[PATCH 0/4] Small Misc psymtabs / objfile cleanups

[PATCH 0/4] Small Misc psymtabs / objfile cleanups

[Simon Marchi]

I spotted these while looking at Tom's psymtabs series, nothing major.

Simon Marchi (4):
  gdb: add intern method to objfile_per_bfd_storage
  gdb: use std::string in partial_symtab::partial_symtab /
    allocate_symtab
  gdb: pass objfile_per_bfd_storage instead of objfile to partial_symtab
  gdb: remove objfile parameter from get_objfile_bfd_data

 gdb/ctfread.c     |  6 +++---
 gdb/dbxread.c     |  4 ++--
 gdb/dwarf2/read.c | 18 ++++++++----------
 gdb/dwarf2/read.h |  4 ++--
 gdb/mdebugread.c  |  6 +++---
 gdb/objfiles.c    |  8 ++++----
 gdb/objfiles.h    | 29 ++++++++++++++++++++++++-----
 gdb/psympriv.h    | 23 ++++++++++++-----------
 gdb/psymtab.c     | 21 ++++++++++-----------
 gdb/symfile.c     | 11 +++++------
 gdb/xcoffread.c   |  4 ++--
 11 files changed, 75 insertions(+), 59 deletions(-)

-- 
2.30.0

[PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Simon Marchi]

From: Simon Marchi <simon.marchi@polymtl.ca>

This allows factoring out the internal implementation details that are
currently in the two objfile::intern methods.

gdb/ChangeLog:

	* objfiles.h (struct objfile_per_bfd_storage) <intern>: New
	method.
	(struct objfile) <intern>: Use
	objfile::objfile_per_bfd_storage::intern.

Change-Id: Ifd54026c5efaeffafac9b84ff84c199acc7ce78a
---
 gdb/objfiles.h | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/gdb/objfiles.h b/gdb/objfiles.h
index 41f8fc913d8..d9aa06636f5 100644
--- a/gdb/objfiles.h
+++ b/gdb/objfiles.h
@@ -270,6 +270,14 @@ struct objfile_per_bfd_storage
 
   ~objfile_per_bfd_storage ();
 
+  /* Intern STRING in this object's string cache and return the unique copy.
+     The copy has the same lifetime as this object.  */
+
+  const char *intern (gdb::string_view str)
+  {
+    return (const char *) string_cache.insert (str.data (), str.size () + 1);
+  }
+
   /* The storage has an obstack of its own.  */
 
   auto_obstack storage_obstack;
@@ -516,15 +524,14 @@ struct objfile
      lifetime as the per-BFD object.  */
   const char *intern (const char *str)
   {
-    return (const char *) per_bfd->string_cache.insert (str, strlen (str) + 1);
+    return per_bfd->intern (str);
   }
 
   /* Intern STRING and return the unique copy.  The copy has the same
      lifetime as the per-BFD object.  */
   const char *intern (const std::string &str)
   {
-    return (const char *) per_bfd->string_cache.insert (str.c_str (),
-							str.size () + 1);
+    return per_bfd->intern (str);
   }
 
   /* Retrieve the gdbarch associated with this objfile.  */
-- 
2.30.0

Re: [PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Christian Biesinger]

On Mon, Mar 22, 2021 at 8:34 PM Simon Marchi via Gdb-patches
<gdb-patches@sourceware.org> wrote:
>
> From: Simon Marchi <simon.marchi@polymtl.ca>
>
> This allows factoring out the internal implementation details that are
> currently in the two objfile::intern methods.
>
> gdb/ChangeLog:
>
>         * objfiles.h (struct objfile_per_bfd_storage) <intern>: New
>         method.
>         (struct objfile) <intern>: Use
>         objfile::objfile_per_bfd_storage::intern.
>
> Change-Id: Ifd54026c5efaeffafac9b84ff84c199acc7ce78a
> ---
>  gdb/objfiles.h | 13 ++++++++++---
>  1 file changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/gdb/objfiles.h b/gdb/objfiles.h
> index 41f8fc913d8..d9aa06636f5 100644
> --- a/gdb/objfiles.h
> +++ b/gdb/objfiles.h
> @@ -270,6 +270,14 @@ struct objfile_per_bfd_storage
>
>    ~objfile_per_bfd_storage ();
>
> +  /* Intern STRING in this object's string cache and return the unique copy.
> +     The copy has the same lifetime as this object.  */
> +
> +  const char *intern (gdb::string_view str)
> +  {
> +    return (const char *) string_cache.insert (str.data (), str.size () + 1);

Is this guaranteed to be nullterminated even without calling c_str()?

Christian

Re: [PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Simon Marchi]

On 2021-03-23 10:47 a.m., Christian Biesinger via Gdb-patches wrote:
> On Mon, Mar 22, 2021 at 8:34 PM Simon Marchi via Gdb-patches
> <gdb-patches@sourceware.org> wrote:
>>
>> From: Simon Marchi <simon.marchi@polymtl.ca>
>>
>> This allows factoring out the internal implementation details that are
>> currently in the two objfile::intern methods.
>>
>> gdb/ChangeLog:
>>
>>         * objfiles.h (struct objfile_per_bfd_storage) <intern>: New
>>         method.
>>         (struct objfile) <intern>: Use
>>         objfile::objfile_per_bfd_storage::intern.
>>
>> Change-Id: Ifd54026c5efaeffafac9b84ff84c199acc7ce78a
>> ---
>>  gdb/objfiles.h | 13 ++++++++++---
>>  1 file changed, 10 insertions(+), 3 deletions(-)
>>
>> diff --git a/gdb/objfiles.h b/gdb/objfiles.h
>> index 41f8fc913d8..d9aa06636f5 100644
>> --- a/gdb/objfiles.h
>> +++ b/gdb/objfiles.h
>> @@ -270,6 +270,14 @@ struct objfile_per_bfd_storage
>>
>>    ~objfile_per_bfd_storage ();
>>
>> +  /* Intern STRING in this object's string cache and return the unique copy.
>> +     The copy has the same lifetime as this object.  */
>> +
>> +  const char *intern (gdb::string_view str)
>> +  {
>> +    return (const char *) string_cache.insert (str.data (), str.size () + 1);
> 
> Is this guaranteed to be nullterminated even without calling c_str()?
> 
> Christian

From what I remember from last time something like this came up (I can't
remember the context) is that we'd only use string_view for strings that
are null-terminated, even though string_view doesn't guarantee it.  Is
that the case?  Or is it a bad use case for string_view?

In this particular case, callers of this intern method have a C string
and an std::string, so we know they are null-terminated.

Simon

Re: [PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Christian Biesinger]

On Tue, Mar 23, 2021 at 4:49 PM Simon Marchi <simon.marchi@polymtl.ca> wrote:
>
> On 2021-03-23 10:47 a.m., Christian Biesinger via Gdb-patches wrote:
> > On Mon, Mar 22, 2021 at 8:34 PM Simon Marchi via Gdb-patches
> > <gdb-patches@sourceware.org> wrote:
> >>
> >> From: Simon Marchi <simon.marchi@polymtl.ca>
> >>
> >> This allows factoring out the internal implementation details that are
> >> currently in the two objfile::intern methods.
> >>
> >> gdb/ChangeLog:
> >>
> >>         * objfiles.h (struct objfile_per_bfd_storage) <intern>: New
> >>         method.
> >>         (struct objfile) <intern>: Use
> >>         objfile::objfile_per_bfd_storage::intern.
> >>
> >> Change-Id: Ifd54026c5efaeffafac9b84ff84c199acc7ce78a
> >> ---
> >>  gdb/objfiles.h | 13 ++++++++++---
> >>  1 file changed, 10 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/gdb/objfiles.h b/gdb/objfiles.h
> >> index 41f8fc913d8..d9aa06636f5 100644
> >> --- a/gdb/objfiles.h
> >> +++ b/gdb/objfiles.h
> >> @@ -270,6 +270,14 @@ struct objfile_per_bfd_storage
> >>
> >>    ~objfile_per_bfd_storage ();
> >>
> >> +  /* Intern STRING in this object's string cache and return the unique copy.
> >> +     The copy has the same lifetime as this object.  */
> >> +
> >> +  const char *intern (gdb::string_view str)
> >> +  {
> >> +    return (const char *) string_cache.insert (str.data (), str.size () + 1);
> >
> > Is this guaranteed to be nullterminated even without calling c_str()?
> >
> > Christian
>
> From what I remember from last time something like this came up (I can't
> remember the context) is that we'd only use string_view for strings that
> are null-terminated, even though string_view doesn't guarantee it.  Is
> that the case?  Or is it a bad use case for string_view?

Hm, I'm not sure we concluded that in the general case. It came up
specifically for general_symbol_info::compute_and_set_names, which can
handle the case where it may or may not be nullterminated depending on
a parameter. IMO you should at least comment that the string view must
be nullterminated here.

> In this particular case, callers of this intern method have a C string
> and an std::string, so we know they are null-terminated.

OK, makes sense.

Christian

Re: [PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Simon Marchi]

>> From what I remember from last time something like this came up (I can't
>> remember the context) is that we'd only use string_view for strings that
>> are null-terminated, even though string_view doesn't guarantee it.  Is
>> that the case?  Or is it a bad use case for string_view?
> 
> Hm, I'm not sure we concluded that in the general case. It came up
> specifically for general_symbol_info::compute_and_set_names, which can
> handle the case where it may or may not be nullterminated depending on
> a parameter. IMO you should at least comment that the string view must
> be nullterminated here.

Good idea, will do.

Simon

Re: [PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Tom Tromey]

>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:

Simon> In this particular case, callers of this intern method have a C string
Simon> and an std::string, so we know they are null-terminated.

It seems a little dangerous to do this, though, in that someone might
add a call without remembering or checking the invariant.
Maybe it's better to just have 2 overloads on the per-bfd object.

Tom

Re: [PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Simon Marchi]

On 2021-04-01 1:44 p.m., Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> Simon> In this particular case, callers of this intern method have a C string
> Simon> and an std::string, so we know they are null-terminated.
> 
> It seems a little dangerous to do this, though, in that someone might
> add a call without remembering or checking the invariant.
> Maybe it's better to just have 2 overloads on the per-bfd object.

Yeah, you're right.  I changed the patch to just move the two overloads
from objfile to objfile_per_bfd_storage.  Here's what I pushed:


From 4a4f97c129b26445ff14d0e5323feeb80610a539 Mon Sep 17 00:00:00 2001
From: Simon Marchi <simon.marchi@polymtl.ca>
Date: Fri, 2 Apr 2021 11:23:52 -0400
Subject: [PATCH] gdb: add intern methods to objfile_per_bfd_storage

This allows keeping the objfile_per_bfd_storage implementation details
into objfile_per_bfd_storage, instead of into objfile.  And this makes
the intern methods usable for code that only has an
objfile_per_bfd_storage to work with.

gdb/ChangeLog:

	* objfiles.h (struct objfile_per_bfd_storage) <intern>: New
	methods.
	(struct objfile) <intern>: Use
	objfile::objfile_per_bfd_storage::intern.

Change-Id: Ifd54026c5efaeffafac9b84ff84c199acc7ce78a
---
 gdb/ChangeLog  |  7 +++++++
 gdb/objfiles.h | 22 +++++++++++++++++++---
 2 files changed, 26 insertions(+), 3 deletions(-)

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index 1bdd652b632a..ac64f5c76099 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,10 @@
+2021-04-02  Simon Marchi  <simon.marchi@polymtl.ca>
+
+	* objfiles.h (struct objfile_per_bfd_storage) <intern>: New
+	methods.
+	(struct objfile) <intern>: Use
+	objfile::objfile_per_bfd_storage::intern.
+
 2021-04-01  Simon Marchi  <simon.marchi@efficios.com>
 
 	* gdbtypes.h (TYPE_FLAG_ENUM): Remove, replace all uses
diff --git a/gdb/objfiles.h b/gdb/objfiles.h
index 4cd392bd7f02..e8a8b5f6de78 100644
--- a/gdb/objfiles.h
+++ b/gdb/objfiles.h
@@ -270,6 +270,23 @@ struct objfile_per_bfd_storage
 
   ~objfile_per_bfd_storage ();
 
+  /* Intern STRING in this object's string cache and return the unique copy.
+     The copy has the same lifetime as this object.
+
+     STRING must be null-terminated.  */
+
+  const char *intern (const char *str)
+  {
+    return (const char *) string_cache.insert (str, strlen (str) + 1);
+  }
+
+  /* Same as the above, but for an std::string.  */
+
+  const char *intern (const std::string &str)
+  {
+    return (const char *) string_cache.insert (str.c_str (), str.size () + 1);
+  }
+
   /* The storage has an obstack of its own.  */
 
   auto_obstack storage_obstack;
@@ -516,15 +533,14 @@ struct objfile
      lifetime as the per-BFD object.  */
   const char *intern (const char *str)
   {
-    return (const char *) per_bfd->string_cache.insert (str, strlen (str) + 1);
+    return per_bfd->intern (str);
   }
 
   /* Intern STRING and return the unique copy.  The copy has the same
      lifetime as the per-BFD object.  */
   const char *intern (const std::string &str)
   {
-    return (const char *) per_bfd->string_cache.insert (str.c_str (),
-							str.size () + 1);
+    return per_bfd->intern (str);
   }
 
   /* Retrieve the gdbarch associated with this objfile.  */
-- 
2.30.1

[PATCH 2/4] gdb: use std::string in partial_symtab::partial_symtab / allocate_symtab

[Simon Marchi]

From: Simon Marchi <simon.marchi@polymtl.ca>

This simplifies the code a bit.

gdb/ChangeLog:

	* psymtab.c (partial_symtab::partial_symtab): Change
	last_objfile_name to be an std::string.
	* symfile.c (allocate_symtab): Likewise.

Change-Id: I3dfe217233ed9346c2abc04a9b1be0df69a90af8
---
 gdb/psymtab.c | 11 +++++------
 gdb/symfile.c | 11 +++++------
 2 files changed, 10 insertions(+), 12 deletions(-)

diff --git a/gdb/psymtab.c b/gdb/psymtab.c
index 597817269c1..2ee05ec415b 100644
--- a/gdb/psymtab.c
+++ b/gdb/psymtab.c
@@ -1575,16 +1575,15 @@ partial_symtab::partial_symtab (const char *filename_,
     {
       /* Be a bit clever with debugging messages, and don't print objfile
 	 every time, only when it changes.  */
-      static char *last_objfile_name = NULL;
+      static std::string last_objfile_name;
+      const char *this_objfile_name = objfile_name (objfile);
 
-      if (last_objfile_name == NULL
-	  || strcmp (last_objfile_name, objfile_name (objfile)) != 0)
+      if (last_objfile_name.empty () || last_objfile_name != this_objfile_name)
 	{
-	  xfree (last_objfile_name);
-	  last_objfile_name = xstrdup (objfile_name (objfile));
+	  last_objfile_name = this_objfile_name;
 	  fprintf_filtered (gdb_stdlog,
 			    "Creating one or more psymtabs for objfile %s ...\n",
-			    last_objfile_name);
+			    this_objfile_name);
 	}
       fprintf_filtered (gdb_stdlog,
 			"Created psymtab %s for module %s.\n",
diff --git a/gdb/symfile.c b/gdb/symfile.c
index adcdc169306..ce696321e25 100644
--- a/gdb/symfile.c
+++ b/gdb/symfile.c
@@ -2772,16 +2772,15 @@ allocate_symtab (struct compunit_symtab *cust, const char *filename)
     {
       /* Be a bit clever with debugging messages, and don't print objfile
 	 every time, only when it changes.  */
-      static char *last_objfile_name = NULL;
+      static std::string last_objfile_name;
+      const char *this_objfile_name = objfile_name (objfile);
 
-      if (last_objfile_name == NULL
-	  || strcmp (last_objfile_name, objfile_name (objfile)) != 0)
+      if (last_objfile_name.empty () || last_objfile_name != this_objfile_name)
 	{
-	  xfree (last_objfile_name);
-	  last_objfile_name = xstrdup (objfile_name (objfile));
+	  last_objfile_name = this_objfile_name;
 	  fprintf_filtered (gdb_stdlog,
 			    "Creating one or more symtabs for objfile %s ...\n",
-			    last_objfile_name);
+			    this_objfile_name);
 	}
       fprintf_filtered (gdb_stdlog,
 			"Created symtab %s for module %s.\n",
-- 
2.30.0

Re: [PATCH 2/4] gdb: use std::string in partial_symtab::partial_symtab / allocate_symtab

[Tom Tromey]

>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:

Simon> From: Simon Marchi <simon.marchi@polymtl.ca>
Simon> This simplifies the code a bit.

Simon> gdb/ChangeLog:

Simon> 	* psymtab.c (partial_symtab::partial_symtab): Change
Simon> 	last_objfile_name to be an std::string.
Simon> 	* symfile.c (allocate_symtab): Likewise.

This looks good to me.

Tom

[PATCH 3/4] gdb: pass objfile_per_bfd_storage instead of objfile to partial_symtab

[Simon Marchi]

From: Simon Marchi <simon.marchi@polymtl.ca>

Since partial_symtab is supposed to be objfile-independent (since series
[1]), I think it would make sense for partial_symtab to not take an
objfile as a parameter in its constructor.

This patch replaces that parameter with an objfile_per_bfd_storage
parameter.

The objfile is used for two things:

 - to get the objfile_name, for debug messages.  We can get that name
   from the bfd instead.
 - to intern the partial symtab filename.  Even though it goes through
   an objfile method, the request is actually forwarded to the
   underlying objfile_per_bfd_storage.  So we can ask the new
   objfile_per_bfd_storage instead.

In order to get a reference to the BFD from the objfile_per_bfd_storage,
the BFD is saved in the objfile_per_bfd_storage object.

[1] https://sourceware.org/pipermail/gdb-patches/2021-February/176625.html

gdb/ChangeLog:

	* psympriv.h (struct partial_symtab) <partial_symtab>: Change
	objfile parameter for objfile_per_bfd_storage, adjust callers.
	(struct standard_psymtab) <standard_psymtab>: Likewise.
	(struct legacy_psymtab) <legacy_psymtab>: Likewise.
	* psymtab.c (partial_symtab::partial_symtab): Likewise.
	* ctfread.c (struct ctf_psymtab): Likewise.
	* dwarf2/read.h (struct dwarf2_psymtab): Likewise.
	* dwarf2/read.c (struct dwarf2_include_psymtab): Likewise.
	(dwarf2_create_include_psymtab): Likewise.
	* objfiles.h (struct objfile_per_bfd_storage)
	<objfile_per_bfd_storage>: Add bfd parameter, adjust callers.
	<bfd_>: New method.
	<m_bfd>: New field.
	* objfiles.c (get_objfile_bfd_data):

Change-Id: I2ed3ab5d2e6f27d034bd4dc26ae2fae7b0b8a2b9
---
 gdb/ctfread.c     |  6 +++---
 gdb/dbxread.c     |  4 ++--
 gdb/dwarf2/read.c | 18 ++++++++----------
 gdb/dwarf2/read.h |  4 ++--
 gdb/mdebugread.c  |  6 +++---
 gdb/objfiles.c    |  2 +-
 gdb/objfiles.h    | 16 ++++++++++++++--
 gdb/psympriv.h    | 23 ++++++++++++-----------
 gdb/psymtab.c     | 20 ++++++++++----------
 gdb/xcoffread.c   |  4 ++--
 10 files changed, 57 insertions(+), 46 deletions(-)

diff --git a/gdb/ctfread.c b/gdb/ctfread.c
index 616464c77df..fae244d4481 100644
--- a/gdb/ctfread.c
+++ b/gdb/ctfread.c
@@ -125,9 +125,9 @@ struct ctf_psymtab : public standard_psymtab
 {
   ctf_psymtab (const char *filename,
 	       psymtab_storage *partial_symtabs,
-	       struct objfile *objfile,
+	       objfile_per_bfd_storage *objfile_per_bfd,
 	       CORE_ADDR addr)
-    : standard_psymtab (filename, partial_symtabs, objfile, addr)
+    : standard_psymtab (filename, partial_symtabs, objfile_per_bfd, addr)
   {
   }
 
@@ -1369,7 +1369,7 @@ create_partial_symtab (const char *name,
   ctf_psymtab *pst;
   struct ctf_context *ccx;
 
-  pst = new ctf_psymtab (name, partial_symtabs, objfile, 0);
+  pst = new ctf_psymtab (name, partial_symtabs, objfile->per_bfd, 0);
 
   ccx = XOBNEW (&objfile->objfile_obstack, struct ctf_context);
   ccx->fp = cfp;
diff --git a/gdb/dbxread.c b/gdb/dbxread.c
index 99a36cafa15..5cf77e9c08a 100644
--- a/gdb/dbxread.c
+++ b/gdb/dbxread.c
@@ -1917,7 +1917,7 @@ start_psymtab (psymtab_storage *partial_symtabs, struct objfile *objfile,
 	       const char *filename, CORE_ADDR textlow, int ldsymoff)
 {
   legacy_psymtab *result = new legacy_psymtab (filename, partial_symtabs,
-					       objfile, textlow);
+					       objfile->per_bfd, textlow);
 
   result->read_symtab_private =
     XOBNEW (&objfile->objfile_obstack, struct symloc);
@@ -2040,7 +2040,7 @@ dbx_end_psymtab (struct objfile *objfile, psymtab_storage *partial_symtabs,
   for (i = 0; i < num_includes; i++)
     {
       legacy_psymtab *subpst =
-	new legacy_psymtab (include_list[i], partial_symtabs, objfile);
+	new legacy_psymtab (include_list[i], partial_symtabs, objfile->per_bfd);
 
       subpst->read_symtab_private =
 	XOBNEW (&objfile->objfile_obstack, struct symloc);
diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c
index 2bfb13d6d0e..58b370ecd4a 100644
--- a/gdb/dwarf2/read.c
+++ b/gdb/dwarf2/read.c
@@ -6245,8 +6245,8 @@ struct dwarf2_include_psymtab : public partial_symtab
 {
   dwarf2_include_psymtab (const char *filename,
 			  psymtab_storage *partial_symtabs,
-			  struct objfile *objfile)
-    : partial_symtab (filename, partial_symtabs, objfile)
+			  objfile_per_bfd_storage *objfile_per_bfd)
+    : partial_symtab (filename, partial_symtabs, objfile_per_bfd)
   {
   }
 
@@ -6302,10 +6302,10 @@ dwarf2_create_include_psymtab (dwarf2_per_bfd *per_bfd,
 			       const char *name,
 			       dwarf2_psymtab *pst,
 			       psymtab_storage *partial_symtabs,
-			       struct objfile *objfile)
+			       objfile_per_bfd_storage *objfile_per_bfd)
 {
   dwarf2_include_psymtab *subpst
-    = new dwarf2_include_psymtab (name, partial_symtabs, objfile);
+    = new dwarf2_include_psymtab (name, partial_symtabs, objfile_per_bfd);
 
   if (!IS_ABSOLUTE_PATH (subpst->filename))
     subpst->dirname = pst->dirname;
@@ -7557,11 +7557,9 @@ create_partial_symtab (dwarf2_per_cu_data *per_cu,
 		       dwarf2_per_objfile *per_objfile,
 		       const char *name)
 {
-  struct objfile *objfile = per_objfile->objfile;
-  dwarf2_psymtab *pst;
-
-  pst = new dwarf2_psymtab (name, per_objfile->per_bfd->partial_symtabs.get (),
-			    objfile, per_cu);
+  dwarf2_psymtab *pst
+    = new dwarf2_psymtab (name, per_objfile->per_bfd->partial_symtabs.get (),
+			  per_objfile->objfile->per_bfd, per_cu);
 
   pst->psymtabs_addrmap_supported = true;
 
@@ -22008,7 +22006,7 @@ dwarf_decode_lines (struct line_header *lh, const char *comp_dir,
 	      dwarf2_create_include_psymtab
 		(cu->per_objfile->per_bfd, include_name, pst,
 		 cu->per_objfile->per_bfd->partial_symtabs.get (),
-		 objfile);
+		 objfile->per_bfd);
 	  }
     }
   else
diff --git a/gdb/dwarf2/read.h b/gdb/dwarf2/read.h
index 82ab3879a5b..416d8eae959 100644
--- a/gdb/dwarf2/read.h
+++ b/gdb/dwarf2/read.h
@@ -408,9 +408,9 @@ struct dwarf2_psymtab : public partial_symtab
 {
   dwarf2_psymtab (const char *filename,
 		  psymtab_storage *partial_symtabs,
-		  struct objfile *objfile,
+		  objfile_per_bfd_storage *objfile_per_bfd,
 		  dwarf2_per_cu_data *per_cu)
-    : partial_symtab (filename, partial_symtabs, objfile, 0),
+    : partial_symtab (filename, partial_symtabs, objfile_per_bfd, 0),
       per_cu_data (per_cu)
   {
   }
diff --git a/gdb/mdebugread.c b/gdb/mdebugread.c
index 7bf4564aecb..026f2ff20da 100644
--- a/gdb/mdebugread.c
+++ b/gdb/mdebugread.c
@@ -2605,8 +2605,8 @@ parse_partial_symbols (minimal_symbol_reader &reader,
 	textlow = fh->adr;
       else
 	textlow = 0;
-      pst = new legacy_psymtab (fdr_name (fh), partial_symtabs, objfile,
-				textlow);
+      pst = new legacy_psymtab (fdr_name (fh), partial_symtabs,
+				objfile->per_bfd, textlow);
       pst->read_symtab_private = XOBNEW (&objfile->objfile_obstack, symloc);
       memset (pst->read_symtab_private, 0, sizeof (struct symloc));
 
@@ -4646,7 +4646,7 @@ new_psymtab (const char *name, psymtab_storage *partial_symtabs,
 {
   legacy_psymtab *psymtab;
 
-  psymtab = new legacy_psymtab (name, partial_symtabs, objfile);
+  psymtab = new legacy_psymtab (name, partial_symtabs, objfile->per_bfd);
 
   /* Keep a backpointer to the file's symbols.  */
 
diff --git a/gdb/objfiles.c b/gdb/objfiles.c
index ed51c31c8b9..702900761f3 100644
--- a/gdb/objfiles.c
+++ b/gdb/objfiles.c
@@ -134,7 +134,7 @@ get_objfile_bfd_data (struct objfile *objfile, struct bfd *abfd)
 
   if (storage == NULL)
     {
-      storage = new objfile_per_bfd_storage;
+      storage = new objfile_per_bfd_storage (abfd);
       /* If the object requires gdb to do relocations, we simply fall
 	 back to not sharing data across users.  These cases are rare
 	 enough that this seems reasonable.  */
diff --git a/gdb/objfiles.h b/gdb/objfiles.h
index d9aa06636f5..e8833794c14 100644
--- a/gdb/objfiles.h
+++ b/gdb/objfiles.h
@@ -264,8 +264,8 @@ struct minimal_symbol_iterator
 
 struct objfile_per_bfd_storage
 {
-  objfile_per_bfd_storage ()
-    : minsyms_read (false)
+  objfile_per_bfd_storage (bfd *bfd)
+    : minsyms_read (false), m_bfd (bfd)
   {}
 
   ~objfile_per_bfd_storage ();
@@ -278,6 +278,13 @@ struct objfile_per_bfd_storage
     return (const char *) string_cache.insert (str.data (), str.size () + 1);
   }
 
+  /* Get the BFD this object is associated to.  */
+
+  bfd *bfd_ () const
+  {
+    return m_bfd;
+  }
+
   /* The storage has an obstack of its own.  */
 
   auto_obstack storage_obstack;
@@ -355,6 +362,11 @@ struct objfile_per_bfd_storage
   /* All the different languages of symbols found in the demangled
      hash table.  */
   std::bitset<nr_languages> demangled_hash_languages;
+
+private:
+  /* The BFD this object is associated to.  */
+
+  bfd *m_bfd;
 };
 
 /* An iterator that first returns a parent objfile, and then each
diff --git a/gdb/psympriv.h b/gdb/psympriv.h
index bbae2fc90e4..f6e7feaf3f4 100644
--- a/gdb/psympriv.h
+++ b/gdb/psympriv.h
@@ -111,7 +111,8 @@ enum class psymbol_placement
 
 struct partial_symtab
 {
-  /* Allocate a new partial symbol table associated with OBJFILE.
+  /* Allocate a new partial symbol table.
+
      FILENAME (which must be non-NULL) is the filename of this partial
      symbol table; it is copied into the appropriate storage.  The
      partial symtab will also be installed using
@@ -119,7 +120,7 @@ struct partial_symtab
 
   partial_symtab (const char *filename,
 		  psymtab_storage *partial_symtabs,
-		  struct objfile *objfile)
+		  objfile_per_bfd_storage *objfile_per_bfd)
     ATTRIBUTE_NONNULL (2) ATTRIBUTE_NONNULL (3);
 
   /* Like the above, but also sets the initial text low and text high
@@ -128,7 +129,7 @@ struct partial_symtab
 
   partial_symtab (const char *filename,
 		  psymtab_storage *partial_symtabs,
-		  struct objfile *objfile,
+		  objfile_per_bfd_storage *objfile_per_bfd,
 		  CORE_ADDR addr)
     ATTRIBUTE_NONNULL (2) ATTRIBUTE_NONNULL (3);
 
@@ -369,16 +370,16 @@ struct standard_psymtab : public partial_symtab
 {
   standard_psymtab (const char *filename,
 		    psymtab_storage *partial_symtabs,
-		    struct objfile *objfile)
-    : partial_symtab (filename, partial_symtabs, objfile)
+		    objfile_per_bfd_storage *objfile_per_bfd)
+    : partial_symtab (filename, partial_symtabs, objfile_per_bfd)
   {
   }
 
   standard_psymtab (const char *filename,
 		    psymtab_storage *partial_symtabs,
-		    struct objfile *objfile,
+		    objfile_per_bfd_storage *objfile_per_bfd,
 		    CORE_ADDR addr)
-    : partial_symtab (filename, partial_symtabs, objfile, addr)
+    : partial_symtab (filename, partial_symtabs, objfile_per_bfd, addr)
   {
   }
 
@@ -411,16 +412,16 @@ struct legacy_psymtab : public standard_psymtab
 {
   legacy_psymtab (const char *filename,
 		  psymtab_storage *partial_symtabs,
-		  struct objfile *objfile)
-    : standard_psymtab (filename, partial_symtabs, objfile)
+		  objfile_per_bfd_storage *objfile_per_bfd)
+    : standard_psymtab (filename, partial_symtabs, objfile_per_bfd)
   {
   }
 
   legacy_psymtab (const char *filename,
 		  psymtab_storage *partial_symtabs,
-		  struct objfile *objfile,
+		  objfile_per_bfd_storage *objfile_per_bfd,
 		  CORE_ADDR addr)
-    : standard_psymtab (filename, partial_symtabs, objfile, addr)
+    : standard_psymtab (filename, partial_symtabs, objfile_per_bfd, addr)
   {
   }
 
diff --git a/gdb/psymtab.c b/gdb/psymtab.c
index 2ee05ec415b..93e2a5c8a01 100644
--- a/gdb/psymtab.c
+++ b/gdb/psymtab.c
@@ -1437,9 +1437,9 @@ psymbol_functions::find_compunit_symtab_by_address (struct objfile *objfile,
 
 partial_symtab::partial_symtab (const char *filename,
 				psymtab_storage *partial_symtabs,
-				struct objfile *objfile,
+				objfile_per_bfd_storage *objfile_per_bfd,
 				CORE_ADDR textlow)
-  : partial_symtab (filename, partial_symtabs, objfile)
+  : partial_symtab (filename, partial_symtabs, objfile_per_bfd)
 {
   set_text_low (textlow);
   set_text_high (raw_text_low ()); /* default */
@@ -1562,28 +1562,28 @@ partial_symtab::add_psymbol (gdb::string_view name, bool copy_name,
 
 partial_symtab::partial_symtab (const char *filename_,
 				psymtab_storage *partial_symtabs,
-				struct objfile *objfile)
+				objfile_per_bfd_storage *objfile_per_bfd)
   : searched_flag (PST_NOT_SEARCHED),
     text_low_valid (0),
     text_high_valid (0)
 {
   partial_symtabs->install_psymtab (this);
 
-  filename = objfile->intern (filename_);
+  filename = objfile_per_bfd->intern (filename_);
 
   if (symtab_create_debug)
     {
       /* Be a bit clever with debugging messages, and don't print objfile
 	 every time, only when it changes.  */
-      static std::string last_objfile_name;
-      const char *this_objfile_name = objfile_name (objfile);
+      static std::string last_bfd_name;
+      const char *this_bfd_name = bfd_get_filename (objfile_per_bfd->bfd_ ());
 
-      if (last_objfile_name.empty () || last_objfile_name != this_objfile_name)
+      if (last_bfd_name.empty () || last_bfd_name != this_bfd_name)
 	{
-	  last_objfile_name = this_objfile_name;
+	  last_bfd_name = this_bfd_name;
 	  fprintf_filtered (gdb_stdlog,
-			    "Creating one or more psymtabs for objfile %s ...\n",
-			    this_objfile_name);
+			    "Creating one or more psymtabs for %s ...\n",
+			    this_bfd_name);
 	}
       fprintf_filtered (gdb_stdlog,
 			"Created psymtab %s for module %s.\n",
diff --git a/gdb/xcoffread.c b/gdb/xcoffread.c
index 30ac876e8e3..368572797f6 100644
--- a/gdb/xcoffread.c
+++ b/gdb/xcoffread.c
@@ -1966,7 +1966,7 @@ xcoff_start_psymtab (psymtab_storage *partial_symtabs,
 {
   /* We fill in textlow later.  */
   legacy_psymtab *result = new legacy_psymtab (filename, partial_symtabs,
-					       objfile, 0);
+					       objfile->per_bfd, 0);
 
   result->read_symtab_private =
     XOBNEW (&objfile->objfile_obstack, struct symloc);
@@ -2022,7 +2022,7 @@ xcoff_end_psymtab (struct objfile *objfile, psymtab_storage *partial_symtabs,
   for (i = 0; i < num_includes; i++)
     {
       legacy_psymtab *subpst =
-	new legacy_psymtab (include_list[i], partial_symtabs, objfile);
+	new legacy_psymtab (include_list[i], partial_symtabs, objfile->per_bfd);
 
       subpst->read_symtab_private = XOBNEW (&objfile->objfile_obstack, symloc);
       ((struct symloc *) subpst->read_symtab_private)->first_symnum = 0;
-- 
2.30.0

Re: [PATCH 3/4] gdb: pass objfile_per_bfd_storage instead of objfile to partial_symtab

[Tom Tromey]

>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:

Simon> Since partial_symtab is supposed to be objfile-independent (since series
Simon> [1]), I think it would make sense for partial_symtab to not take an
Simon> objfile as a parameter in its constructor.

Definitely.  Thanks for doing this.

I had one nit:

Simon> +  /* Get the BFD this object is associated to.  */
Simon> +
Simon> +  bfd *bfd_ () const
Simon> +  {
Simon> +    return m_bfd;
Simon> +  }

This seems like an unusual choice of name for an accessor.
Maybe 'get_bfd' instead?

Tom

Re: [PATCH 3/4] gdb: pass objfile_per_bfd_storage instead of objfile to partial_symtab

[Simon Marchi]

On 2021-04-01 1:47 p.m., Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> Simon> Since partial_symtab is supposed to be objfile-independent (since series
> Simon> [1]), I think it would make sense for partial_symtab to not take an
> Simon> objfile as a parameter in its constructor.
> 
> Definitely.  Thanks for doing this.
> 
> I had one nit:
> 
> Simon> +  /* Get the BFD this object is associated to.  */
> Simon> +
> Simon> +  bfd *bfd_ () const
> Simon> +  {
> Simon> +    return m_bfd;
> Simon> +  }
> 
> This seems like an unusual choice of name for an accessor.
> Maybe 'get_bfd' instead?

Hmm yeah.  I generally like to not use `get_` for getters.  For example,
just use `foo.name ()` for a getter that gets the name.  But I wouldn't
let me use `bfd` here.  Anyway, I'll push with `get_bfd`.

Thanks,

Simon

[PATCH 4/4] gdb: remove objfile parameter from get_objfile_bfd_data

[Simon Marchi]

From: Simon Marchi <simon.marchi@polymtl.ca>

I noticed it was unused.  I think that makes sense, as it shows that
objfile_per_bfd_storage is not specific to one objfile (it can be shared
by multiple objfiles that have the same bfd).

There is one thing I wonder though, maybe I'm missing something.  If
the BFD doesn't require relocation, get_objfile_bfd_data stores the
newly allocated object in objfiles_bfd_data, so we can assume that
objfiles_bfd_data is the owner of the object.  When the bfd's refcount
drops to 0, the corresponding objfile_per_bfd_storage object in
objfiles_bfd_data is deleted.

But if the BFD requires relocation, get_objfile_bfd_data returns a newly
allocated object that isn't kept anywhere else (and isn't shared).  So
the objfile becomes the owner of the objfile_per_bfd_storage object.  In
objfile::~objfile, we have this:

    if (obfd)
      gdb_bfd_unref (obfd);
    else
      delete per_bfd;

I'm thinking that obfd could be non-nullptr, and it could require
relocation.  In that case, it would never be freed.  Anyway, that's not
really connected to this patch.

gdb/ChangeLog:

	* objfiles.c (get_objfile_bfd_data): Remove objfile parameter,
	adjust callers.

Change-Id: Ifa3158074ea6b42686780ba09d0c964b0cf14cf1
---
 gdb/objfiles.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/gdb/objfiles.c b/gdb/objfiles.c
index 702900761f3..f68e97bd08f 100644
--- a/gdb/objfiles.c
+++ b/gdb/objfiles.c
@@ -125,7 +125,7 @@ objfile_per_bfd_storage::~objfile_per_bfd_storage ()
    only be called when allocating or re-initializing OBJFILE.  */
 
 static struct objfile_per_bfd_storage *
-get_objfile_bfd_data (struct objfile *objfile, struct bfd *abfd)
+get_objfile_bfd_data (bfd *abfd)
 {
   struct objfile_per_bfd_storage *storage = NULL;
 
@@ -154,7 +154,7 @@ get_objfile_bfd_data (struct objfile *objfile, struct bfd *abfd)
 void
 set_objfile_per_bfd (struct objfile *objfile)
 {
-  objfile->per_bfd = get_objfile_bfd_data (objfile, objfile->obfd);
+  objfile->per_bfd = get_objfile_bfd_data (objfile->obfd);
 }
 
 /* Set the objfile's per-BFD notion of the "main" name and
@@ -363,7 +363,7 @@ objfile::objfile (bfd *abfd, const char *name, objfile_flags flags_)
       build_objfile_section_table (this);
     }
 
-  per_bfd = get_objfile_bfd_data (this, abfd);
+  per_bfd = get_objfile_bfd_data (abfd);
 }
 
 /* If there is a valid and known entry point, function fills *ENTRY_P with it
-- 
2.30.0

Re: [PATCH 4/4] gdb: remove objfile parameter from get_objfile_bfd_data

[Tom Tromey]

>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:

Simon> But if the BFD requires relocation, get_objfile_bfd_data returns a newly
Simon> allocated object that isn't kept anywhere else (and isn't shared).  So
Simon> the objfile becomes the owner of the objfile_per_bfd_storage object.  In
Simon> objfile::~objfile, we have this:

Simon>     if (obfd)
Simon>       gdb_bfd_unref (obfd);
Simon>     else
Simon>       delete per_bfd;

Simon> I'm thinking that obfd could be non-nullptr, and it could require
Simon> relocation.  In that case, it would never be freed.  Anyway, that's not
Simon> really connected to this patch.

Yes, this seems like a bug.  Probably objfile should just have an
explicit "owns_per_bfd" boolean somewhere.

Simon> gdb/ChangeLog:

Simon> 	* objfiles.c (get_objfile_bfd_data): Remove objfile parameter,
Simon> 	adjust callers.

This looks good to me.  Thank you.

Tom

Re: [PATCH 4/4] gdb: remove objfile parameter from get_objfile_bfd_data

[Simon Marchi]

On 2021-04-01 1:52 p.m., Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> Simon> But if the BFD requires relocation, get_objfile_bfd_data returns a newly
> Simon> allocated object that isn't kept anywhere else (and isn't shared).  So
> Simon> the objfile becomes the owner of the objfile_per_bfd_storage object.  In
> Simon> objfile::~objfile, we have this:
> 
> Simon>     if (obfd)
> Simon>       gdb_bfd_unref (obfd);
> Simon>     else
> Simon>       delete per_bfd;
> 
> Simon> I'm thinking that obfd could be non-nullptr, and it could require
> Simon> relocation.  In that case, it would never be freed.  Anyway, that's not
> Simon> really connected to this patch.
> 
> Yes, this seems like a bug.  Probably objfile should just have an
> explicit "owns_per_bfd" boolean somewhere.
> 
> Simon> gdb/ChangeLog:
> 
> Simon> 	* objfiles.c (get_objfile_bfd_data): Remove objfile parameter,
> Simon> 	adjust callers.
> 
> This looks good to me.  Thank you.

Thanks, all pushed now.  I removed the last sentence of the
get_objfile_bfd_data comment, it didn't seem appropriate anymore.

Simon