[PATCH 0/7] [gdb/build] Fix -std=c++20 issues

[PATCH 0/7] [gdb/build] Fix -std=c++20 issues

[Tom de Vries]

I tried building gdb using -std=c++20, an ran into a few issues.

This series contains fixes for most of them.

I ran into a problem in a libguile header file, I've worked around that by
disabling guile support.  I'll file a PR about this.

The final patch, "[gdb/build, c++20] Handle deprecated
std::allocator::construct", exposes two problems where a gdb::byte_vector is
returned and consequently copied.

I have fix for one of those, patch "[gdb/build] Return reference in
target_read_auxv".

But the other one, in remote_target::thread_info_to_thread_handle remains
unfixed:
...
gdb::byte_vector
remote_target::thread_info_to_thread_handle (struct thread_info *tp)
{
  remote_thread_info *priv = get_remote_thread_info (tp);
  return priv->thread_handle;
}
...

I wrote a patch for this, which allowed me to finish the build with
-std=c++20, but it causes a regression in gdb.python/py-thrhandle.exp, so it's
not included in this series.

Tested on x86_64-linux, with:
- gcc 7.5.0 -std=c++11, and
- gcc 12.3.0 -std=c++17.

Build on x86_64-linux, with:
- gcc 12.3.0 -std=c++20, which runs into the single error mentioned above.

Tom de Vries (7):
  [gdb/build, c++20] Fix Wdeprecated-enum-enum-conversion
  [gdb/build, c++20] Stop using deprecated is_pod
  [gdb/build, c++20] Fix DISABLE_COPY_AND_ASSIGN use in ui_out_emit_type
  [gdb/build, c++20] Fix deprecated implicit capture of this
  [gdb/build, c++20] Fix invalid conversion in test_symbols
  [gdb/build] Return reference in target_read_auxv
  [gdb/build, c++20] Handle deprecated std::allocator::construct

 gdb/ada-lang.c                  |  2 +-
 gdb/arm-fbsd-tdep.c             |  2 +-
 gdb/auxv.c                      |  6 +++---
 gdb/auxv.h                      |  2 +-
 gdb/dwarf2/read.c               |  2 +-
 gdb/ravenscar-thread.c          |  2 +-
 gdb/remote.c                    | 14 +++++++++-----
 gdb/rs6000-tdep.c               |  3 ++-
 gdb/ui-out.h                    |  2 +-
 gdb/unittests/ptid-selftests.c  |  4 +++-
 gdbsupport/common-exceptions.h  |  4 +++-
 gdbsupport/default-init-alloc.h | 13 ++++++++++++-
 gdbsupport/poison.h             |  2 +-
 13 files changed, 39 insertions(+), 19 deletions(-)


base-commit: 2a3f442df9e9a71f9e3c4c5999a2c53bf93e488d
-- 
2.35.3

[PATCH 1/7] [gdb/build, c++20] Fix Wdeprecated-enum-enum-conversion

[Tom de Vries]

When building gdb with clang 15 and -std=c++20, I run into:
...
gdbsupport/common-exceptions.h:203:32: error: arithmetic between different \
  enumeration types ('const enum return_reason' and 'const enum errors') is \
  deprecated [-Werror,-Wdeprecated-enum-enum-conversion]
    size_t result = exc.reason + exc.error;
                    ~~~~~~~~~~ ^ ~~~~~~~~~
...

Fix this by using to_underlying.

Likewise in a few other places.

Tested on x86_64-linux.
---
 gdb/remote.c                   | 12 ++++++++----
 gdb/rs6000-tdep.c              |  3 ++-
 gdbsupport/common-exceptions.h |  4 +++-
 3 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/gdb/remote.c b/gdb/remote.c
index 6fefabac0ce..e01da795ec8 100644
--- a/gdb/remote.c
+++ b/gdb/remote.c
@@ -10850,7 +10850,8 @@ remote_target::insert_watchpoint (CORE_ADDR addr, int len,
   char *p;
   enum Z_packet_type packet = watchpoint_to_Z_packet (type);
 
-  if (m_features.packet_support (PACKET_Z0 + packet) == PACKET_DISABLE)
+  if (m_features.packet_support ((to_underlying (PACKET_Z0)
+				  + to_underlying (packet))) == PACKET_DISABLE)
     return 1;
 
   /* Make sure the remote is pointing at the right process, if
@@ -10867,7 +10868,8 @@ remote_target::insert_watchpoint (CORE_ADDR addr, int len,
   putpkt (rs->buf);
   getpkt (&rs->buf, 0);
 
-  switch (m_features.packet_ok (rs->buf, PACKET_Z0 + packet))
+  switch (m_features.packet_ok (rs->buf, (to_underlying (PACKET_Z0)
+					  + to_underlying (packet))))
     {
     case PACKET_ERROR:
       return -1;
@@ -10898,7 +10900,8 @@ remote_target::remove_watchpoint (CORE_ADDR addr, int len,
   char *p;
   enum Z_packet_type packet = watchpoint_to_Z_packet (type);
 
-  if (m_features.packet_support (PACKET_Z0 + packet) == PACKET_DISABLE)
+  if (m_features.packet_support ((to_underlying (PACKET_Z0)
+				  + to_underlying (packet))) == PACKET_DISABLE)
     return -1;
 
   /* Make sure the remote is pointing at the right process, if
@@ -10914,7 +10917,8 @@ remote_target::remove_watchpoint (CORE_ADDR addr, int len,
   putpkt (rs->buf);
   getpkt (&rs->buf, 0);
 
-  switch (m_features.packet_ok (rs->buf, PACKET_Z0 + packet))
+  switch (m_features.packet_ok (rs->buf, (to_underlying (PACKET_Z0)
+					  + to_underlying (packet))))
     {
     case PACKET_ERROR:
     case PACKET_UNKNOWN:
diff --git a/gdb/rs6000-tdep.c b/gdb/rs6000-tdep.c
index ff83743da36..71390513ad2 100644
--- a/gdb/rs6000-tdep.c
+++ b/gdb/rs6000-tdep.c
@@ -2495,7 +2495,8 @@ rs6000_register_name (struct gdbarch *gdbarch, int regno)
 
   /* Hide the upper halves of the cvs0~cvs31 registers.  */
   if (PPC_CVSR0_UPPER_REGNUM <= regno
-      && regno < PPC_CVSR0_UPPER_REGNUM + ppc_num_gprs)
+      && regno < (to_underlying (PPC_CVSR0_UPPER_REGNUM)
+		  + to_underlying (ppc_num_gprs)))
     return "";
 
   /* Check if the SPE pseudo registers are available.  */
diff --git a/gdbsupport/common-exceptions.h b/gdbsupport/common-exceptions.h
index f9b59ece736..e4750211448 100644
--- a/gdbsupport/common-exceptions.h
+++ b/gdbsupport/common-exceptions.h
@@ -26,6 +26,8 @@
 #include <string>
 #include <functional>
 
+#include "gdbsupport/underlying.h"
+
 /* Reasons for calling throw_exceptions().  NOTE: all reason values
    must be different from zero.  enum value 0 is reserved for internal
    use as the return value from an initial setjmp().  */
@@ -200,7 +202,7 @@ struct hash<gdb_exception>
 {
   size_t operator() (const gdb_exception &exc) const
   {
-    size_t result = exc.reason + exc.error;
+    size_t result = to_underlying (exc.reason) + to_underlying (exc.error);
     if (exc.message != nullptr)
       result += std::hash<std::string> {} (*exc.message);
     return result;
-- 
2.35.3

[PATCH 2/7] [gdb/build, c++20] Stop using deprecated is_pod

[Tom de Vries]

When building gdb with clang 15 and -std=c++20, I run into:
...
gdbsupport/poison.h:52:11: error: 'is_pod<timeval>' is deprecated: use \
  is_standard_layout && is_trivial instead [-Werror,-Wdeprecated-declarations]
            std::is_pod<T>>
                 ^
...

Fix this by following the suggestion.

Likewise in gdb/unittests/ptid-selftests.c.

Tested on x86_64-linux.
---
 gdb/unittests/ptid-selftests.c | 4 +++-
 gdbsupport/poison.h            | 2 +-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/gdb/unittests/ptid-selftests.c b/gdb/unittests/ptid-selftests.c
index 5af73b0da53..2713b619741 100644
--- a/gdb/unittests/ptid-selftests.c
+++ b/gdb/unittests/ptid-selftests.c
@@ -29,7 +29,9 @@ namespace ptid {
    This is a requirement for as long as we have ptids embedded in
    structures allocated with malloc. */
 
-static_assert (std::is_pod<ptid_t>::value, "ptid_t is POD");
+static_assert (gdb::And<std::is_standard_layout<ptid_t>,
+			std::is_trivial<ptid_t>>::value,
+	       "ptid_t is POD");
 
 /* We want to avoid implicit conversion from int to ptid_t.  */
 
diff --git a/gdbsupport/poison.h b/gdbsupport/poison.h
index 956c3d856d8..63fccb30cb9 100644
--- a/gdbsupport/poison.h
+++ b/gdbsupport/poison.h
@@ -49,7 +49,7 @@ be a compile-time error.  */
 template<typename T>
 struct IsMemsettable
   : gdb::Or<std::is_void<T>,
-	    std::is_pod<T>>
+	    gdb::And<std::is_standard_layout<T>, std::is_trivial<T>>>
 {};
 
 template <typename T,
-- 
2.35.3

[PATCH 3/7] [gdb/build, c++20] Fix DISABLE_COPY_AND_ASSIGN use in ui_out_emit_type

[Tom de Vries]

When building gdb with -std=c++20, I run into:
...
include/ansidecl.h:342:9: error: expected unqualified-id before ‘const’
  342 |   TYPE (const TYPE&) = delete;                  \
      |         ^~~~~
gdb/ui-out.h:412:3: note: in expansion of macro ‘DISABLE_COPY_AND_ASSIGN’
  412 |   DISABLE_COPY_AND_ASSIGN (ui_out_emit_type<Type>);
      |   ^~~~~~~~~~~~~~~~~~~~~~~
...

Fix this by using "DISABLE_COPY_AND_ASSIGN (ui_out_emit_type)".

Tested on x86_64-linux.
---
 gdb/ui-out.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/gdb/ui-out.h b/gdb/ui-out.h
index ba5b1de68ab..07567a1df35 100644
--- a/gdb/ui-out.h
+++ b/gdb/ui-out.h
@@ -409,7 +409,7 @@ class ui_out_emit_type
     m_uiout->end (Type);
   }
 
-  DISABLE_COPY_AND_ASSIGN (ui_out_emit_type<Type>);
+  DISABLE_COPY_AND_ASSIGN (ui_out_emit_type);
 
 private:
 
-- 
2.35.3

[PATCH 4/7] [gdb/build, c++20] Fix deprecated implicit capture of this

[Tom de Vries]

When building gdb with -std=c++20 I run into:
...
gdb/ada-lang.c:10713:16: error: implicit capture of ‘this’ via ‘[=]’ is \
  deprecated in C++20 [-Werror=deprecated]
10713 |   auto do_op = [=] (LONGEST x, LONGEST y)
      |                ^
gdb/ada-lang.c:10713:16: note: add explicit ‘this’ or ‘*this’ capture
...

Fix this by using "[this]".

Likewise in two more spots.

Tested on x86_64-linux.
---
 gdb/ada-lang.c         | 2 +-
 gdb/ravenscar-thread.c | 2 +-
 gdb/remote.c           | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/gdb/ada-lang.c b/gdb/ada-lang.c
index 1261ee8fa05..018adfcfb70 100644
--- a/gdb/ada-lang.c
+++ b/gdb/ada-lang.c
@@ -10710,7 +10710,7 @@ ada_binop_addsub_operation::evaluate (struct type *expect_type,
   value *arg1 = std::get<1> (m_storage)->evaluate_with_coercion (exp, noside);
   value *arg2 = std::get<2> (m_storage)->evaluate_with_coercion (exp, noside);
 
-  auto do_op = [=] (LONGEST x, LONGEST y)
+  auto do_op = [this] (LONGEST x, LONGEST y)
     {
       if (std::get<0> (m_storage) == BINOP_ADD)
 	return x + y;
diff --git a/gdb/ravenscar-thread.c b/gdb/ravenscar-thread.c
index 52016133889..68d336bee99 100644
--- a/gdb/ravenscar-thread.c
+++ b/gdb/ravenscar-thread.c
@@ -473,7 +473,7 @@ ravenscar_thread_target::update_thread_list ()
      (m_base_ptid) and the running thread, that may not have been included
      to system.tasking.debug's list yet.  */
 
-  iterate_over_live_ada_tasks ([=] (struct ada_task_info *task)
+  iterate_over_live_ada_tasks ([this] (struct ada_task_info *task)
 			       {
 				 this->add_thread (task);
 			       });
diff --git a/gdb/remote.c b/gdb/remote.c
index e01da795ec8..7abe08439b5 100644
--- a/gdb/remote.c
+++ b/gdb/remote.c
@@ -11609,7 +11609,7 @@ remote_target::search_memory (CORE_ADDR start_addr, ULONGEST search_space_len,
   int found;
   ULONGEST found_addr;
 
-  auto read_memory = [=] (CORE_ADDR addr, gdb_byte *result, size_t len)
+  auto read_memory = [this] (CORE_ADDR addr, gdb_byte *result, size_t len)
     {
       return (target_read (this, TARGET_OBJECT_MEMORY, NULL, result, addr, len)
 	      == len);
-- 
2.35.3

[PATCH 5/7] [gdb/build, c++20] Fix invalid conversion in test_symbols

[Tom de Vries]

When building gdb with -std=c++20, I run into:
...
gdb/dwarf2/read.c:2709:3: error: invalid conversion from ‘const char8_t*’ to \
  ‘const char*’ [-fpermissive]
 2709 |   u8"u8função",
      |   ^~~~~~~~~~~~
      |   |
      |   const char8_t*
...

Fix this by making the conversion explicit.

Tested on x86_64-linux.
---
 gdb/dwarf2/read.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c
index a64f82bd24a..5b13bbda31d 100644
--- a/gdb/dwarf2/read.c
+++ b/gdb/dwarf2/read.c
@@ -2706,7 +2706,7 @@ static const char *test_symbols[] = {
   /* A UTF-8 name with multi-byte sequences to make sure that
      cp-name-parser understands this as a single identifier ("função"
      is "function" in PT).  */
-  u8"u8função",
+  (const char *)u8"u8função",
 
   /* Test a symbol name that ends with a 0xff character, which is a
      valid character in non-UTF-8 source character sets (e.g. Latin1
-- 
2.35.3

[PATCH 6/7] [gdb/build] Return reference in target_read_auxv

[Tom de Vries]

In target_read_auxv we return a copy of an object:
...
gdb::optional<gdb::byte_vector>
target_read_auxv ()
{
  ...
  return info->data;
}
...

Return a reference instead, saving a copy.

This is exposed by using std::pmr::polymorphic_allocator instead of
std::allocator in default_init_allocator.

Tested on x86_64-linux.
---
 gdb/arm-fbsd-tdep.c | 2 +-
 gdb/auxv.c          | 6 +++---
 gdb/auxv.h          | 2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/gdb/arm-fbsd-tdep.c b/gdb/arm-fbsd-tdep.c
index acafc4a8bcd..fe1c7f453c9 100644
--- a/gdb/arm-fbsd-tdep.c
+++ b/gdb/arm-fbsd-tdep.c
@@ -244,7 +244,7 @@ arm_fbsd_read_description_auxv (const gdb::optional<gdb::byte_vector> &auxv,
 const struct target_desc *
 arm_fbsd_read_description_auxv (bool tls)
 {
-  gdb::optional<gdb::byte_vector> auxv = target_read_auxv ();
+  gdb::optional<gdb::byte_vector> &auxv = target_read_auxv ();
   return arm_fbsd_read_description_auxv (auxv,
 					 current_inferior ()->top_target (),
 					 current_inferior ()->gdbarch,
diff --git a/gdb/auxv.c b/gdb/auxv.c
index 812b2807554..02b4ab4cdd0 100644
--- a/gdb/auxv.c
+++ b/gdb/auxv.c
@@ -354,7 +354,7 @@ invalidate_auxv_cache (void)
 
 /* See auxv.h.  */
 
-gdb::optional<gdb::byte_vector>
+gdb::optional<gdb::byte_vector> &
 target_read_auxv ()
 {
   inferior *inf = current_inferior ();
@@ -410,7 +410,7 @@ target_auxv_search (const gdb::byte_vector &auxv, target_ops *ops,
 int
 target_auxv_search (CORE_ADDR match, CORE_ADDR *valp)
 {
-  gdb::optional<gdb::byte_vector> auxv = target_read_auxv ();
+  gdb::optional<gdb::byte_vector> &auxv = target_read_auxv ();
 
   if (!auxv.has_value ())
     return -1;
@@ -564,7 +564,7 @@ fprint_target_auxv (struct ui_file *file)
   struct gdbarch *gdbarch = target_gdbarch ();
   CORE_ADDR type, val;
   int ents = 0;
-  gdb::optional<gdb::byte_vector> auxv = target_read_auxv ();
+  gdb::optional<gdb::byte_vector> &auxv = target_read_auxv ();
 
   if (!auxv.has_value ())
     return -1;
diff --git a/gdb/auxv.h b/gdb/auxv.h
index 2441b6c02ab..f42efa4fa7b 100644
--- a/gdb/auxv.h
+++ b/gdb/auxv.h
@@ -48,7 +48,7 @@ extern int svr4_auxv_parse (struct gdbarch *gdbarch, const gdb_byte **readptr,
 
 /* Read auxv data from the current inferior's target stack.  */
 
-extern gdb::optional<gdb::byte_vector> target_read_auxv ();
+extern gdb::optional<gdb::byte_vector> &target_read_auxv ();
 
 /* Read auxv data from OPS.  */
 
-- 
2.35.3

[PATCH 7/7] [gdb/build, c++20] Handle deprecated std::allocator::construct

[Tom de Vries]

When building gdb with -std=c++20, I run into:
...
gdbsupport/default-init-alloc.h:52:12: error: ‘construct’ has not been \
  declared in ‘class std::allocator<unsigned char>’
   52 |   using A::construct;
      |            ^~~~~~~~~
...

Indeed, std::allocator::construct has been deprecated in c++17 and removed in
c++20.

Fix this by using instead std::pmr::polymorphic_allocator for c++20.

Tested on x86_64-linux.
---
 gdbsupport/default-init-alloc.h | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/gdbsupport/default-init-alloc.h b/gdbsupport/default-init-alloc.h
index 9c8b5a2c689..e2d3bd24516 100644
--- a/gdbsupport/default-init-alloc.h
+++ b/gdbsupport/default-init-alloc.h
@@ -18,6 +18,10 @@
 #ifndef COMMON_DEFAULT_INIT_ALLOC_H
 #define COMMON_DEFAULT_INIT_ALLOC_H
 
+#if __cplusplus >= 202002L
+#include <memory_resource>
+#endif
+
 namespace gdb {
 
 /* An allocator that default constructs using default-initialization
@@ -29,7 +33,14 @@ namespace gdb {
    adapter that given an allocator A, overrides 'A::construct()'.  'A'
    defaults to std::allocator<T>.  */
 
-template<typename T, typename A = std::allocator<T>>
+template<typename T,
+	 typename A
+#if __cplusplus >= 202002L
+	 = std::pmr::polymorphic_allocator<T>
+#else
+	 = std::allocator<T>
+#endif
+	 >
 class default_init_allocator : public A
 {
 public:
-- 
2.35.3

Re: [PATCH 6/7] [gdb/build] Return reference in target_read_auxv

[Tom Tromey]

>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:

Tom> In target_read_auxv we return a copy of an object:
Tom> ...
Tom> gdb::optional<gdb::byte_vector>
Tom> target_read_auxv ()
Tom> {
Tom>   ...
Tom>   return info->data;
Tom> }
Tom> ...

Tom> Return a reference instead, saving a copy.

I suspect it ought to return a const reference instead.

Tom

Re: [PATCH 0/7] [gdb/build] Fix -std=c++20 issues

[Tom Tromey]

>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:

Tom> I tried building gdb using -std=c++20, an ran into a few issues.
Tom> This series contains fixes for most of them.

Thanks for doing this.  I read these & sent one comment.
The rest look ok to me.

Tom> But the other one, in remote_target::thread_info_to_thread_handle remains
Tom> unfixed:
Tom> ...
Tom> gdb::byte_vector
Tom> remote_target::thread_info_to_thread_handle (struct thread_info *tp)
Tom> {
Tom>   remote_thread_info *priv = get_remote_thread_info (tp);
Tom>   return priv->thread_handle;

First, I wonder if we even care about this copying.

But if we do, it could return a const reference and linux-thread-db
could be modified to cache the byte_vector in thread_db_thread_info.

Tom

Re: [PATCH 0/7] [gdb/build] Fix -std=c++20 issues

[Tom de Vries]

On 8/16/23 20:27, Tom Tromey wrote:
>>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> Tom> I tried building gdb using -std=c++20, an ran into a few issues.
> Tom> This series contains fixes for most of them.
> 
> Thanks for doing this.  I read these & sent one comment.
> The rest look ok to me.
> 
> Tom> But the other one, in remote_target::thread_info_to_thread_handle remains
> Tom> unfixed:
> Tom> ...
> Tom> gdb::byte_vector
> Tom> remote_target::thread_info_to_thread_handle (struct thread_info *tp)
> Tom> {
> Tom>   remote_thread_info *priv = get_remote_thread_info (tp);
> Tom>   return priv->thread_handle;
> 
> First, I wonder if we even care about this copying.
> 

Well, I care because I run into a build error:
...
In file included from /usr/include/c++/12/ext/alloc_traits.h:34,
                  from /usr/include/c++/12/unordered_map:41,
                  from /usr/include/c++/12/functional:61,
                  from /data/vries/gdb/src/gdb/../gdbsupport/ptid.h:35,
                  from 
/data/vries/gdb/src/gdb/../gdbsupport/common-defs.h:206,
                  from /data/vries/gdb/src/gdb/defs.h:26,
                  from /data/vries/gdb/src/gdb/remote.c:22:
/usr/include/c++/12/bits/alloc_traits.h: In instantiation of ‘static 
constexpr _Alloc std::allocator_traits< <template-parameter-1-1> 
 >::select_on_container_copy_construction(const _Alloc&) [with _Alloc = 
gdb::default_init_allocator<unsigned char, 
std::pmr::polymorphic_allocator<unsigned char> >]’:
/usr/include/c++/12/ext/alloc_traits.h:98:63:   required from ‘static 
constexpr _Alloc __gnu_cxx::__alloc_traits<_Alloc, 
<template-parameter-1-2> >::_S_select_on_copy(const _Alloc&) [with 
_Alloc = gdb::default_init_allocator<unsigned char, 
std::pmr::polymorphic_allocator<unsigned char> >; 
<template-parameter-1-2> = unsigned char]’
/usr/include/c++/12/bits/stl_vector.h:598:34:   required from ‘constexpr 
std::vector<_Tp, _Alloc>::vector(const std::vector<_Tp, _Alloc>&) [with 
_Tp = unsigned char; _Alloc = gdb::default_init_allocator<unsigned char, 
std::pmr::polymorphic_allocator<unsigned char> >]’
/data/vries/gdb/src/gdb/remote.c:14557:16:   required from here
/usr/include/c++/12/bits/alloc_traits.h:402:25: error: could not convert 
‘std::allocator_traits<gdb::default_init_allocator<unsigned char, 
std::pmr::polymorphic_allocator<unsigned char> > >::_S_select<const 
gdb::default_init_allocator<unsigned char, 
std::pmr::polymorphic_allocator<unsigned char> > >((* & __rhs), 0)’ from 
‘std::pmr::polymorphic_allocator<unsigned char>’ to 
‘gdb::default_init_allocator<unsigned char, 
std::pmr::polymorphic_allocator<unsigned char> >’
   402 |       { return _S_select(__rhs, 0); }
       |                ~~~~~~~~~^~~~~~~~~~
       |                         |
       | 
std::pmr::polymorphic_allocator<unsigned char>
...

> But if we do, it could return a const reference and linux-thread-db
> could be modified to cache the byte_vector in thread_db_thread_info.

I've managed to get that working, submitted here ( 
https://sourceware.org/pipermail/gdb-patches/2023-August/201727.html ).

Thanks,
- Tom