[3/3] btrace: Remove ui_out cleanups

  This patch replaces the cleanups that close the list and tuple of the
btrace instruction history output with ui_out_emit_tuple and
ui_out_emit_list.

This allows removing make_cleanup_ui_out_tuple_begin_end and
make_cleanup_ui_out_list_begin_end.

This patch (along with the previous ones in the series) was regtested on
the buildbot.

gdb/ChangeLog:

	* record-btrace.c (btrace_print_lines): Replace cleanup
	parameter with RAII equivalents.
	(btrace_insn_history): Replace cleanup with RAII equivalents.
	* ui-out.h (make_cleanup_ui_out_list_begin_end,
	make_cleanup_ui_out_tuple_begin_end): Remove.
	* ui-out.c (struct ui_out_end_cleanup_data, do_cleanup_end,
	make_cleanup_ui_out_end, make_cleanup_ui_out_tuple_begin_end,
	make_cleanup_ui_out_list_begin_end): Remove.
---
 gdb/record-btrace.c | 55 ++++++++++++++++++++++++-----------------------------
 gdb/ui-out.c        | 44 ------------------------------------------
 gdb/ui-out.h        |  8 --------
 3 files changed, 25 insertions(+), 82 deletions(-)
  

Hello Simon,


> This patch replaces the cleanups that close the list and tuple of the
> btrace instruction history output with ui_out_emit_tuple and
> ui_out_emit_list.
> 
> This allows removing make_cleanup_ui_out_tuple_begin_end and
> make_cleanup_ui_out_list_begin_end.
> 
> This patch (along with the previous ones in the series) was regtested on
> the buildbot.


> diff --git a/gdb/record-btrace.c b/gdb/record-btrace.c
> index 15ce760f5a..ddd15c4781 100644
> --- a/gdb/record-btrace.c
> +++ b/gdb/record-btrace.c
> @@ -620,26 +620,25 @@ btrace_find_line_range (CORE_ADDR pc)
> 
>  static void
>  btrace_print_lines (struct btrace_line_range lines, struct ui_out *uiout,
> -		    struct cleanup **ui_item_chain, int flags)
> +		    gdb::optional<ui_out_emit_tuple> *src_and_asm_tuple,
> +		    gdb::optional<ui_out_emit_list> *asm_list,

Reference instead of pointer?

 
> -  for (line = lines.begin; line < lines.end; ++line)
> +  for (int line = lines.begin; line < lines.end; ++line)
>      {
> -      if (*ui_item_chain != NULL)
> -	do_cleanups (*ui_item_chain);
> +      asm_list->reset ();
> +      src_and_asm_tuple->reset ();

The SRC_AND_ASM_TUPLE reset () shouldn't be necessary; it is reset () in emplace ().


> -      *ui_item_chain
> -	= make_cleanup_ui_out_tuple_begin_end (uiout, "src_and_asm_line");
> +      src_and_asm_tuple->emplace (uiout, "src_and_asm_line");
> 
>        print_source_lines (lines.symtab, line, line + 1, psl_flags);
> 
> -      make_cleanup_ui_out_list_begin_end (uiout, "line_asm_insn");
> +      asm_list->emplace (uiout, "line_asm_insn");
>      }
>  }

Looks good to me, otherwise.

I have not tested it, though.  Did you run the gdb.btrace test suite?  You said
you ran buildbot but that might skip gdb.btrace when run on VMs or on old
hardware.

Please let me know if you want me to run the gdb.btrace tests for you.  A user
branch would be nice in that case.

Regards,
Markus.

Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928
  

On 2018-03-05 07:39, Metzger, Markus T wrote:
> Hello Simon,
> 
> 
>> This patch replaces the cleanups that close the list and tuple of the
>> btrace instruction history output with ui_out_emit_tuple and
>> ui_out_emit_list.
>> 
>> This allows removing make_cleanup_ui_out_tuple_begin_end and
>> make_cleanup_ui_out_list_begin_end.
>> 
>> This patch (along with the previous ones in the series) was regtested 
>> on
>> the buildbot.
> 
> 
>> diff --git a/gdb/record-btrace.c b/gdb/record-btrace.c
>> index 15ce760f5a..ddd15c4781 100644
>> --- a/gdb/record-btrace.c
>> +++ b/gdb/record-btrace.c
>> @@ -620,26 +620,25 @@ btrace_find_line_range (CORE_ADDR pc)
>> 
>>  static void
>>  btrace_print_lines (struct btrace_line_range lines, struct ui_out 
>> *uiout,
>> -		    struct cleanup **ui_item_chain, int flags)
>> +		    gdb::optional<ui_out_emit_tuple> *src_and_asm_tuple,
>> +		    gdb::optional<ui_out_emit_list> *asm_list,
> 
> Reference instead of pointer?

I once pointed this out on one of Tom's patches, and he said that in the 
caller code, it's more obvious that object is meant to be modified if 
you do:

   function_that_modifies_object (&object);

instead of

   function_that_modifies_object (object);

And I kind of agree with that it is, which is why I've been using 
pointers when references would have worked (ok, here the function name 
makes it obvious but it's not always the case).  In the implementation 
of the function, since you use object->field instead of object.field, it 
also hints that you're not modifying a local object.  But I also agree 
that it's really not C++-y to do it this way, so I'll happily change it.

>> -  for (line = lines.begin; line < lines.end; ++line)
>> +  for (int line = lines.begin; line < lines.end; ++line)
>>      {
>> -      if (*ui_item_chain != NULL)
>> -	do_cleanups (*ui_item_chain);
>> +      asm_list->reset ();
>> +      src_and_asm_tuple->reset ();
> 
> The SRC_AND_ASM_TUPLE reset () shouldn't be necessary; it is reset ()
> in emplace ().

Ok.

>> -      *ui_item_chain
>> -	= make_cleanup_ui_out_tuple_begin_end (uiout, "src_and_asm_line");
>> +      src_and_asm_tuple->emplace (uiout, "src_and_asm_line");
>> 
>>        print_source_lines (lines.symtab, line, line + 1, psl_flags);
>> 
>> -      make_cleanup_ui_out_list_begin_end (uiout, "line_asm_insn");
>> +      asm_list->emplace (uiout, "line_asm_insn");
>>      }
>>  }
> 
> Looks good to me, otherwise.
> 
> I have not tested it, though.  Did you run the gdb.btrace test suite?  
> You said
> you ran buildbot but that might skip gdb.btrace when run on VMs or on 
> old
> hardware.
> 
> Please let me know if you want me to run the gdb.btrace tests for you.  
> A user
> branch would be nice in that case.

Yes, I have ran the gdb.btrace/*.exp tests locally on two different 
machines and saw no regressions.  However, the processors may be a bit 
old (Q6600 from 2007 and i5-4310U from 2014), so it's possible that not 
all required features are available, and therefore some tests may be 
skipped.  So if you want to be sure, here's a branch for you to test:

users/simark/btrace-cleanups
https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=shortlog;h=refs/heads/users/simark/btrace-cleanups

Thanks!

Simon
  

Hello Simon,

> >>  static void
> >>  btrace_print_lines (struct btrace_line_range lines, struct ui_out
> >> *uiout,
> >> -		    struct cleanup **ui_item_chain, int flags)
> >> +		    gdb::optional<ui_out_emit_tuple> *src_and_asm_tuple,
> >> +		    gdb::optional<ui_out_emit_list> *asm_list,
> >
> > Reference instead of pointer?
> 
> I once pointed this out on one of Tom's patches, and he said that in the
> caller code, it's more obvious that object is meant to be modified if
> you do:
> 
>    function_that_modifies_object (&object);
> 
> instead of
> 
>    function_that_modifies_object (object);
> 
> And I kind of agree with that it is, which is why I've been using
> pointers when references would have worked (ok, here the function name
> makes it obvious but it's not always the case).  In the implementation
> of the function, since you use object->field instead of object.field, it
> also hints that you're not modifying a local object.  But I also agree
> that it's really not C++-y to do it this way, so I'll happily change it.

I prefer consistency.  I we agreed to use pointers instead of references
in other parts of GDB, let's do so everywhere.


> Yes, I have ran the gdb.btrace/*.exp tests locally on two different
> machines and saw no regressions.  However, the processors may be a bit
> old (Q6600 from 2007 and i5-4310U from 2014), so it's possible that not
> all required features are available, and therefore some tests may be
> skipped.  So if you want to be sure, here's a branch for you to test:

You would get an "untested" if btrace tests are skipped.  As long as
you're not getting all "untested", you should be fine.  There is only
one test, tsx.exp, that requires recent hardware and compiler.

It would use the method that is available on your target preferring
PT over BTS.  But this change is not related to trace decode so it
shouldn't matter.

I ran the tests on recent hardware using PT and everything passes.

Thanks,
Markus.

Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928
  

On 2018-03-06 02:30, Metzger, Markus T wrote:
> I prefer consistency.  I we agreed to use pointers instead of 
> references
> in other parts of GDB, let's do so everywhere.

There was no formal decision, I would just say it's the current trend.  
But it would be a good idea to formalize it, so we don't have to wonder 
about it again, I'll send a proposal in a separate mail.  For reference, 
I checked the Google C++ style guide, and they forbid non-const 
reference:

https://google.github.io/styleguide/cppguide.html#Reference_Arguments

I'll push this patch with pointers then.

>> Yes, I have ran the gdb.btrace/*.exp tests locally on two different
>> machines and saw no regressions.  However, the processors may be a bit
>> old (Q6600 from 2007 and i5-4310U from 2014), so it's possible that 
>> not
>> all required features are available, and therefore some tests may be
>> skipped.  So if you want to be sure, here's a branch for you to test:
> 
> You would get an "untested" if btrace tests are skipped.  As long as
> you're not getting all "untested", you should be fine.  There is only
> one test, tsx.exp, that requires recent hardware and compiler.
> 
> It would use the method that is available on your target preferring
> PT over BTS.  But this change is not related to trace decode so it
> shouldn't matter.
> 
> I ran the tests on recent hardware using PT and everything passes.

Ok, thanks!

Simon