Re: [PATCH 2/2] gdb: Preserve is-stmt lines when switch between files

[Andrew Burgess <andrew.burgess@embecosm.com>]

* Andrew Burgess <andrew.burgess@embecosm.com> [2020-05-27 14:10:22 +0100]:

> Below is an updated version of the previous work in progress patch.
> I'm posting it here so there's a record, however, I don't think this
> patch should be merged into upstream.  While working on this solution
> the problems with it became apparent to me.  I've tried to describe my
> issues with this approach in the commit message of the patch.
> 
> It is now my intention to merge the patch I posted here:
> 
>    https://sourceware.org/pipermail/gdb-patches/2020-May/168611.html
> 
> which I'll do in a couple of days unless anyone would like to discuss
> this further.

I have now merged the patch mentioned above.

Tom - this should resolve the regression you are seeing in your MinGW
test.  Do let me know if you continue to see any problems related to
this series.

Thanks,
Andrew



> 
> Thanks,
> Andrew
> 
> ---
> 
> commit 79719063c19b3906fda9d9e7fa0b8c009716b53a
> Author: Andrew Burgess <andrew.burgess@embecosm.com>
> Date:   Wed May 20 19:16:05 2020 +0100
> 
>     gdb: Don't delete empty lines when changing symtabs
>     
>     ***** NOT FOR MERGING INTO UPSTREAM *****
>     
>     This patch is based on the work of Bernd Edlinger here:
>     
>       https://sourceware.org/pipermail/gdb-patches/2020-May/168591.html
>     
>     After the is-stmt support commit:
>     
>         commit 8c95582da858ac981f689a6f599acacb8c5c490f
>         Date:   Mon Dec 30 21:04:51 2019 +0000
>     
>             gdb: Add support for tracking the DWARF line table is-stmt field
>     
>     A regression was observed where a breakpoint could no longer be placed
>     in some cases.
>     
>     Consider a line table like this:
>     
>       File 1: test.c
>       File 2: test.h
>     
>       | Addr | File | Line | Stmt |
>       |------|------|------|------|
>       | 1    | 1    | 16   | Y    |
>       | 2    | 1    | 17   | Y    |
>       | 3    | 2    | 21   | Y    |
>       | 4    | 2    | 22   | Y    |
>       | 4    | 1    | 18   | N    |
>       | 5    | 2    | 23   | N    |
>       | 6    | 1    | 24   | Y    |
>       | 7    | 1    | END  | Y    |
>       |------|------|------|------|
>     
>     Before the is-stmt patch GDB would ignore any non-stmt lines, so GDB
>     built two line table structures:
>     
>       File 1                 File 2
>       ------                 ------
>     
>       | Addr | Line |        | Addr | Line |
>       |------|------|        |------|------|
>       | 1    | 16   |        | 3    | 21   |
>       | 2    | 17   |        | 4    | 22   |
>       | 3    | END  |        | 6    | END  |
>       | 6    | 24   |        |------|------|
>       | 7    | END  |
>       |------|------|
>     
>     After the is-stmt patch GDB now records non-stmt lines, so the
>     generated line table structures look like this:
>     
>       File 1                   File 2
>       ------                   ------
>     
>       | Addr | Line | Stmt |  | Addr | Line | Stmt |
>       |------|------|------|  |------|------|------|
>       | 1    | 16   | Y    |  | 3    | 21   | Y    |
>       | 2    | 17   | Y    |  | 4    | 22   | Y    |
>       | 3    | END  | Y    |  | 4    | END  | Y    |
>       | 4    | 18   | N    |  | 5    | 23   | N    |
>       | 5    | END  | Y    |  | 6    | END  | Y    |
>       | 6    | 24   | Y    |  |------|------|------|
>       | 7    | END  | Y    |
>       |------|------|------|
>     
>     The problem is that in 'File 2', end END marker at address 4 causes
>     the previous line table entry to be discarded, so we actually end up
>     with this:
>     
>       File 2
>       ------
>     
>       | Addr | Line | Stmt |
>       |------|------|------|
>       | 3    | 21   | Y    |
>       | 4    | END  | Y    |
>       | 5    | 23   | N    |
>       | 6    | END  | Y    |
>       |------|------|------|
>     
>     When a user tries to place a breakpoint in file 2 at line 22, this is
>     no longer possible.
>     
>     In order to understand the solution presented here, we must first
>     remind ourselves why we delete line table entries that occur at the
>     same address as an END marker.
>     
>     The problem this (deleting lines) is trying to solve is that if the
>     last real line within a line table is empty (maybe due to
>     optimisation) then the line will be at the same address as the end
>     marker.  If the user tries to place a breakpoint at this address then
>     we actually end up with a breakpoint at whatever is _after_ the
>     location of that empty line, which is likely some other random
>     function.
>     
>     However, when a line table switches subfiles we also insert end
>     markers, this is because the GDB internal line table is split
>     per-subfile, so the end markers allow us to see and "end" for the line
>     just before the switch.  Without this, the line table entry just
>     before the we switch subfiles would appear to extend over the entire
>     region covered by the subfile we switch too.
>     
>     If we further consider the end markers inserted for switching
>     subfiles, there are two common cases when this can occur, these are,
>     first when one file is fully included inside another, and secondly
>     when a function from one (header) file is inlined within a function
>     in a different file.
>     
>     In the first of these situations GDB is exposed to the same risk as
>     for the classic end marker case, if the last line before the subfile
>     switch is empty, then when we switch subfiles we should be deleting
>     the empty line, otherwise a user placing a breakpoint on this line,
>     will end up with a breakpoint in a potentially different function.
>     
>     However, for the case of the inlined function the argument for
>     deleting lines is harder to make, indeed if we consider a function
>     inlined from another file compared to a function inlined from within
>     the same file then it is obvious that in the case where the inlined
>     callee is within the same file there will be no subfile switch, and
>     hence no end of sequence markers inserted, and as a consequence no
>     deleted line entries.
>     
>     It is tempting then to think that, when we switch subfiles as part of
>     an inlined function we would like to preserve line table entries where
>     possible.
>     
>     The problem is how to distinguish between the case of an inlined
>     function subfile switch, and a general inclusion subfile switch?
>     Though I'm sure we can solve this problem, I think it is worth
>     considering if this is even a problem worth solving.
>     
>     First, the act of deleting line table entries only occurs when entries
>     in the original debug information "pile up" and occur at the same
>     address, consider this imagined input line table:
>     
>       | Addr | File | Line | Stmt |
>       |------|------|------|------|
>       | 1    | 1    | 10   | Y    |
>       | 2    | 2    | 100  | Y    |
>       | 3    | 1    | 11   | Y    |
>     
>     In this case no entries will be deleted, as every line table item
>     occurs at a separate address.  Compare it to this:
>     
>       | Addr | File | Line | Stmt |
>       |------|------|------|------|
>       | 1    | 1    | 10   | Y    |
>       | 1    | 2    | 100  | Y    |
>       | 2    | 1    | 11   | Y    |
>     
>     Here, the switch to subfile 2 will cause the line item from subfile
>     1 (also at address 1) to be deleted.  This kind of makes sense, the
>     debug is claiming the address 1 represents both file 1, line 10 and
>     file 2, line 100.  If we stop at this address which line should GDB
>     report?  If the same situation was imagined within a single subfile,
>     like this:
>     
>       | Addr | File | Line | Stmt |
>       |------|------|------|------|
>       | 1    | 1    | 10   | Y    |
>       | 1    | 1    | 100  | Y    |
>       | 2    | 1    | 11   | Y    |
>     
>     GDB would report line 100 at address 1, and line 11 at address 2.  By
>     deleting the line in the two subfile case we get the same behaviour,
>     but we protect against the empty line at the end of a subfile
>     problem.
>     
>     It is for this reason, that I think this commit should not be merged
>     with GDB.
>     
>     One problem that came from this commit was the test
>     gdb.cp/step-and-next-inline.exp, which broke in several places.  After
>     looking at this test again I think that in some cases this test was
>     only ever passing by pure luck.  The debug GCC is producing for this
>     test is pretty broken.  I raised this GCC bug:
>     
>       https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94474
>     
>     for this and disabled one entire half of the test.  There are still
>     some cases in here that do pass, and if/when GCC is fixed it would be
>     great to enable this test again.  Also the possibility of extending
>     GDB to handle GCC's broken debug information has been discussed.
>     
>     gdb/ChangeLog:
>     
>             * buildsym.c (buildsym_compunit::record_line): Extend header
>             comment.  Add some static assertions.  Handle different end of
>             sequence types.
>             * buildsym.h (struct buildsym_compunit) <end_of_sequence>: New
>             constant.
>             <change_of_subfile>: New constant.
>             * dwarf2/read.c (dwarf_finish_line): Add extra parameter, use this
>             to decide which end of sequence marker to send through as the line
>             number.
>             (lnp_state_machine::record_line): Pass extra parameter to
>             dwarf_finish_line.
>             * symtab.c (find_pc_sect_line): The previous line is better if it
>             was marked is-stmt, and the new possible best is not.
>             * symtab.h (struct linetable_entry) <line>: Extend comment.
>     
>     gdb/testsuite/ChangeLog:
>     
>             * gdb.cp/step-and-next-inline.exp (do_test): Skip all tests in the
>             use_header case.
>             * gdb.dwarf2/dw2-inline-header-1.exp: New file.
>             * gdb.dwarf2/dw2-inline-header-2.exp: New file.
>             * gdb.dwarf2/dw2-inline-header-3.exp: New file.
>             * gdb.dwarf2/dw2-inline-header-lbls.c: New file.
>             * gdb.dwarf2/dw2-inline-header.c: New file.
>             * gdb.dwarf2/dw2-inline-header.h: New file.
> 
> diff --git a/gdb/buildsym.c b/gdb/buildsym.c
> index 33bf6523e90..37fec4f53b6 100644
> --- a/gdb/buildsym.c
> +++ b/gdb/buildsym.c
> @@ -663,7 +663,42 @@ buildsym_compunit::pop_subfile ()
>  }
>  \f
>  /* Add a linetable entry for line number LINE and address PC to the
> -   line vector for SUBFILE.  */
> +   line vector for SUBFILE.
> +
> +   Pass the constant end_of_sequence for LINE when an end of sequence
> +   marker is encountered in the original debug information.  If this case
> +   any previous line table entries already at this address will be deleted,
> +   these lines are by definition empty, and attempting to place a
> +   breakpoint at this address will result in a breakpoint being placed on
> +   whatever is next in the executable, which is never going to be the
> +   correct behaviour.
> +
> +   Alternatively, pass the constant change_of_subfile for LINE to indicate
> +   that the original debug line table switched to a different subfile.
> +   This still results in an end of sequence marker being placed in GDB's
> +   internal line table, but in this case we don't delete previous line
> +   entries at this address.  This case arises from two common situations,
> +   the most common is inlined functions, in this case the switching between
> +   subfiles occurs within a single function.  If a real line entry and a
> +   subfile switch occur at the same address we don't want to delete the
> +   real line entry in this case as we end up loosing too much useful debug
> +   information.
> +
> +   The other situation where a subfile switch is commonly seen is when one
> +   source file is included into another, in this case the original line
> +   table will include information for one subfile before switching to the
> +   second subfile.  If the last line of the last function in the first
> +   subfile is at the same address as the subfile switch then it is possible
> +   that a breakpoint placed on this line would actually be placed in the
> +   subsequent function by mistake.
> +
> +   We could improve on this second situation by entering all end of
> +   sequence markers here without deleting any line table entries, then
> +   cross check the line table with the list of function blocks, we would
> +   then delete any line table entries that occurred at an end of sequence
> +   marker that was also at the end of a function.  Right now we don't do
> +   this, and live with the (small) risk of a breakpoint being misplaced;
> +   the benefit of preserving the extra line table entries is worth it.  */
>  
>  void
>  buildsym_compunit::record_line (struct subfile *subfile, int line,
> @@ -671,6 +706,13 @@ buildsym_compunit::record_line (struct subfile *subfile, int line,
>  {
>    struct linetable_entry *e;
>  
> +  /* The END_OF_SEQUENCE must be 0 as the constant 0 is used when
> +     processing the line table as the magic end of sequence marker.  The
> +     CHANGE_OF_SUBFILE marker must be less than 0 as any value greater than
> +     0 is an actual line number.  */
> +  gdb_static_assert (buildsym_compunit::end_of_sequence == 0);
> +  gdb_static_assert (buildsym_compunit::change_of_subfile < 0);
> +
>    /* Make sure line vector exists and is big enough.  */
>    if (!subfile->line_vector)
>      {
> @@ -705,7 +747,7 @@ buildsym_compunit::record_line (struct subfile *subfile, int line,
>       end of sequence markers.  All we lose is the ability to set
>       breakpoints at some lines which contain no instructions
>       anyway.  */
> -  if (line == 0)
> +  if (line == buildsym_compunit::end_of_sequence)
>      {
>        while (subfile->line_vector->nitems > 0)
>  	{
> @@ -715,8 +757,11 @@ buildsym_compunit::record_line (struct subfile *subfile, int line,
>  	  subfile->line_vector->nitems--;
>  	}
>      }
> +  else if (line == buildsym_compunit::change_of_subfile)
> +    line = buildsym_compunit::end_of_sequence;
>  
>    e = subfile->line_vector->item + subfile->line_vector->nitems++;
> +  gdb_assert (line >= 0);
>    e->line = line;
>    e->is_stmt = is_stmt ? 1 : 0;
>    e->pc = pc;
> diff --git a/gdb/buildsym.h b/gdb/buildsym.h
> index c768a4c2dae..3e22d0b4494 100644
> --- a/gdb/buildsym.h
> +++ b/gdb/buildsym.h
> @@ -190,6 +190,14 @@ struct buildsym_compunit
>    void record_line (struct subfile *subfile, int line, CORE_ADDR pc,
>  		    bool is_stmt);
>  
> +  /* These two constants can be passed to RECORD_LINE as the value of LINE
> +     to indicate special events in the line table.  END_OF_SEQUENCE marks
> +     an actual end of sequence marked in the original debug's line table,
> +     while CHANGE_OF_SUBFILE indicates an end marker due to switching
> +     between subfiles.  */
> +  static const int end_of_sequence = 0;
> +  static const int change_of_subfile = -1;
> +
>    struct compunit_symtab *get_compunit_symtab ()
>    {
>      return m_compunit_symtab;
> diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c
> index ec3844188ee..ed0f247b6a1 100644
> --- a/gdb/dwarf2/read.c
> +++ b/gdb/dwarf2/read.c
> @@ -20067,7 +20067,7 @@ dwarf_record_line_1 (struct gdbarch *gdbarch, struct subfile *subfile,
>  
>  static void
>  dwarf_finish_line (struct gdbarch *gdbarch, struct subfile *subfile,
> -		   CORE_ADDR address, struct dwarf2_cu *cu)
> +		   CORE_ADDR address, struct dwarf2_cu *cu, bool end_sequence)
>  {
>    if (subfile == NULL)
>      return;
> @@ -20080,7 +20080,10 @@ dwarf_finish_line (struct gdbarch *gdbarch, struct subfile *subfile,
>  			  paddress (gdbarch, address));
>      }
>  
> -  dwarf_record_line_1 (gdbarch, subfile, 0, address, true, cu);
> +  int lineno = (end_sequence
> +		? buildsym_compunit::end_of_sequence
> +		: buildsym_compunit::change_of_subfile);
> +  dwarf_record_line_1 (gdbarch, subfile, lineno, address, true, cu);
>  }
>  
>  void
> @@ -20113,7 +20116,8 @@ lnp_state_machine::record_line (bool end_sequence)
>  	      || end_sequence)
>  	    {
>  	      dwarf_finish_line (m_gdbarch, m_last_subfile, m_address,
> -				 m_currently_recording_lines ? m_cu : nullptr);
> +				 m_currently_recording_lines ? m_cu : nullptr,
> +				 end_sequence);
>  	    }
>  
>  	  if (!end_sequence)
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 5c4e282c024..f6ebae45b27 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -3254,7 +3254,9 @@ find_pc_sect_line (CORE_ADDR pc, struct obj_section *section, int notcurrent)
>           save prev if it represents the end of a function (i.e. line number
>           0) instead of a real line.  */
>  
> -      if (prev && prev->line && (!best || prev->pc > best->pc))
> +      if (prev && prev->line && (!best || prev->pc > best->pc
> +				 || (prev->pc == best->pc && !best->is_stmt
> +				     && prev->is_stmt)))
>  	{
>  	  best = prev;
>  	  best_symtab = iter_s;
> diff --git a/gdb/symtab.h b/gdb/symtab.h
> index 9972e8125ba..62bedb45de7 100644
> --- a/gdb/symtab.h
> +++ b/gdb/symtab.h
> @@ -1306,7 +1306,10 @@ struct rust_vtable_symbol : public symbol
>  
>  struct linetable_entry
>  {
> -  /* The line number for this entry.  */
> +  /* The line number for this entry.  This should be a value greater than
> +     0 to indicate an actual line number, or the magic value 0 is used to
> +     indicate the end of a sequence of linetable_entry structures in a
> +     list.  */
>    int line;
>  
>    /* True if this PC is a good location to place a breakpoint for LINE.  */
> diff --git a/gdb/testsuite/gdb.cp/step-and-next-inline.exp b/gdb/testsuite/gdb.cp/step-and-next-inline.exp
> index 3733fa75570..a95e21194f9 100644
> --- a/gdb/testsuite/gdb.cp/step-and-next-inline.exp
> +++ b/gdb/testsuite/gdb.cp/step-and-next-inline.exp
> @@ -24,6 +24,13 @@ if { ![supports_statement_frontiers] } {
>  proc do_test { use_header } {
>      global srcfile testfile
>  
> +    if { $use_header } {
> +	# This test will not pass due to poor debug information
> +	# generated by GCC (at least upto 10.x).  See
> +	# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94474
> +	return
> +    }
> +
>      set options {c++ debug nowarnings optimize=-O2\ -gstatement-frontiers}
>      if { $use_header } {
>  	lappend options additional_flags=-DUSE_NEXT_INLINE_H
> diff --git a/gdb/testsuite/gdb.dwarf2/dw2-inline-header-1.exp b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-1.exp
> new file mode 100644
> index 00000000000..dc7ec929236
> --- /dev/null
> +++ b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-1.exp
> @@ -0,0 +1,186 @@
> +# Copyright 2020 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/>.
> +
> +# Setup a line table where:
> +#
> +# |      |      |      |      | Inline | Inline |
> +# | Addr | File | Line | Stmt | Rng A  | Rng B  |
> +# |------|------|------|------|--------|--------|
> +# | 1    | 1    | 16   | Y    |        |        |
> +# | 2    | 1    | 17   | Y    |        |        |
> +# | 3    | 2    | 21   | Y    | X      |        |
> +# | 4    | 2    | 22   | Y    | X      |        |
> +# | 4    | 1    | 18   | N    | X      |        |
> +# | 5    | 2    | 23   | N    | X      | X      |
> +# | 6    | 1    | 24   | Y    |        |        |
> +# | 7    | 1    | END  | Y    |        |        |
> +# |------|------|------|------|--------|--------|
> +#
> +# Places a brekpoint at file 2, line 22.  Previously GDB would discard
> +# the line table entry for this line due to switching files for the
> +# file 1, line 18 non-statement line.  After patching however, GDB now
> +# discards the file 1, line 18 entry instead, and the breakpoint at
> +# line 22 should succeed.
> +#
> +# The two inlined subroutine ranges 'A' and 'B' represent two possible
> +# ways that a compiler might represent this siuatio in the DWARF.
> +#
> +# Range 'B' is something that has been seen in the wild using GCC 8.2.
> +# In this case the compilers range information is clearly wrong, but
> +# this shouldn't impact the main point of the test.
> +#
> +# Range 'A' is a hypothetical case of how the compiler might choose to
> +# represent this range, this has never been seen in the wild, but is
> +# an improved debug experiece over range 'B'.  However, if we ever run
> +# in to the situation where GDB can support the range 'A' test, or
> +# support some real DWARF seen in the wild, then the range 'A' case
> +# should be dropped in favour of supporting real world cases.  This is
> +# included here as it "just worked" once the range 'B' case was
> +# working.
> +
> +load_lib dwarf.exp
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +if {![dwarf2_support]} {
> +    return 0
> +}
> +
> +# The .c files use __attribute__.
> +if [get_compiler_info] {
> +    return -1
> +}
> +if !$gcc_compiled {
> +    return 0
> +}
> +
> +# Prepare and run the test.
> +proc do_test { start_label func_name tag } {
> +    global srcfile srcfile2 srcfile3 srcfile4 testfile
> +
> +    standard_testfile dw2-inline-header-lbls.c dw2-inline-header-${tag}.S \
> +	dw2-inline-header.c dw2-inline-header.h
> +
> +    set build_options {nodebug optimize=-O1}
> +
> +    set asm_file [standard_output_file $srcfile2]
> +    Dwarf::assemble $asm_file {
> +	global srcdir subdir srcfile srcfile3 srcfile4 testfile
> +	upvar build_options build_options
> +	upvar start_label start_label
> +	declare_labels lines_label callee_subprog_label
> +
> +	get_func_info main $build_options
> +
> +	cu {} {
> +	    compile_unit {
> +		{producer "gcc" }
> +		{language @DW_LANG_C}
> +		{name ${srcfile3}}
> +		{low_pc 0 addr}
> +		{stmt_list ${lines_label} DW_FORM_sec_offset}
> +	    } {
> +		callee_subprog_label: subprogram {
> +		    {external 1 flag}
> +		    {name callee}
> +		    {inline 3 data1}
> +		}
> +		subprogram {
> +		    {external 1 flag}
> +		    {name main}
> +		    {low_pc $main_start addr}
> +		    {high_pc "$main_start + $main_len" addr}
> +		} {
> +		    inlined_subroutine {
> +			{abstract_origin %$callee_subprog_label}
> +			{low_pc $start_label addr}
> +			{high_pc line_label_6 addr}
> +			{call_file 1 data1}
> +			{call_line 18 data1}
> +		    }
> +		}
> +	    }
> +	}
> +
> +	lines {version 2 default_is_stmt 1} lines_label {
> +	    include_dir "${srcdir}/${subdir}"
> +	    file_name "$srcfile3" 1
> +	    file_name "$srcfile4" 1
> +
> +	    program {
> +		{DW_LNE_set_address line_label_1}
> +		{DW_LNS_advance_line 15}
> +		{DW_LNS_copy}
> +
> +		{DW_LNE_set_address line_label_2}
> +		{DW_LNS_advance_line 1}
> +		{DW_LNS_copy}
> +
> +		{DW_LNS_set_file 2}
> +		{DW_LNE_set_address line_label_3}
> +		{DW_LNS_advance_line 4}
> +		{DW_LNS_copy}
> +
> +		{DW_LNE_set_address line_label_4}
> +		{DW_LNS_advance_line 1}
> +		{DW_LNS_copy}
> +
> +		{DW_LNS_advance_line -4}
> +		{DW_LNS_set_file 1}
> +		{DW_LNS_negate_stmt}
> +		{DW_LNS_copy}
> +
> +		{DW_LNS_set_file 2}
> +		{DW_LNE_set_address line_label_5}
> +		{DW_LNS_advance_line 5}
> +		{DW_LNS_copy}
> +
> +		{DW_LNS_negate_stmt}
> +		{DW_LNS_set_file 1}
> +		{DW_LNE_set_address line_label_6}
> +		{DW_LNS_advance_line 1}
> +		{DW_LNS_copy}
> +
> +		{DW_LNE_set_address line_label_7}
> +		{DW_LNE_end_sequence}
> +	    }
> +	}
> +    }
> +
> +    if { [prepare_for_testing "failed to prepare" ${testfile}-${tag} \
> +	      [list $srcfile $asm_file] $build_options] } {
> +	return -1
> +    }
> +
> +    if ![runto_main] {
> +	return -1
> +    }
> +
> +    # Delete all breakpoints so that the output of "info breakpoints"
> +    # below will only contain a single breakpoint.
> +    delete_breakpoints
> +
> +    # Place a breakpoint within the function in the header file.
> +    gdb_breakpoint "${srcfile4}:22"
> +
> +    # Check that the breakpoint was placed where we expected.  It should
> +    # appear at the requested line.  When the bug in GDB was present the
> +    # breakpoint would be placed on one of the following lines instead.
> +    gdb_test "info breakpoints" \
> +	".* in $func_name at \[^\r\n\]+${srcfile4}:22\\y.*" \
> +	"info breakpoints, $tag"
> +}
> +
> +do_test line_label_3 "callee" "range-a"
> +do_test line_label_5 "main" "range-b"
> diff --git a/gdb/testsuite/gdb.dwarf2/dw2-inline-header-2.exp b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-2.exp
> new file mode 100644
> index 00000000000..9f09f353273
> --- /dev/null
> +++ b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-2.exp
> @@ -0,0 +1,189 @@
> +# Copyright 2020 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/>.
> +
> +# Setup a line table where:
> +#
> +# | Addr | File | Line | Stmt | Inline |
> +# |------|------|------|------|--------|
> +# | 1    | 1    | 16   | Y    |        |
> +# | 2    | 1    | 17   | Y    |        |
> +# | 3    | 2    | 21   | Y    | X      |
> +# | 4    | 2    | 22   | Y    | X      |
> +# | 4    | 1    | 18   | N    | X      |
> +# | 5    | 1    | 19   | Y    |        |
> +# | 6    | 1    | 20   | Y    |        |
> +# | 7    | 1    | END  | Y    |        |
> +# |------|------|------|------|--------|
> +#
> +#
> +# Place the first brekpoint at file 2, line 22 and a second breakpoint
> +# at file 1, line 19.  A third breakpoint is placed at file 1, line
> +# 18, but as this line table entry will have been discarded[1] the
> +# third breakpoint will actually be placed at the same location as the
> +# second breakpoint.
> +#
> +# This test is designed to test GDB's internal behaviour with respect
> +# to discarding particular line table entries.  GCC and DWARF are
> +# starting to introduce the idea of line table views.  As the views
> +# information becomes better supported within GDB it is likely that
> +# this will become out of date.  This is fine, the test will have
> +# served its purpose by that point and can be deleted.
> +#
> +# [1] The entry for file 1, line 18 is discarded because it is at the
> +# same address as the previous entry, but the previous entry is-stmt,
> +# while line 18 is a non-stmt.
> +
> +load_lib dwarf.exp
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +if {![dwarf2_support]} {
> +    return 0
> +}
> +
> +# The .c files use __attribute__.
> +if [get_compiler_info] {
> +    return -1
> +}
> +if !$gcc_compiled {
> +    return 0
> +}
> +
> +standard_testfile dw2-inline-header-lbls.c dw2-inline-header.S \
> +    dw2-inline-header.c dw2-inline-header.h
> +
> +set build_options {nodebug optimize=-O1}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> +    global srcdir subdir srcfile srcfile3 srcfile4
> +    global build_options
> +    declare_labels lines_label callee_subprog_label
> +
> +    get_func_info main $build_options
> +
> +    cu {} {
> +	compile_unit {
> +	    {producer "gcc" }
> +	    {language @DW_LANG_C}
> +	    {name ${srcfile3}}
> +	    {low_pc 0 addr}
> +	    {stmt_list ${lines_label} DW_FORM_sec_offset}
> +	} {
> +	    callee_subprog_label: subprogram {
> +		{external 1 flag}
> +		{name callee}
> +		{inline 3 data1}
> +	    }
> +	    subprogram {
> +		{external 1 flag}
> +		{name main}
> +		{low_pc $main_start addr}
> +		{high_pc "$main_start + $main_len" addr}
> +	    } {
> +		inlined_subroutine {
> +		    {abstract_origin %$callee_subprog_label}
> +		    {low_pc line_label_3 addr}
> +		    {high_pc line_label_5 addr}
> +		    {call_file 1 data1}
> +		    {call_line 18 data1}
> +		}
> +	    }
> +	}
> +    }
> +
> +    lines {version 2 default_is_stmt 1} lines_label {
> +	include_dir "${srcdir}/${subdir}"
> +	file_name "$srcfile3" 1
> +	file_name "$srcfile4" 1
> +
> +	program {
> +	    {DW_LNE_set_address line_label_1}
> +	    {DW_LNS_advance_line 15}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_2}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNS_set_file 2}
> +	    {DW_LNE_set_address line_label_3}
> +	    {DW_LNS_advance_line 4}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_4}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNS_advance_line -4}
> +	    {DW_LNS_set_file 1}
> +	    {DW_LNS_negate_stmt}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_5}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_negate_stmt}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_6}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_7}
> +	    {DW_LNE_end_sequence}
> +	}
> +    }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> +	  [list $srcfile $asm_file] $build_options] } {
> +    return -1
> +}
> +
> +if ![runto_main] {
> +    return -1
> +}
> +
> +# Delete all breakpoints so that the output of "info breakpoints"
> +# below will only contain a single breakpoint.
> +delete_breakpoints
> +
> +# Place a breakpoint within the function in the header file.
> +gdb_breakpoint "${srcfile4}:22"
> +
> +# Check that the breakpoint was placed where we expected.  It should
> +# appear at the requested line.  When the bug in GDB was present the
> +# breakpoint would be placed on one of the following lines instead.
> +gdb_test "info breakpoints" \
> +    ".* in callee at \[^\r\n\]+${srcfile4}:22\\y.*" \
> +    "check for breakpoint at ${srcfile4}"
> +
> +# Delete all breakpoints so that the output of "info breakpoints"
> +# below will only contain a single breakpoint.
> +delete_breakpoints
> +
> +# Place a breakpoint within the function in the header file.
> +gdb_breakpoint "${srcfile3}:19"
> +
> +# Check that the breakpoint was placed where we expected.  It should
> +# appear at the requested line.  When the bug in GDB was present the
> +# breakpoint would be placed on one of the following lines instead.
> +gdb_test "info breakpoints" \
> +    ".* in main at \[^\r\n\]+${srcfile3}:19\\y.*" \
> +    "check for breakpoint at ${srcfile3}"
> +
> +# Line table entry for line 18 will have been discarded, so this
> +# brekpoint will be at the same location as line 19.
> +gdb_test "break ${srcfile3}:18" \
> +    "Note: breakpoint $decimal also set at pc $hex.*"
> diff --git a/gdb/testsuite/gdb.dwarf2/dw2-inline-header-3.exp b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-3.exp
> new file mode 100644
> index 00000000000..a3820f16d57
> --- /dev/null
> +++ b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-3.exp
> @@ -0,0 +1,193 @@
> +# Copyright 2020 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/>.
> +
> +# Setup a line table where:
> +#
> +# | Addr | File | Line | Stmt | Inline |
> +# |------|------|------|------|--------|
> +# | 1    | 1    | 16   | Y    |        |
> +# | 2    | 1    | 17   | Y    |        |
> +# | 3    | 2    | 21   | Y    | X      |
> +# | 4    | 2    | 22   | Y    | X      |
> +# | 4    | 1    | 18   | N    |        |
> +# | 5    | 1    | 19   | N    |        |
> +# | 6    | 1    | 20   | Y    |        |
> +# | 7    | 1    | END  | Y    |        |
> +# |------|------|------|------|--------|
> +#
> +# Break at file 2, line 22, then single instruction step forward.  We
> +# should pass through line 19 and then encounter line 20.
> +#
> +# Currently we don't expect GDB to see file 1, line 18, as this is a
> +# non-stmt line in a different file at the same address as the
> +# previous is-stmt line.
> +
> +load_lib dwarf.exp
> +
> +# This test can only be run on targets which support DWARF-2 and use gas.
> +if {![dwarf2_support]} {
> +    return 0
> +}
> +
> +# The .c files use __attribute__.
> +if [get_compiler_info] {
> +    return -1
> +}
> +if !$gcc_compiled {
> +    return 0
> +}
> +
> +standard_testfile dw2-inline-header-lbls.c dw2-inline-header.S \
> +    dw2-inline-header.c dw2-inline-header.h
> +
> +set build_options {nodebug optimize=-O1}
> +
> +set asm_file [standard_output_file $srcfile2]
> +Dwarf::assemble $asm_file {
> +    global srcdir subdir srcfile srcfile3 srcfile4
> +    global build_options
> +    declare_labels lines_label callee_subprog_label
> +
> +    get_func_info main $build_options
> +
> +    cu {} {
> +	compile_unit {
> +	    {producer "gcc" }
> +	    {language @DW_LANG_C}
> +	    {name ${srcfile3}}
> +	    {low_pc 0 addr}
> +	    {stmt_list ${lines_label} DW_FORM_sec_offset}
> +	} {
> +	    callee_subprog_label: subprogram {
> +		{external 1 flag}
> +		{name callee}
> +		{inline 3 data1}
> +	    }
> +	    subprogram {
> +		{external 1 flag}
> +		{name main}
> +		{low_pc $main_start addr}
> +		{high_pc "$main_start + $main_len" addr}
> +	    } {
> +		inlined_subroutine {
> +		    {abstract_origin %$callee_subprog_label}
> +		    {low_pc line_label_3 addr}
> +		    {high_pc line_label_5 addr}
> +		    {call_file 1 data1}
> +		    {call_line 18 data1}
> +		}
> +	    }
> +	}
> +    }
> +
> +    lines {version 2 default_is_stmt 1} lines_label {
> +	include_dir "${srcdir}/${subdir}"
> +	file_name "$srcfile3" 1
> +	file_name "$srcfile4" 1
> +
> +	program {
> +	    {DW_LNE_set_address line_label_1}
> +	    {DW_LNS_advance_line 15}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_2}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNS_set_file 2}
> +	    {DW_LNE_set_address line_label_3}
> +	    {DW_LNS_advance_line 4}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_4}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNS_advance_line -4}
> +	    {DW_LNS_set_file 1}
> +	    {DW_LNS_negate_stmt}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_5}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_6}
> +	    {DW_LNS_advance_line 1}
> +	    {DW_LNS_negate_stmt}
> +	    {DW_LNS_copy}
> +
> +	    {DW_LNE_set_address line_label_7}
> +	    {DW_LNE_end_sequence}
> +	}
> +    }
> +}
> +
> +if { [prepare_for_testing "failed to prepare" ${testfile} \
> +	  [list $srcfile $asm_file] $build_options] } {
> +    return -1
> +}
> +
> +if ![runto_main] {
> +    return -1
> +}
> +
> +# Delete all breakpoints so that the output of "info breakpoints"
> +# below will only contain a single breakpoint.
> +delete_breakpoints
> +
> +# Place a breakpoint within the function in the header file.
> +gdb_breakpoint "${srcfile4}:22"
> +
> +# Check that the breakpoint was placed where we expected.  It should
> +# appear at the requested line.  When the bug in GDB was present the
> +# breakpoint would be placed on one of the following lines instead.
> +gdb_test "info breakpoints" \
> +    ".* in callee at \[^\r\n\]+${srcfile4}:22\\y.*"
> +
> +gdb_continue_to_breakpoint "${srcfile4}:22" \
> +    ".* ${srcfile4} : 22 .*"
> +
> +# Now single instruction step forward.  Eventually we should hit
> +# ${srcfile3}:20, but before we do we should hit the non-statement
> +# line ${srcfile3}:19.
> +#
> +# We don't know how many instructions we'll need to step, but 100
> +# should be enough for everyone (surely), and this stops us looping
> +# forever if something goes wrong.
> +set found_line_19 0
> +set found_line_20 0
> +set keep_going 1
> +for { set i 0 } { $i < 100 && $keep_going } { incr i } {
> +    set keep_going 0
> +    gdb_test_multiple "stepi" "stepi ${i}" {
> +	-re "${srcfile3} : 19 .*${gdb_prompt} " {
> +	    set found_line_19 1
> +	    set keep_going 1
> +	}
> +
> +	-re "${srcfile3} : 20 .*${gdb_prompt} " {
> +	    set found_line_20 1
> +	}
> +
> +	-re "${srcfile4} : 22 .*${gdb_prompt} " {
> +	    # Not left line 22 yet.
> +	    set keep_going 1
> +	}
> +    }
> +}
> +
> +gdb_assert { $found_line_19 && $found_line_20 } \
> +    "found line 19 and 20"
> diff --git a/gdb/testsuite/gdb.dwarf2/dw2-inline-header-lbls.c b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-lbls.c
> new file mode 100644
> index 00000000000..a1b7b17cbeb
> --- /dev/null
> +++ b/gdb/testsuite/gdb.dwarf2/dw2-inline-header-lbls.c
> @@ -0,0 +1,46 @@
> +/* Copyright 2020 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/>.  */
> +
> +/* Used to insert labels with which we can build a fake line table.  */
> +#define LL(N) asm ("line_label_" #N ": .globl line_label_" #N)
> +
> +volatile int var;
> +volatile int bar;
> +
> +/* Generate some code to take up some space.  */
> +#define FILLER do { \
> +    var = 99;	    \
> +} while (0)
> +
> +int
> +main ()
> +{					/* main prologue */
> +  asm ("main_label: .globl main_label");
> +  LL (1);	// F1, Ln 16
> +  FILLER;
> +  LL (2);	// F1, Ln 17
> +  FILLER;
> +  LL (3);	// F2, Ln 21
> +  FILLER;
> +  LL (4);	// F2, Ln 22 // F1, Ln 18, !S
> +  FILLER;
> +  LL (5);	// F1, Ln 19 !S
> +  FILLER;
> +  LL (6);	// F1, Ln 20
> +  FILLER;
> +  LL (7);
> +  FILLER;
> +  return 0;				/* main end */
> +}
> diff --git a/gdb/testsuite/gdb.dwarf2/dw2-inline-header.c b/gdb/testsuite/gdb.dwarf2/dw2-inline-header.c
> new file mode 100644
> index 00000000000..a8331268a09
> --- /dev/null
> +++ b/gdb/testsuite/gdb.dwarf2/dw2-inline-header.c
> @@ -0,0 +1,24 @@
> +/* Copyright 2020 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/>.  */
> +
> +/* dw2-inline-header.c : 16 */
> +/* dw2-inline-header.c : 17 */
> +/* dw2-inline-header.c : 18 */
> +/* dw2-inline-header.c : 19 */
> +/* dw2-inline-header.c : 20 */
> +/* dw2-inline-header.c : 21 */
> +/* dw2-inline-header.c : 22 */
> +/* dw2-inline-header.c : 23 */
> +/* dw2-inline-header.c : 24 */
> diff --git a/gdb/testsuite/gdb.dwarf2/dw2-inline-header.h b/gdb/testsuite/gdb.dwarf2/dw2-inline-header.h
> new file mode 100644
> index 00000000000..7233acbcd76
> --- /dev/null
> +++ b/gdb/testsuite/gdb.dwarf2/dw2-inline-header.h
> @@ -0,0 +1,24 @@
> +/* Copyright 2020 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/>.  */
> +
> +/* dw2-inline-header.h : 16 */
> +/* dw2-inline-header.h : 17 */
> +/* dw2-inline-header.h : 18 */
> +/* dw2-inline-header.h : 19 */
> +/* dw2-inline-header.h : 20 */
> +/* dw2-inline-header.h : 21 */
> +/* dw2-inline-header.h : 22 */
> +/* dw2-inline-header.h : 23 */
> +/* dw2-inline-header.h : 24 */