[V4,2/2] CTF: multi-CU and archive support

Message ID 1619485254-7340-3-git-send-email-weimin.pan@oracle.com
State New
Headers show
Series
  • CTF: bug fixes and new features.
Related show

Commit Message

Mike Frysinger via Gdb-patches April 27, 2021, 1 a.m.
Now gdb is capable of debugging executable, which consists of multiple 
compilation units, with CTF. An executable could potentially have one
or more archives, which, in CTF context, contain conflicting types.   
When comparing to DWARF2, there is a major difference in the type sections, 

all changes were made in ctfread.c where elfctf_build_psymtabs and 
scan_partial_symbols, with the new ctf_per_tu_data struct, were modify to 
handle archives which were treated as dependencies in gdb, via the ctf 
archive iterator and its callback build_ctf_archive_member, and were then 
expanded with expand_dependencies when psymtabs were expanded. 

Also changes were made to handle CTF's data object section and function
info section which now share the same format for their contents - an array 
of type IDs. New functions ctf_psymtab_add_stt_entries, which is called by 
ctf_psymtab_add_stt_obj and ctf_psymtab_add_stt_func, and add_stt_entries, 
which is called by add_stt_obj and add_stt_func when setting up psymtabs 
and full symtab, respectively.

---
 gdb/ctfread.c                             | 317 ++++++++++++++++++------------
 gdb/testsuite/gdb.ctf/cross-tu-cyclic-1.c |  18 ++
 gdb/testsuite/gdb.ctf/cross-tu-cyclic-2.c |  16 ++
 gdb/testsuite/gdb.ctf/cross-tu-cyclic-3.c |   3 +
 gdb/testsuite/gdb.ctf/cross-tu-cyclic-4.c |   4 +
 gdb/testsuite/gdb.ctf/cross-tu-cyclic.exp |  43 ++++
 gdb/testsuite/gdb.ctf/ctf-a.c             |  32 +++
 gdb/testsuite/gdb.ctf/ctf-a.h             |  22 +++
 gdb/testsuite/gdb.ctf/ctf-b.c             |  25 +++
 gdb/testsuite/gdb.ctf/ctf-b.h             |  22 +++
 gdb/testsuite/gdb.ctf/ctf-c.c             |  25 +++
 gdb/testsuite/gdb.ctf/ctf-c.h             |  21 ++
 gdb/testsuite/gdb.ctf/multi.exp           |  42 ++++
 13 files changed, 464 insertions(+), 126 deletions(-)
 create mode 100644 gdb/testsuite/gdb.ctf/cross-tu-cyclic-1.c
 create mode 100644 gdb/testsuite/gdb.ctf/cross-tu-cyclic-2.c
 create mode 100644 gdb/testsuite/gdb.ctf/cross-tu-cyclic-3.c
 create mode 100644 gdb/testsuite/gdb.ctf/cross-tu-cyclic-4.c
 create mode 100644 gdb/testsuite/gdb.ctf/cross-tu-cyclic.exp
 create mode 100644 gdb/testsuite/gdb.ctf/ctf-a.c
 create mode 100644 gdb/testsuite/gdb.ctf/ctf-a.h
 create mode 100644 gdb/testsuite/gdb.ctf/ctf-b.c
 create mode 100644 gdb/testsuite/gdb.ctf/ctf-b.h
 create mode 100644 gdb/testsuite/gdb.ctf/ctf-c.c
 create mode 100644 gdb/testsuite/gdb.ctf/ctf-c.h
 create mode 100644 gdb/testsuite/gdb.ctf/multi.exp

-- 
1.8.3.1

Comments

Tom Tromey May 7, 2021, 8:29 p.m. | #1
>>>>> ">" == Weimin Pan via Gdb-patches <gdb-patches@sourceware.org> writes:


>> Now gdb is capable of debugging executable, which consists of multiple 

>> compilation units, with CTF. An executable could potentially have one

>> or more archives, which, in CTF context, contain conflicting types.   

>> When comparing to DWARF2, there is a major difference in the type sections, 


I didn't understand the last sentence here.

I'm afraid there's a lot about this patch I don't really understand.
This doesn't mean anything about the patch, it's more my ignorance of
CTF.  But, I'd like to at least have some idea before approving it.

>> +/* Persistent data held for a translation unit.  */

>> +

>> +struct ctf_per_tu_data


I don't think "persistent" is the right term here, because one of these
is created in elfctf_build_psymtabs and then discarded when the function
returns.

>> +  for (int i = 0; i < pst->number_of_dependencies; i++)

>> +    {

>> +      dp = (ctf_psymtab *)pst->dependencies[i];


Should be a space after the cast.

>> +      pst->add_psymbol (tname, true,

>> +			tdomain, aclass, -1,

>> +			psymbol_placement::GLOBAL,

>> +			0, language_c, pst->context->partial_symtabs, of);


I don't know what ctf_symbol_next does, but I wonder if 'tname' truly
needs to be copied here, or whether it's possible to change true->false
in the call.

>> +  ctf_dict_close (ccp->fp);


I didn't understand the placement of this addition.
I expected it to be paired with some kind of dict_open call, but I
didn't see that.  Could you explain it?

>> +/* Callback to build the psymtab for archive member NAME.  */

>> +

>> +static int

>> +build_ctf_archive_member (ctf_dict_t *ctf, const char *name, void *arg)

>> +{

>> +  struct ctf_per_tu_data *tup = (struct ctf_per_tu_data *) arg;

>> +  ctf_dict_t *parent = tup->fp;

>> +  ctf_psymtab *pst;

>> +  int isparent = 0;

>> +

>> +  if (strcmp (name, ".ctf") != 0)

>> +    {

>> +      ctf_import (ctf, parent);

>>      }


Remove the brace here.

I don't know what the ".ctf" thing means here, but I guess it's some
special marker?  Could you explain the layout here a little?

>> +  if (isparent)

>> +    tup->parent_psymtab = pst;

>> +  else

>> +    tup->imported_symtabs.push_back (pst);


Like, how many psymtabs might be marked as a parent, and how many would
be pushed as dependencies?

Maybe it would be helpful to know how the new test cases would look
here.

thanks,
Tom
Tom Tromey May 8, 2021, 8:13 p.m. | #2
>> +      for (int i = 0; i < arch_cnt; ++i)

>> +	{

>> +	  parent_pst->dependencies[i]

>> +	    = pcu.imported_symtabs.at (i);

>> +	}


I forgot to mention -- don't use .at(), just use [] instead.
This loop could probably just be a memcpy or a std::copy, though an
explicit loop is also fine.

Tom
Mike Frysinger via Gdb-patches May 10, 2021, 3:11 a.m. | #3
On 5/7/2021 1:29 PM, Tom Tromey wrote:
>>>>>> ">" == Weimin Pan via Gdb-patches <gdb-patches@sourceware.org> writes:

>>> Now gdb is capable of debugging executable, which consists of multiple

>>> compilation units, with CTF. An executable could potentially have one

>>> or more archives, which, in CTF context, contain conflicting types.

>>> When comparing to DWARF2, there is a major difference in the type sections,

> I didn't understand the last sentence here.

>

> I'm afraid there's a lot about this patch I don't really understand.

> This doesn't mean anything about the patch, it's more my ignorance of

> CTF.  But, I'd like to at least have some idea before approving it.


Hi Tom,

Thank you for taking the time. Please see below for some explanation 
about CTF
and let me know if you need more info.

>>> +/* Persistent data held for a translation unit.  */

>>> +

>>> +struct ctf_per_tu_data

> I don't think "persistent" is the right term here, because one of these

> is created in elfctf_build_psymtabs and then discarded when the function

> returns.


Makes sense, will drop it.

>

>>> +  for (int i = 0; i < pst->number_of_dependencies; i++)

>>> +    {

>>> +      dp = (ctf_psymtab *)pst->dependencies[i];

> Should be a space after the cast.


Done.

>

>>> +      pst->add_psymbol (tname, true,

>>> +			tdomain, aclass, -1,

>>> +			psymbol_placement::GLOBAL,

>>> +			0, language_c, pst->context->partial_symtabs, of);

> I don't know what ctf_symbol_next does, but I wonder if 'tname' truly

> needs to be copied here, or whether it's possible to change true->false

> in the call.


I think you're right. Will check with Nick Alcock who is maintaining 
libctf to confirm.

>

>>> +  ctf_dict_close (ccp->fp);

> I didn't understand the placement of this addition.

> I expected it to be paired with some kind of dict_open call, but I

> didn't see that.  Could you explain it?


This file closing call is incorrect and will be removed. The proper 
place for
closing should be done in ctf_fp_info's destructor, which corresponds to
ctf_dict_open in elfctf_build_psymtabs.

>>> +/* Callback to build the psymtab for archive member NAME.  */

>>> +

>>> +static int

>>> +build_ctf_archive_member (ctf_dict_t *ctf, const char *name, void *arg)

>>> +{

>>> +  struct ctf_per_tu_data *tup = (struct ctf_per_tu_data *) arg;

>>> +  ctf_dict_t *parent = tup->fp;

>>> +  ctf_psymtab *pst;

>>> +  int isparent = 0;

>>> +

>>> +  if (strcmp (name, ".ctf") != 0)

>>> +    {

>>> +      ctf_import (ctf, parent);

>>>       }

> Remove the brace here.


OK.

>

> I don't know what the ".ctf" thing means here, but I guess it's some

> special marker?  Could you explain the layout here a little?

>

>>> +  if (isparent)

>>> +    tup->parent_psymtab = pst;

>>> +  else

>>> +    tup->imported_symtabs.push_back (pst);

> Like, how many psymtabs might be marked as a parent, and how many would

> be pushed as dependencies?

>

> Maybe it would be helpful to know how the new test cases would look

> here.


A CTF archive is a CU which contains all CTF sections. The difference
is the "Types" section in an archive contains conflicting types only.
It's been treated like a dependency and gdb traverses each archive,
via ctf_archive_iter, to extract its CTF info and build its internal
partial/full symtabs. And expand_dependencies is called to expand
the archives when its parent is expanded.

Here is an example in which "struct A" is being defined multiple times:

% cat t.c
struct A;
struct B
{
   int foo;
   struct A *bar;
};

struct A
{
   long a;
   struct B *foo;
};
static struct A *foo __attribute__((used));

int main() { }

% cat t1.c
struct B;
struct A
{
   long a;
   struct B *foo1;
   struct C *bar1;
};
static struct A *bar __attribute__((used));

% cat t.2
struct A { struct B *foo2; };
static struct A *a __attribute__((__used__));
static struct A *conflicty __attribute__((__used__));

% gcc -gt -Wl,--export-dynamic *.c
% objdump --ctf=.ctf a.out
a.out:     file format elf64-x86-64

Contents of CTF section .ctf:

   Header:
     Magic number: dff2
     Version: 4 (CTF_VERSION_3)
     Flags: 0xe (CTF_F_NEWFUNCINFO, CTF_F_IDXSORTED, CTF_F_DYNSTR)
     Function info section:      0x0 -- 0xb (0xc bytes)
     Type section:       0xc -- 0x9f (0x94 bytes)
     String section:     0xa0 -- 0xbc (0x1d bytes)

   Labels:

   Data objects:

   Function objects:
     main -> int (*) (...)

   Variables:

   Types:
      1: long int [0x0:0x40] (size 0x8)
         [0x0] (ID 0x1) (kind 1) long int:64 (aligned at 0x8, format 
0x1, offset:bits 0x0:0x40)
      2: struct B (size 0x10)
         [0x0] (ID 0x2) (kind 6) struct B (aligned at 0x4)
             [0x0] (ID 0x3) (kind 1) int foo:32 (aligned at 0x4, format 
0x1, offset:bits 0x0:0x20)
             [0x40] (ID 0x5) (kind 3) struct A * bar (aligned at 0x8)
      3: int [0x0:0x20] (size 0x4)
         [0x0] (ID 0x3) (kind 1) int:32 (aligned at 0x4, format 0x1, 
offset:bits 0x0:0x20)
      4: struct A (size 0x6)
         [0x0] (ID 0x4) (kind 9) struct A (aligned at 0x6)
      5: struct A * (size 0x8) -> 4: struct A (size 0x6)
         [0x0] (ID 0x5) (kind 3) struct A * (aligned at 0x8)
      6: struct B * (size 0x8) -> 2: struct B (size 0x10)
         [0x0] (ID 0x6) (kind 3) struct B * (aligned at 0x8)
      7: int (*) (...) (size 0x0)
         [0x0] (ID 0x7) (kind 5) int (*) (...) (aligned at 0x8)
      8: struct C (size 0x6)
         [0x0] (ID 0x8) (kind 9) struct C (aligned at 0x6)
      9: struct C * (size 0x8) -> 8: struct C (size 0x6)
         [0x0] (ID 0x9) (kind 3) struct C * (aligned at 0x8)

   Strings:
     0:
     1: A
     3: B
     5: C
     7: bar
     b: foo
     f: int
     13: long int

CTF archive member: /tests/ctf/cross/t.c:

   Header:
     Magic number: dff2
     Version: 4 (CTF_VERSION_3)
     Flags: 0xe (CTF_F_NEWFUNCINFO, CTF_F_IDXSORTED, CTF_F_DYNSTR)
     Parent name: .ctf
     Compilation unit name: /tests/ctf/cross/t.c
     Type section:       0x0 -- 0x23 (0x24 bytes)
     String section:     0x24 -- 0x58 (0x35 bytes)

   Labels:

   Data objects:

   Function objects:

   Variables:

   Types:
      80000001: struct A (size 0x10)
         [0x0] (ID 0x80000001) (kind 6) struct A (aligned at 0x8)
             [0x0] (ID 0x1) (kind 1) long int a:64 (aligned at 0x8, 
format 0x1, offset:bits 0x0:0x40)
             [0x40] (ID 0x6) (kind 3) struct B * foo (aligned at 0x8)

   Strings:
     0:
     1: .ctf
     6: /tests/ctf/cross/t.c
     2c: A
     2e: a
     30: foo

CTF archive member: /tests/ctf/cross/t1.c:

   Header:
     Magic number: dff2
     Version: 4 (CTF_VERSION_3)
     Flags: 0xe (CTF_F_NEWFUNCINFO, CTF_F_IDXSORTED, CTF_F_DYNSTR)
     Parent name: .ctf
     Compilation unit name: /tests/ctf/cross/t1.c
     Type section:       0x0 -- 0x2f (0x30 bytes)
     String section:     0x30 -- 0x6b (0x3c bytes)

   Labels:

   Data objects:

   Function objects:

   Variables:

   Types:
      80000001: struct A (size 0x18)
         [0x0] (ID 0x80000001) (kind 6) struct A (aligned at 0x8)
             [0x0] (ID 0x1) (kind 1) long int a:64 (aligned at 0x8, 
format 0x1, offset:bits 0x0:0x40)
             [0x40] (ID 0x6) (kind 3) struct B * foo1 (aligned at 0x8)
             [0x80] (ID 0x9) (kind 3) struct C * bar1 (aligned at 0x8)

   Strings:
     0:
     1: .ctf
     6: /tests/ctf/cross/t1.c
     2d: A
     2f: a
     31: bar1
     36: foo1

CTF archive member: /tests/ctf/cross/t2.c:

   Header:
     Magic number: dff2
     Version: 4 (CTF_VERSION_3)
     Flags: 0xe (CTF_F_NEWFUNCINFO, CTF_F_IDXSORTED, CTF_F_DYNSTR)
     Parent name: .ctf
     Compilation unit name: /tests/ctf/cross/t2.c
     Type section:       0x0 -- 0x17 (0x18 bytes)
     String section:     0x18 -- 0x4c (0x35 bytes)

   Labels:

   Data objects:

   Function objects:

   Variables:

   Types:
      80000001: struct A (size 0x8)
         [0x0] (ID 0x80000001) (kind 6) struct A (aligned at 0x8)
             [0x0] (ID 0x6) (kind 3) struct B * foo2 (aligned at 0x8)

   Strings:
     0:
     1: .ctf
     6: /tests/ctf/cross/t2.c
     2d: A
     2f: foo2

Where each archive has the CU as its name and ".ctf" as its parent's name
(emitted by the linker.) In terms of psymtabs, there is only one parent
which can have any number of archives.

Thanks again.
>

> thanks,

> Tom
Mike Frysinger via Gdb-patches May 10, 2021, 3:12 a.m. | #4
On 5/8/2021 1:13 PM, Tom Tromey wrote:
>>> +      for (int i = 0; i < arch_cnt; ++i)

>>> +	{

>>> +	  parent_pst->dependencies[i]

>>> +	    = pcu.imported_symtabs.at (i);

>>> +	}

> I forgot to mention -- don't use .at(), just use [] instead.

> This loop could probably just be a memcpy or a std::copy, though an

> explicit loop is also fine.


OK, thanks.

>

> Tom
Tom Tromey May 10, 2021, 9:27 p.m. | #5
>>>>> ">" == Wei-min Pan <weimin.pan@oracle.com> writes:


>> Thank you for taking the time. Please see below for some explanation

>> about CTF and let me know if you need more info.


Thank you.

>> Where each archive has the CU as its name and ".ctf" as its parent's name

>> (emitted by the linker.) In terms of psymtabs, there is only one parent

>> which can have any number of archives.


I think this will probably yield over-expansion in gdb.  What I mean is
that if any symbol is found in CTF, all types and symbols will be
expanded.  Normally this isn't what you want, though -- you only really
need or want to expand the psymtab that has the type (or whatever) that
was asked for.

Normally what you want is for a single psymtab to correspond to a single
source file.  In gdb the dependencies are used for #includes, which I
guess comes from how stabs were originally implemented.

Tom
Mike Frysinger via Gdb-patches May 11, 2021, 12:33 a.m. | #6
On 5/10/2021 2:27 PM, Tom Tromey wrote:
>>>>>> ">" == Wei-min Pan <weimin.pan@oracle.com> writes:

>>> Thank you for taking the time. Please see below for some explanation

>>> about CTF and let me know if you need more info.

> Thank you.

>

>>> Where each archive has the CU as its name and ".ctf" as its parent's name

>>> (emitted by the linker.) In terms of psymtabs, there is only one parent

>>> which can have any number of archives.

> I think this will probably yield over-expansion in gdb.  What I mean is

> that if any symbol is found in CTF, all types and symbols will be

> expanded.  Normally this isn't what you want, though -- you only really

> need or want to expand the psymtab that has the type (or whatever) that

> was asked for.

>

> Normally what you want is for a single psymtab to correspond to a single

> source file.  In gdb the dependencies are used for #includes, which I

> guess comes from how stabs were originally implemented.


Yes, that's what happens with the CTF layout. While there is some
benefit of generating psymtabs, it's quite limited as you pointed out.
And I doubt that even getting rid of emitting psymtabs altogether
will not help much. Do you have any suggestions?

Thanks.

>

> Tom
Mike Frysinger via Gdb-patches May 11, 2021, 6:58 p.m. | #7
On 5/10/2021 2:27 PM, Tom Tromey wrote:
>>>>>> ">" == Wei-min Pan <weimin.pan@oracle.com> writes:

>>> Thank you for taking the time. Please see below for some explanation

>>> about CTF and let me know if you need more info.

> Thank you.

>

>>> Where each archive has the CU as its name and ".ctf" as its parent's name

>>> (emitted by the linker.) In terms of psymtabs, there is only one parent

>>> which can have any number of archives.

> I think this will probably yield over-expansion in gdb.  What I mean is

> that if any symbol is found in CTF, all types and symbols will be

> expanded.  Normally this isn't what you want, though -- you only really

> need or want to expand the psymtab that has the type (or whatever) that

> was asked for.

>

> Normally what you want is for a single psymtab to correspond to a single

> source file.  In gdb the dependencies are used for #includes, which I

> guess comes from how stabs were originally implemented.


Or we can treat CTF archives as CUs, as opposed to dependencies.

>

> Tom

Patch

diff --git a/gdb/ctfread.c b/gdb/ctfread.c
index 97355bf..284cf91 100644
--- a/gdb/ctfread.c
+++ b/gdb/ctfread.c
@@ -117,6 +117,7 @@  struct ctf_context
   struct objfile *of;
   psymtab_storage *partial_symtabs;
   partial_symtab *pst;
+  ctf_archive_t *arc;
   struct buildsym_compunit *builder;
 };
 
@@ -166,6 +167,18 @@  struct ctf_field_info
   std::vector<struct decl_field> nested_types_list;
 };
 
+/* Persistent data held for a translation unit.  */
+
+struct ctf_per_tu_data
+{
+  ctf_dict_t *fp;
+  struct objfile *of;
+  ctf_archive_t *arc;
+  ctf_psymtab *parent_psymtab;
+  psymtab_storage *pss;
+  psymbol_functions *psf;
+  std::vector <struct partial_symtab *> imported_symtabs;
+};
 
 /* Local function prototypes */
 
@@ -245,10 +258,8 @@  struct ctf_tid_and_type
   ids.tid = tid;
   ids.type = typ;
   slot = (struct ctf_tid_and_type **) htab_find_slot (htab, &ids, INSERT);
-  if (*slot)
-    complaint (_("An internal GDB problem: ctf_ id_t %ld type already set"),
-	       (tid));
-  *slot = XOBNEW (&of->objfile_obstack, struct ctf_tid_and_type);
+  if (*slot == nullptr)
+    *slot = XOBNEW (&of->objfile_obstack, struct ctf_tid_and_type);
   **slot = ids;
   return typ;
 }
@@ -1144,7 +1155,8 @@  struct ctf_tid_and_type
 	if (type)
 	  {
 	    sym = new_symbol (ccp, type, id);
-	    sym->compute_and_set_names (name, false, ccp->of->per_bfd);
+	    if (sym)
+	      sym->compute_and_set_names (name, false, ccp->of->per_bfd);
 	  }
 	break;
       case CTF_K_STRUCT:
@@ -1174,80 +1186,48 @@  struct ctf_tid_and_type
   return 0;
 }
 
-/* Add an ELF STT_OBJ symbol with index IDX to the symbol table.  */
+/* Add entries in either data objects or function info section, controlled
+   by FUNCTIONS.  */
 
-static struct symbol *
-add_stt_obj (struct ctf_context *ccp, unsigned long idx)
+static void
+add_stt_entries (struct ctf_context *ccp, int functions)
 {
-  struct symbol *sym;
-  struct type *type;
+  ctf_next_t *i = nullptr;
+  const char *tname;
   ctf_id_t tid;
+  struct symbol *sym = nullptr;
+  struct type *type;
 
-  if ((tid = ctf_lookup_by_symbol (ccp->fp, idx)) == CTF_ERR)
-    return nullptr;
-
-  type = fetch_tid_type (ccp, tid);
-  if (type == nullptr)
-    return nullptr;
-
-  sym = new_symbol (ccp, type, tid);
-
-  return sym;
+  while ((tid = ctf_symbol_next (ccp->fp, &i, &tname, functions)) != CTF_ERR)
+    {
+      type = get_tid_type (ccp->of, tid);
+      if (type == nullptr)
+	continue;
+      sym = new (&ccp->of->objfile_obstack) symbol;
+      OBJSTAT (ccp->of, n_syms++);
+      SYMBOL_TYPE (sym) = type;
+      SYMBOL_DOMAIN (sym) = VAR_DOMAIN;
+      SYMBOL_ACLASS_INDEX (sym) = LOC_STATIC;
+      sym->compute_and_set_names (tname, false, ccp->of->per_bfd);
+      add_symbol_to_list (sym, ccp->builder->get_global_symbols ());
+      set_symbol_address (ccp->of, sym, tname);
+    }
 }
 
-/* Add an ELF STT_FUNC symbol with index IDX to the symbol table.  */
+/* Add entries in data objects section.  */
 
-static struct symbol *
-add_stt_func (struct ctf_context *ccp, unsigned long idx)
+static void
+add_stt_obj (struct ctf_context *ccp)
 {
-  struct type *ftype, *atyp, *rettyp;
-  struct symbol *sym;
-  ctf_funcinfo_t finfo;
-  ctf_id_t argv[32];
-  uint32_t argc;
-  ctf_id_t tid;
-  struct type *void_type = objfile_type (ccp->of)->builtin_void;
-
-  if (ctf_func_info (ccp->fp, idx, &finfo) == CTF_ERR)
-    return nullptr;
-
-  argc = finfo.ctc_argc;
-  if (ctf_func_args (ccp->fp, idx, argc, argv) == CTF_ERR)
-    return nullptr;
-
-  if (ctf_type_name_raw (ccp->fp, idx) == nullptr)
-    return nullptr;
-
-  tid = ctf_lookup_by_symbol (ccp->fp, idx);
-  ftype = fetch_tid_type (ccp, tid);
-  if ((finfo.ctc_flags & CTF_FUNC_VARARG) != 0)
-    ftype->set_has_varargs (true);
-  ftype->set_num_fields (argc);
-
-  /* If argc is 0, it has a "void" type.  */
-  if (argc != 0)
-    ftype->set_fields
-      ((struct field *) TYPE_ZALLOC (ftype, argc * sizeof (struct field)));
-
-  /* TYPE_FIELD_TYPE must never be NULL.  Fill it with void_type, if failed
-     to find the argument type.  */
-  for (int iparam = 0; iparam < argc; iparam++)
-    {
-      atyp = fetch_tid_type (ccp, argv[iparam]);
-      if (atyp)
-	ftype->field (iparam).set_type (atyp);
-      else
-	ftype->field (iparam).set_type (void_type);
-    }
+  add_stt_entries (ccp, 0);
+}
 
-  sym = new_symbol (ccp, ftype, tid);
-  rettyp = fetch_tid_type (ccp, finfo.ctc_return);
-  if (rettyp != nullptr)
-    SYMBOL_TYPE (sym) = rettyp;
-  else
-    SYMBOL_TYPE (sym) = void_type;
+/* Add entries in function info section.  */
 
-  return sym;
+static void
+add_stt_func (struct ctf_context *ccp)
+{
+  add_stt_entries (ccp, 1);
 }
 
 /* Get text segment base for OBJFILE, TSIZE contains the segment size.  */
@@ -1270,12 +1250,20 @@  struct ctf_tid_and_type
 		  struct objfile *of, CORE_ADDR text_offset)
 {
   struct ctf_context *ccp;
+  ctf_psymtab *dp;
 
   ccp = pst->context;
   ccp->builder = new buildsym_compunit
 		       (of, of->original_name, nullptr,
 		       language_c, text_offset);
   ccp->builder->record_debugformat ("ctf");
+
+  for (int i = 0; i < pst->number_of_dependencies; i++)
+    {
+      dp = (ctf_psymtab *)pst->dependencies[i];
+      dp->context->builder = ccp->builder;
+      dp->context->of = of;
+    }
 }
 
 /* Finish reading symbol/type definitions in CTF format.
@@ -1317,17 +1305,77 @@  struct ctf_tid_and_type
 	       ctf_errmsg (ctf_errno (ccp->fp)));
 }
 
+/* Add entries in either data objects or function info section, controlled
+   by FUNCTIONS, to psymtab.  */
+
+static void
+ctf_psymtab_add_stt_entries (ctf_dict_t *cfp, ctf_psymtab *pst,
+			     struct objfile *of, int functions)
+{
+  ctf_next_t *i = nullptr;
+  ctf_id_t tid;
+  const char *tname;
+
+  while ((tid = ctf_symbol_next (cfp, &i, &tname, functions)) != CTF_ERR)
+    {
+      uint32_t kind = ctf_type_kind (cfp, tid);
+      address_class aclass;
+      domain_enum tdomain;
+      switch (kind)
+	{
+	  case CTF_K_STRUCT:
+	  case CTF_K_UNION:
+	  case CTF_K_ENUM:
+	    tdomain = STRUCT_DOMAIN;
+	    break;
+	  default:
+	    tdomain = VAR_DOMAIN;
+	    break;
+	}
+
+      if (kind == CTF_K_FUNCTION)
+	aclass = LOC_STATIC;
+      else if (kind == CTF_K_CONST)
+	aclass = LOC_CONST;
+      else
+	aclass = LOC_TYPEDEF;
+
+      pst->add_psymbol (tname, true,
+			tdomain, aclass, -1,
+			psymbol_placement::GLOBAL,
+			0, language_c, pst->context->partial_symtabs, of);
+    }
+}
+
+/* Add entries in data objects section to psymtab.  */
+
+static void
+ctf_psymtab_add_stt_obj (ctf_dict_t *cfp, ctf_psymtab *pst,
+			 struct objfile *of)
+{
+  ctf_psymtab_add_stt_entries (cfp, pst, of, 0);
+}
+
+/* Add entries in function info section to psymtab.  */
+
+static void
+ctf_psymtab_add_stt_func (ctf_dict_t *cfp, ctf_psymtab *pst,
+			  struct objfile *of)
+{
+  ctf_psymtab_add_stt_entries (cfp, pst, of, 1);
+}
+
 /* Read in full symbols for PST, and anything it depends on.  */
 
 void
 ctf_psymtab::expand_psymtab (struct objfile *objfile)
 {
-  struct symbol *sym;
   struct ctf_context *ccp;
 
   gdb_assert (!readin);
 
   ccp = context;
+  expand_dependencies (objfile);
 
   /* Iterate over entries in data types section.  */
   if (ctf_type_iter (ccp->fp, ctf_add_type_cb, ccp) == CTF_ERR)
@@ -1341,22 +1389,10 @@  struct ctf_tid_and_type
 	       ctf_errmsg (ctf_errno (ccp->fp)));
 
   /* Add entries in data objects and function info sections.  */
-  for (unsigned long i = 0; ; i++)
-    {
-      sym = add_stt_obj (ccp, i);
-      if (sym == nullptr)
-	{
-	  if (ctf_errno (ccp->fp) == EINVAL
-	      || ctf_errno (ccp->fp) == ECTF_NOSYMTAB)
-	    break;
-	  sym = add_stt_func (ccp, i);
-	}
-      if (sym == nullptr)
-	continue;
-
-      set_symbol_address (ccp->of, sym, sym->linkage_name ());
-    }
+  add_stt_obj (ccp);
+  add_stt_func (ccp);
 
+  ctf_dict_close (ccp->fp);
   readin = true;
 }
 
@@ -1409,6 +1445,7 @@  struct ctf_tid_and_type
 
 static ctf_psymtab *
 create_partial_symtab (const char *name,
+		       ctf_archive_t *arc,
 		       ctf_dict_t *cfp,
 		       psymtab_storage *partial_symtabs,
 		       struct objfile *objfile)
@@ -1419,11 +1456,12 @@  struct ctf_tid_and_type
   pst = new ctf_psymtab (name, partial_symtabs, objfile->per_bfd, 0);
 
   ccx = XOBNEW (&objfile->objfile_obstack, struct ctf_context);
+  ccx->arc = arc;
   ccx->fp = cfp;
+  ctf_ref (cfp);
   ccx->of = objfile;
   ccx->partial_symtabs = partial_symtabs;
   ccx->pst = pst;
-  ccx->builder = nullptr;
   pst->context = ccx;
 
   return pst;
@@ -1509,13 +1547,15 @@  struct ctf_tid_and_type
 /* Setup partial_symtab's describing each source file for which
    debugging information is available.  */
 
-static void
+static ctf_psymtab *
 scan_partial_symbols (ctf_dict_t *cfp, psymtab_storage *partial_symtabs,
-		      struct objfile *of)
+		      struct ctf_per_tu_data *tup)
 {
+  struct objfile *of = tup->of;
   bfd *abfd = of->obfd;
   const char *name = bfd_get_filename (abfd);
-  ctf_psymtab *pst = create_partial_symtab (name, cfp, partial_symtabs, of);
+  ctf_psymtab *pst = create_partial_symtab (name, tup->arc, cfp,
+			partial_symtabs, of);
 
   struct ctf_context *ccx = pst->context;
 
@@ -1530,46 +1570,46 @@  struct ctf_tid_and_type
   /* Scan CTF object and function sections which correspond to each
      STT_FUNC or STT_OBJECT entry in the symbol table,
      pick up what init_symtab has done.  */
-  for (unsigned long idx = 0; ; idx++)
-    {
-      ctf_id_t tid;
-      if ((tid = ctf_lookup_by_symbol (cfp, idx)) == CTF_ERR)
-	{
-	if (ctf_errno (cfp) == EINVAL || ctf_errno (cfp) == ECTF_NOSYMTAB)
-	  break;	// Done, reach end of the section.
-	else
-	  continue;
-	}
-      const char *tname = ctf_type_name_raw (cfp, tid);
-      uint32_t kind = ctf_type_kind (cfp, tid);
-      address_class aclass;
-      domain_enum tdomain;
-      switch (kind)
-	{
-	  case CTF_K_STRUCT:
-	  case CTF_K_UNION:
-	  case CTF_K_ENUM:
-	    tdomain = STRUCT_DOMAIN;
-	    break;
-	  default:
-	    tdomain = VAR_DOMAIN;
-	    break;
-	}
+  ctf_psymtab_add_stt_obj (cfp, pst, of);
+  ctf_psymtab_add_stt_func (cfp, pst, of);
 
-      if (kind == CTF_K_FUNCTION)
-	aclass = LOC_STATIC;
-      else if (kind == CTF_K_CONST)
-	aclass = LOC_CONST;
-      else
-	aclass = LOC_TYPEDEF;
+  pst->end ();
 
-      pst->add_psymbol (tname, true,
-			tdomain, aclass, -1,
-			psymbol_placement::STATIC,
-			0, language_c, partial_symtabs, of);
+  return pst;
+}
+
+/* Callback to build the psymtab for archive member NAME.  */
+
+static int
+build_ctf_archive_member (ctf_dict_t *ctf, const char *name, void *arg)
+{
+  struct ctf_per_tu_data *tup = (struct ctf_per_tu_data *) arg;
+  ctf_dict_t *parent = tup->fp;
+  ctf_psymtab *pst;
+  int isparent = 0;
+
+  if (strcmp (name, ".ctf") != 0)
+    {
+      ctf_import (ctf, parent);
     }
+  else
+    isparent = 1;
 
-  pst->end ();
+  if (info_verbose)
+    {
+      printf_filtered (_("Scanning archive member %s..."), name);
+      gdb_flush (gdb_stdout);
+    }
+
+  psymtab_storage *pss = tup->psf->get_partial_symtabs ().get ();
+  pst = scan_partial_symbols (ctf, pss, tup);
+
+  if (isparent)
+    tup->parent_psymtab = pst;
+  else
+    tup->imported_symtabs.push_back (pst);
+
+  return 0;
 }
 
 /* Read CTF debugging information from a BFD section.  This is
@@ -1579,6 +1619,7 @@  struct ctf_tid_and_type
 void
 elfctf_build_psymtabs (struct objfile *of)
 {
+  struct ctf_per_tu_data pcu;
   bfd *abfd = of->obfd;
   int err;
 
@@ -1593,10 +1634,34 @@  struct ctf_tid_and_type
 	   bfd_get_filename (abfd), ctf_errmsg (err));
   ctf_dict_key.emplace (of, fp);
 
+  pcu.fp = fp;
+  pcu.of = of;
+  pcu.arc = arc;
+
   psymbol_functions *psf = new psymbol_functions ();
   psymtab_storage *partial_symtabs = psf->get_partial_symtabs ().get ();
   of->qf.emplace_front (psf);
-  scan_partial_symbols (fp, partial_symtabs, of);
+  pcu.psf = psf;
+
+  if (ctf_archive_iter (arc, build_ctf_archive_member, &pcu) < 0)
+    error (_("ctf_archive_iter failed in input file %s: - %s"),
+	   bfd_get_filename (abfd), ctf_errmsg (err));
+
+  int arch_cnt = pcu.imported_symtabs.size ();
+  struct partial_symtab *parent_pst = pcu.parent_psymtab;
+
+  if (parent_pst && arch_cnt)
+    {
+      /* Fill in the 'dependencies'.  */
+      parent_pst->number_of_dependencies = arch_cnt;
+      parent_pst->dependencies
+	= partial_symtabs->allocate_dependencies (arch_cnt);
+      for (int i = 0; i < arch_cnt; ++i)
+	{
+	  parent_pst->dependencies[i]
+	    = pcu.imported_symtabs.at (i);
+	}
+    }
 }
 
 #else
diff --git a/gdb/testsuite/gdb.ctf/cross-tu-cyclic-1.c b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-1.c
new file mode 100644
index 0000000..fe52b9e
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-1.c
@@ -0,0 +1,18 @@ 
+struct A;
+struct B
+{
+  int foo;
+  struct A *bar;
+};
+
+struct A
+{
+  long a;
+  struct B *foo;
+};
+
+static struct A *foo __attribute__((used));
+
+int main()
+{
+}
diff --git a/gdb/testsuite/gdb.ctf/cross-tu-cyclic-2.c b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-2.c
new file mode 100644
index 0000000..aa2d177
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-2.c
@@ -0,0 +1,16 @@ 
+struct B;
+struct A
+{
+  long a;
+  struct B *foo;
+  struct C *bar;
+};
+
+struct C
+{
+  struct B *foo;
+  int b;
+};
+
+static struct C *foo __attribute__((used));
+static struct A *bar __attribute__((used));
diff --git a/gdb/testsuite/gdb.ctf/cross-tu-cyclic-3.c b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-3.c
new file mode 100644
index 0000000..19947e8
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-3.c
@@ -0,0 +1,3 @@ 
+struct A { struct B *foo; };
+static struct A *a __attribute__((__used__));
+static struct A *conflicty __attribute__((__used__));
diff --git a/gdb/testsuite/gdb.ctf/cross-tu-cyclic-4.c b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-4.c
new file mode 100644
index 0000000..6e0c957
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/cross-tu-cyclic-4.c
@@ -0,0 +1,4 @@ 
+struct A { struct B *foo; };
+struct B { struct B *next; };
+static struct A *a __attribute__((__used__));
+static struct B *conflicty __attribute__((__used__));
diff --git a/gdb/testsuite/gdb.ctf/cross-tu-cyclic.exp b/gdb/testsuite/gdb.ctf/cross-tu-cyclic.exp
new file mode 100644
index 0000000..734892d
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/cross-tu-cyclic.exp
@@ -0,0 +1,43 @@ 
+# Copyright 2021 Free Software Foundation, Inc.
+
+# 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/>.
+
+# This file is a subset of ptype.exp written by Rob Savoye. (rob@cygnus.com)
+
+if [skip_ctf_tests] {
+    unsupported "no CTF debug format support, or CTF disabled in GDB"
+    return 0
+}
+
+standard_testfile cross-tu-cyclic-1.c  cross-tu-cyclic-2.c \
+	cross-tu-cyclic-3.c  cross-tu-cyclic-4.c
+
+# Using `-gt` generates full-fledged CTF debug information.
+set opts "additional_flags=-gt -Wl,--export-dynamic"
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+	  [list $srcfile $srcfile2 $srcfile3 $srcfile4] \
+	  [list $opts nowarnings]] } {
+    return 0
+}
+
+# Create and source the file that provides information about the compiler
+# used to compile the test case.
+if [get_compiler_info] {
+    return -1
+}
+
+# Same thing with struct and union.
+gdb_test "ptype struct A" "type = struct A \{\[\r\n\]+\[ \t\]+long a;\[\r\n\]+\[ \t\]+struct B \\*foo;\[\r\n\]+\}.*" "ptype structure A"
+gdb_test "ptype struct B" "type = struct B \{\[\r\n\]+\[ \t\]+int foo;\[\r\n\]+\[ \t\]+struct A \\*bar;\[\r\n\]+\}.*" "ptype structure B"
+gdb_test "ptype struct C" "type = struct C \{\[\r\n\]+\[ \t\]+struct B \\*foo;\[\r\n\]+\[ \t\]+int b;\[\r\n\]+\}.*" "ptype structure C"
diff --git a/gdb/testsuite/gdb.ctf/ctf-a.c b/gdb/testsuite/gdb.ctf/ctf-a.c
new file mode 100644
index 0000000..9aa2a8f
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/ctf-a.c
@@ -0,0 +1,32 @@ 
+/* This test program is part of GDB, the GNU debugger.
+
+   Copyright 2021 Free Software Foundation, Inc.
+
+   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 "ctf-a.h"
+
+static struct A a __attribute__((used));
+
+extern struct C *foo ();
+extern int bar ();
+
+int main ()
+{
+  struct C *cp;
+  cp = foo ();
+  if (cp)
+    return bar ();
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.ctf/ctf-a.h b/gdb/testsuite/gdb.ctf/ctf-a.h
new file mode 100644
index 0000000..297d740
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/ctf-a.h
@@ -0,0 +1,22 @@ 
+/* This test program is part of GDB, the GNU debugger.
+
+   Copyright 2021 Free Software Foundation, Inc.
+
+   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/>.  */
+
+struct A {
+  struct B *b;
+  struct A *next;
+};
+
diff --git a/gdb/testsuite/gdb.ctf/ctf-b.c b/gdb/testsuite/gdb.ctf/ctf-b.c
new file mode 100644
index 0000000..c3a8ce5
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/ctf-b.c
@@ -0,0 +1,25 @@ 
+/* This test program is part of GDB, the GNU debugger.
+
+   Copyright 2021 Free Software Foundation, Inc.
+
+   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 "ctf-b.h"
+
+static struct B b __attribute__((used));
+
+int bar ()
+{
+  return b.wombat;
+}
diff --git a/gdb/testsuite/gdb.ctf/ctf-b.h b/gdb/testsuite/gdb.ctf/ctf-b.h
new file mode 100644
index 0000000..9dbdd7d
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/ctf-b.h
@@ -0,0 +1,22 @@ 
+/* This test program is part of GDB, the GNU debugger.
+
+   Copyright 2021 Free Software Foundation, Inc.
+
+   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/>.  */
+
+struct B {
+  struct C *c;
+  int wombat;
+};
+
diff --git a/gdb/testsuite/gdb.ctf/ctf-c.c b/gdb/testsuite/gdb.ctf/ctf-c.c
new file mode 100644
index 0000000..b4051b3
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/ctf-c.c
@@ -0,0 +1,25 @@ 
+/* This test program is part of GDB, the GNU debugger.
+
+   Copyright 2021 Free Software Foundation, Inc.
+
+   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 "ctf-c.h"
+
+static struct C c __attribute__((used));
+
+struct C * foo ()
+{
+  return &c;
+}
diff --git a/gdb/testsuite/gdb.ctf/ctf-c.h b/gdb/testsuite/gdb.ctf/ctf-c.h
new file mode 100644
index 0000000..fb18157
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/ctf-c.h
@@ -0,0 +1,21 @@ 
+/* This test program is part of GDB, the GNU debugger.
+
+   Copyright 2021 Free Software Foundation, Inc.
+
+   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/>.  */
+
+struct C {
+  struct A *a;
+  int b;
+};
diff --git a/gdb/testsuite/gdb.ctf/multi.exp b/gdb/testsuite/gdb.ctf/multi.exp
new file mode 100644
index 0000000..973115e
--- /dev/null
+++ b/gdb/testsuite/gdb.ctf/multi.exp
@@ -0,0 +1,42 @@ 
+# Copyright 2021 Free Software Foundation, Inc.
+
+# 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/>.
+
+# This file is a subset of ptype.exp written by Rob Savoye. (rob@cygnus.com)
+
+if [skip_ctf_tests] {
+    unsupported "no CTF debug format support, or CTF disabled in GDB"
+    return 0
+}
+
+standard_testfile ctf-a.c ctf-b.c ctf-c.c
+
+# Using `-gt` generates full-fledged CTF debug information.
+set opts "additional_flags=-gt -Wl,--export-dynamic"
+if { [prepare_for_testing "failed to prepare" ${testfile} \
+	  [list $srcfile $srcfile2 $srcfile3] \
+	  [list $opts nowarnings]] } {
+    return 0
+}
+
+# Create and source the file that provides information about the compiler
+# used to compile the test case.
+if [get_compiler_info] {
+    return -1
+}
+
+# Same thing with struct and union.
+gdb_test "ptype struct A" "type = struct A \{\[\r\n\]+\[ \t\]+struct B \\*b;\[\r\n\]+\[ \t\]+struct A \\*next;\[\r\n\]+\}.*" "ptype structure A"
+gdb_test "ptype struct B" "type = struct B \{\[\r\n\]+\[ \t\]+struct C \\*c;\[\r\n\]+\[ \t\]+int \\wombat;\[\r\n\]+\}.*" "ptype structure B"
+gdb_test "ptype struct C" "type = struct C \{\[\r\n\]+\[ \t\]+struct A \\*a;\[\r\n\]+\[ \t\]+int b;\[\r\n\]+\}.*" "ptype structure C"