1580 /* comp_unit_count - The number of CUs in the CU list. */
1581 header
.append_uint (4, dwarf5_byte_order
,
1582 per_objfile
->per_bfd
->all_comp_units
.size ());
1584 /* local_type_unit_count - The number of TUs in the local TU
1586 header
.append_uint (4, dwarf5_byte_order
,
1587 per_objfile
->per_bfd
->all_type_units
.size ());
1589 /* foreign_type_unit_count - The number of TUs in the foreign TU
1591 header
.append_uint (4, dwarf5_byte_order
, 0);
1593 /* bucket_count - The number of hash buckets in the hash lookup
1595 header
.append_uint (4, dwarf5_byte_order
, nametable
.bucket_count ());
1597 /* name_count - The number of unique names in the index. */
1598 header
.append_uint (4, dwarf5_byte_order
, nametable
.name_count ());
1600 /* abbrev_table_size - The size in bytes of the abbreviations
1602 header
.append_uint (4, dwarf5_byte_order
, nametable
.abbrev_table_bytes ());
1604 /* augmentation_string_size - The size in bytes of the augmentation
1605 string. This value is rounded up to a multiple of 4. */
1606 static_assert (sizeof (dwarf5_gdb_augmentation
) % 4 == 0, "");
1607 header
.append_uint (4, dwarf5_byte_order
, sizeof (dwarf5_gdb_augmentation
));
1608 header
.append_data (dwarf5_gdb_augmentation
);
1610 gdb_assert (header
.size () == bytes_of_header
);
1612 header
.file_write (out_file
);
1613 cu_list
.file_write (out_file
);
1614 types_cu_list
.file_write (out_file
);
1615 nametable
.file_write (out_file
, out_file_str
);
1617 assert_file_size (out_file
, expected_bytes
);
1620 /* This represents an index file being written (work-in-progress).
1622 The data is initially written to a temporary file. When the finalize method
1623 is called, the file is closed and moved to its final location.
1625 On failure (if this object is being destroyed with having called finalize),
1626 the temporary file is closed and deleted. */
1628 struct index_wip_file
1630 index_wip_file (const char *dir
, const char *basename
,
1633 filename
= (std::string (dir
) + SLASH_STRING
+ basename
1636 filename_temp
= make_temp_filename (filename
);
1638 scoped_fd
out_file_fd (gdb_mkostemp_cloexec (filename_temp
.data (),
1640 if (out_file_fd
.get () == -1)
1641 perror_with_name (("mkstemp"));
1643 out_file
= out_file_fd
.to_file ("wb");
1645 if (out_file
== nullptr)
1646 error (_("Can't open `%s' for writing"), filename_temp
.data ());
1648 unlink_file
.emplace (filename_temp
.data ());
1653 /* We want to keep the file. */
1654 unlink_file
->keep ();
1656 /* Close and move the str file in place. */
1657 unlink_file
.reset ();
1658 if (rename (filename_temp
.data (), filename
.c_str ()) != 0)
1659 perror_with_name (("rename"));
1662 std::string filename
;
1663 gdb::char_vector filename_temp
;
1665 /* Order matters here; we want FILE to be closed before
1666 FILENAME_TEMP is unlinked, because on MS-Windows one cannot
1667 delete a file that is still open. So, we wrap the unlinker in an
1668 optional and emplace it once we know the file name. */
1669 gdb::optional
<gdb::unlinker
> unlink_file
;
1671 gdb_file_up out_file
;
1674 /* See dwarf-index-write.h. */
1677 write_psymtabs_to_index (dwarf2_per_objfile
*per_objfile
, const char *dir
,
1678 const char *basename
, const char *dwz_basename
,
1679 dw_index_kind index_kind
)
1681 struct objfile
*objfile
= per_objfile
->objfile
;
1683 if (per_objfile
->per_bfd
->using_index
)
1684 error (_("Cannot use an index to create the index"));
1686 if (per_objfile
->per_bfd
->types
.size () > 1)
1687 error (_("Cannot make an index when the file has multiple .debug_types sections"));
1689 if (!objfile
->partial_symtabs
->psymtabs
1690 || !objfile
->partial_symtabs
->psymtabs_addrmap
)
1694 if (stat (objfile_name (objfile
), &st
) < 0)
1695 perror_with_name (objfile_name (objfile
));
1697 const char *index_suffix
= (index_kind
== dw_index_kind::DEBUG_NAMES
1698 ? INDEX5_SUFFIX
: INDEX4_SUFFIX
);
1700 index_wip_file
objfile_index_wip (dir
, basename
, index_suffix
);
1701 gdb::optional
<index_wip_file
> dwz_index_wip
;
1703 if (dwz_basename
!= NULL
)
1704 dwz_index_wip
.emplace (dir
, dwz_basename
, index_suffix
);
1706 if (index_kind
== dw_index_kind::DEBUG_NAMES
)
1708 index_wip_file
str_wip_file (dir
, basename
, DEBUG_STR_SUFFIX
);
1710 write_debug_names (per_objfile
, objfile_index_wip
.out_file
.get (),
1711 str_wip_file
.out_file
.get ());
1713 str_wip_file
.finalize ();
1716 write_gdbindex (per_objfile
, objfile_index_wip
.out_file
.get (),
1717 (dwz_index_wip
.has_value ()
1718 ? dwz_index_wip
->out_file
.get () : NULL
));
1720 objfile_index_wip
.finalize ();
1722 if (dwz_index_wip
.has_value ())
1723 dwz_index_wip
->finalize ();
1726 /* Implementation of the `save gdb-index' command.
1728 Note that the .gdb_index file format used by this command is
1729 documented in the GDB manual. Any changes here must be documented
1733 save_gdb_index_command (const char *arg
, int from_tty
)
1735 const char dwarf5space
[] = "-dwarf-5 ";
1736 dw_index_kind index_kind
= dw_index_kind::GDB_INDEX
;
1741 arg
= skip_spaces (arg
);
1742 if (strncmp (arg
, dwarf5space
, strlen (dwarf5space
)) == 0)
1744 index_kind
= dw_index_kind::DEBUG_NAMES
;
1745 arg
+= strlen (dwarf5space
);
1746 arg
= skip_spaces (arg
);
1750 error (_("usage: save gdb-index [-dwarf-5] DIRECTORY"));
1752 for (objfile
*objfile
: current_program_space
->objfiles ())
1756 /* If the objfile does not correspond to an actual file, skip it. */
1757 if (stat (objfile_name (objfile
), &st
) < 0)
1760 dwarf2_per_objfile
*per_objfile
= get_dwarf2_per_objfile (objfile
);
1762 if (per_objfile
!= NULL
)
1766 const char *basename
= lbasename (objfile_name (objfile
));
1767 const dwz_file
*dwz
= dwarf2_get_dwz_file (per_objfile
->per_bfd
);
1768 const char *dwz_basename
= NULL
;
1771 dwz_basename
= lbasename (dwz
->filename ());
1773 write_psymtabs_to_index (per_objfile
, arg
, basename
, dwz_basename
,
1776 catch (const gdb_exception_error
&except
)
1778 exception_fprintf (gdb_stderr
, except
,
1779 _("Error while writing index for `%s': "),
1780 objfile_name (objfile
));
1787 void _initialize_dwarf_index_write ();
1789 _initialize_dwarf_index_write ()
1791 cmd_list_element
*c
= add_cmd ("gdb-index", class_files
,
1792 save_gdb_index_command
, _("\
1793 Save a gdb-index file.\n\
1794 Usage: save gdb-index [-dwarf-5] DIRECTORY\n\
1796 No options create one file with .gdb-index extension for pre-DWARF-5\n\
1797 compatible .gdb_index section. With -dwarf-5 creates two files with\n\
1798 extension .debug_names and .debug_str for DWARF-5 .debug_names section."),
1800 set_cmd_completer (c
, filename_completer
);