[PATCH 0/2] Fix crash in 'bt' with Ada

[PATCH 0/2] Fix crash in 'bt' with Ada

[Tom Tromey]

GDB was crashing when doing a 'bt' of a complicated Ada program.
I tracked this down to a combination of two bugs.

I couldn't write a test that would reproduce the crash, but I did
manage to write one that shows a related problem.  This is included in
the second patch.

The first patch is not Ada-specific.  I couldn't come up with an
isolated test for this change.

Regression tested on x86-64 Fedora 32.

Tom

[PATCH 1/2] Use common_val_print_checked in print_variable_and_value

[Tom Tromey]

GDB was crashing in "bt full" for a large Ada program.  I tracked this
down to a combination of bugs.

The first bug, fixed here, is that print_variable_and_value calls
common_val_print.  Normally only the value-printing implementation
should call this; from the top level, common_val_print_checked is
preferred, because it correctly handles values that are wholly
optimized out.

I wasn't able to write a test case for this.

gdb/ChangeLog
2021-04-01  Tom Tromey  <tromey@adacore.com>

	* printcmd.c (print_variable_and_value): Use
	common_val_print_checked.
---
 gdb/ChangeLog  | 5 +++++
 gdb/printcmd.c | 2 +-
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/gdb/printcmd.c b/gdb/printcmd.c
index c82f709136f..a4f62f207d0 100644
--- a/gdb/printcmd.c
+++ b/gdb/printcmd.c
@@ -2387,7 +2387,7 @@ print_variable_and_value (const char *name, struct symbol *var,
       val = read_var_value (var, NULL, frame);
       get_user_print_options (&opts);
       opts.deref_ref = 1;
-      common_val_print (val, stream, indent, &opts, current_language);
+      common_val_print_checked (val, stream, indent, &opts, current_language);
 
       /* common_val_print invalidates FRAME when a pretty printer calls inferior
 	 function.  */
-- 
2.26.2

Re: [PATCH 1/2] Use common_val_print_checked in print_variable_and_value

[Tom de Vries]

On 4/1/21 7:58 PM, Tom Tromey wrote:
> GDB was crashing in "bt full" for a large Ada program.  I tracked this
> down to a combination of bugs.
> 
> The first bug, fixed here, is that print_variable_and_value calls
> common_val_print.  Normally only the value-printing implementation
> should call this; from the top level, common_val_print_checked is
> preferred, because it correctly handles values that are wholly
> optimized out.
> 
> I wasn't able to write a test case for this.
> 

Caused:
...
FAIL: gdb.opt/inline-locals.exp: info locals above bar 2
FAIL: gdb.opt/inline-locals.exp: info locals above bar 3
...
filed as https://sourceware.org/bugzilla/show_bug.cgi?id=27742 .

Thanks,
- Tom

> gdb/ChangeLog
> 2021-04-01  Tom Tromey  <tromey@adacore.com>
> 
> 	* printcmd.c (print_variable_and_value): Use
> 	common_val_print_checked.
> ---
>  gdb/ChangeLog  | 5 +++++
>  gdb/printcmd.c | 2 +-
>  2 files changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/gdb/printcmd.c b/gdb/printcmd.c
> index c82f709136f..a4f62f207d0 100644
> --- a/gdb/printcmd.c
> +++ b/gdb/printcmd.c
> @@ -2387,7 +2387,7 @@ print_variable_and_value (const char *name, struct symbol *var,
>        val = read_var_value (var, NULL, frame);
>        get_user_print_options (&opts);
>        opts.deref_ref = 1;
> -      common_val_print (val, stream, indent, &opts, current_language);
> +      common_val_print_checked (val, stream, indent, &opts, current_language);
>  
>        /* common_val_print invalidates FRAME when a pretty printer calls inferior
>  	 function.  */
> 

Re: [PATCH 1/2] Use common_val_print_checked in print_variable_and_value

[Tom Tromey]

Tom> Caused:
Tom> ...
Tom> FAIL: gdb.opt/inline-locals.exp: info locals above bar 2
Tom> FAIL: gdb.opt/inline-locals.exp: info locals above bar 3
Tom> ...
Tom> filed as https://sourceware.org/bugzilla/show_bug.cgi?id=27742 .

I don't have the compiler needed to reproduce.
Could you try the appended?

thanks,
Tom

commit 0d00bb54e39ae86c9344ea3a509720255fb6f147
Author: Tom Tromey <tromey@adacore.com>
Date:   Fri Apr 16 14:08:49 2021 -0600

    Fix kfail patterns in inline-locals.exp
    
    PR gdb/27742 points out that my recent change to
    print_variable_and_value caused a regression in inline-locals.exp.  I
    can't reproduce this, but I came up with this patch based on the
    output shown in the bug.
    
    gdb/testsuite/ChangeLog
    2021-04-16  Tom Tromey  <tromey@adacore.com>
    
            PR gdb/27742:
            * gdb.opt/inline-locals.exp: Update kfail patterns.

diff --git a/gdb/testsuite/ChangeLog b/gdb/testsuite/ChangeLog
index 212f2ae0c5b..2fa302cf338 100644
--- a/gdb/testsuite/ChangeLog
+++ b/gdb/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2021-04-16  Tom Tromey  <tromey@adacore.com>
+
+	PR gdb/27742:
+	* gdb.opt/inline-locals.exp: Update kfail patterns.
+
 2021-04-16  Simon Marchi  <simon.marchi@efficios.com>
 
 	* boards/simavr.exp: Set debug_flags.
diff --git a/gdb/testsuite/gdb.opt/inline-locals.exp b/gdb/testsuite/gdb.opt/inline-locals.exp
index d0acb4ae8b5..ff9dd1b49ed 100644
--- a/gdb/testsuite/gdb.opt/inline-locals.exp
+++ b/gdb/testsuite/gdb.opt/inline-locals.exp
@@ -45,7 +45,7 @@ if { ! $no_frames } {
     gdb_test "info frame" ".*inlined into frame.*" "func1 inlined 2"
     set pass_re "array = \\{0 <repeats 64 times>\\}"
     set kfail_re [multi_line $pass_re \
-		      "array = {<optimized out> <repeats 64 times>}"]
+		      "array = <optimized out>"]
     gdb_test_multiple "info locals" "info locals above bar 2" {
 	-re -wrap $pass_re {
 	    pass $gdb_test_name
@@ -94,7 +94,7 @@ if { ! $no_frames } {
     gdb_test "info frame" ".*inlined into frame.*" "func1 inlined 3"
     set pass_re "array = {$decimal, \[^\r\n\]*}"
     set kfail_re [multi_line $pass_re \
-		      "array = {<optimized out> <repeats 64 times>}"]
+		      "array = <optimized out>"]
     gdb_test_multiple "info locals" "info locals above bar 3" {
 	-re -wrap $pass_re {
 	    pass $gdb_test_name

Re: [PATCH 1/2] Use common_val_print_checked in print_variable_and_value

[Tom de Vries]

On 4/16/21 10:11 PM, Tom Tromey wrote:
> Tom> Caused:
> Tom> ...
> Tom> FAIL: gdb.opt/inline-locals.exp: info locals above bar 2
> Tom> FAIL: gdb.opt/inline-locals.exp: info locals above bar 3
> Tom> ...
> Tom> filed as https://sourceware.org/bugzilla/show_bug.cgi?id=27742 .
> 
> I don't have the compiler needed to reproduce.
> Could you try the appended?
> 

Ack, that fixes it.

Thanks,
- Tom

[PATCH 2/2] Avoid crash in Ada value printing with optimized-out array

[Tom Tromey]

The Ada value-printing code could crash when printing an array which
had been optimized out.

The crash is difficult to reproduce, but I did manage to write a test
that at least shows that the previous behavior was incorrect -- before
the patch, the array is printed as if it is valid and every value is 0.

gdb/ChangeLog
2021-04-01  Tom Tromey  <tromey@adacore.com>

	* ada-valprint.c (ada_value_print_array): Handle optimized-out
	arrays.

gdb/testsuite/ChangeLog
2021-04-01  Tom Tromey  <tromey@adacore.com>

	* gdb.dwarf2/arr-stride.exp: Add test.
---
 gdb/ChangeLog                           |  5 ++++
 gdb/ada-valprint.c                      |  5 +++-
 gdb/testsuite/ChangeLog                 |  4 +++
 gdb/testsuite/gdb.dwarf2/arr-stride.exp | 33 ++++++++++++++++++++++++-
 4 files changed, 45 insertions(+), 2 deletions(-)

diff --git a/gdb/ada-valprint.c b/gdb/ada-valprint.c
index 0d5b6d73076..61c903bbed5 100644
--- a/gdb/ada-valprint.c
+++ b/gdb/ada-valprint.c
@@ -897,7 +897,10 @@ ada_value_print_array (struct value *val, struct ui_file *stream, int recurse,
 
   fprintf_filtered (stream, "(");
   print_optional_low_bound (stream, type, options);
-  if (TYPE_FIELD_BITSIZE (type, 0) > 0)
+
+  if (value_entirely_optimized_out (val))
+    val_print_optimized_out (val, stream);
+  else if (TYPE_FIELD_BITSIZE (type, 0) > 0)
     {
       const gdb_byte *valaddr = value_contents_for_printing (val);
       int offset_aligned = ada_aligned_value_addr (type, valaddr) - valaddr;
diff --git a/gdb/testsuite/gdb.dwarf2/arr-stride.exp b/gdb/testsuite/gdb.dwarf2/arr-stride.exp
index bf30f10c5c8..f25518f4558 100644
--- a/gdb/testsuite/gdb.dwarf2/arr-stride.exp
+++ b/gdb/testsuite/gdb.dwarf2/arr-stride.exp
@@ -31,7 +31,7 @@ Dwarf::assemble $asm_file {
                 {DW_AT_comp_dir /tmp}
         } {
 	    declare_labels integer_label array_elt_label array_label \
-                big_array_label
+                big_array_label struct_label
 
             integer_label: DW_TAG_base_type {
                 {DW_AT_byte_size 4 DW_FORM_sdata}
@@ -79,6 +79,34 @@ Dwarf::assemble $asm_file {
 		    {DW_AT_upper_bound 4 DW_FORM_data1}
 		}
 	    }
+
+	    struct_label: DW_TAG_structure_type {
+		{name struct_type}
+		{byte_size 16 DW_FORM_sdata}
+	    } {
+		member {
+		    {name intfield}
+		    {type :$integer_label}
+		    {data_member_location 0 DW_FORM_sdata}
+		}
+		member {
+		    {name arrayfield}
+		    {type :$array_label}
+		    {data_member_location 4 DW_FORM_sdata}
+		}
+	    }
+
+	    DW_TAG_variable {
+		{name the_struct}
+		{external 1 DW_FORM_flag}
+		{location {
+		    DW_OP_const1u 1
+		    DW_OP_stack_value
+		    DW_OP_piece 4
+		    DW_OP_piece 12
+		} SPECIAL_expr}
+		{type :$struct_label}
+	    }
 	}
     }
 }
@@ -95,3 +123,6 @@ gdb_test "ptype pck.table" \
 
 gdb_test "ptype pck.big_table" \
          "type = array \\(0 \\.\\. 4\\) of pck\\.item <packed: 8-bit elements>"
+
+gdb_test "print the_struct" \
+    "\\(intfield => 1, arrayfield => \\(0 => <optimized out>\\)\\)"
-- 
2.26.2

Re: [PATCH 0/2] Fix crash in 'bt' with Ada

[Tom Tromey]

>>>>> "Tom" == Tom Tromey <tromey@adacore.com> writes:

Tom> GDB was crashing when doing a 'bt' of a complicated Ada program.
Tom> I tracked this down to a combination of two bugs.

I'm going to check these in shortly.

Tom