Discussion:
[PATCH v3] Logical short circuiting with argument lists
Richard Bunt
2018-09-14 08:08:58 UTC
Permalink
Version 3 addresses review comments:

* Removed forward declaration of skip_undetermined_arglist.
* Made definition of skip_undetermined_arglist static.
* Added introductory comment to skip_undetermined_arglist.
* Dropped (*pos)+=3 into the block.
* Adjusted Tcl bracing.
* Removed string splitting in the test case.

---

When evaluating Fortran expressions such as the following:

print truth_table(1,1) .OR. truth_table(2,1)

where truth_table(1,1) evaluates to true, the debugger would report that
it could not perform substring operations on this type. This patch
addresses this issue.

Investigation revealed that EVAL_SKIP was not being handled correctly
for all types serviced by the OP_F77_UNDETERMINED_ARGLIST case in
evaluate_subexp_standard. While skipping an undetermined argument list
the type is resolved to be an integer (as this is what evaluate_subexp
returns when skipping) and so it was not possible to delegate to the
appropriate case (e.g. array, function call).

The solution implemented here updates OP_VAR_VALUE to return correct
type information when skipping. This way OP_F77_UNDETERMINED_ARGLIST
can delegate the skipping to the appropriate case or routine, which
should know how to skip/evaluate the type in question.

koenig.exp was updated to include a testcase which exercises the
modified skip logic in OP_VAR_VALUE, as it falls through from
OP_ADL_FUNC.

This patch has been tested for regressions with GCC 7.3 on aarch64,
ppc64le and x86_64.

2018-08-13 Richard Bunt <***@arm.com>
Chris January <***@arm.com>

* eval.c (skip_undetermined_arglist): Skip argument list helper.
(evaluate_subexp_standard): Return a dummy type when
honoring EVAL_SKIP in OP_VAR_VALUE and handle skipping in the
OP_F77_UNDETERMINED_ARGLIST case.
* expression.h (enum noside): Update comment.

gdb/testsuite/ChangeLog:

2018-08-13 Richard Bunt <***@arm.com>
Chris January <***@arm.com>

* gdb.cp/koenig.exp: Extend to test logical short circuiting.
* gdb.fortran/short-circuit-argument-list.exp: New file.
* gdb.fortran/short-circuit-argument-list.f90: New test.
---
gdb/eval.c | 46 +++++++--
gdb/expression.h | 4 +-
gdb/testsuite/gdb.cp/koenig.exp | 3 +
.../gdb.fortran/short-circuit-argument-list.exp | 106 +++++++++++++++++++++
.../gdb.fortran/short-circuit-argument-list.f90 | 78 +++++++++++++++
5 files changed, 229 insertions(+), 8 deletions(-)
create mode 100644 gdb/testsuite/gdb.fortran/short-circuit-argument-list.exp
create mode 100644 gdb/testsuite/gdb.fortran/short-circuit-argument-list.f90

diff --git a/gdb/eval.c b/gdb/eval.c
index 2e08e9355f5e1ba8bf0ec9818e2291e23676100c..991c6f505cf648d40938e3033564f42d6ea661f3 100644
--- a/gdb/eval.c
+++ b/gdb/eval.c
@@ -1234,6 +1234,19 @@ evaluate_funcall (type *expect_type, expression *exp, int *pos,
return eval_call (exp, noside, nargs, argvec, var_func_name, expect_type);
}

+/* Helper for skipping all the arguments in an undetermined argument list.
+ This function was designed for use in the OP_F77_UNDETERMINED_ARGLIST
+ case of evaluate_subexp_standard as multiple, but not all, code paths
+ require a generic skip. */
+
+static void
+skip_undetermined_arglist (int nargs, struct expression *exp, int *pos,
+ enum noside noside)
+{
+ for (int i = 0; i < nargs; ++i)
+ evaluate_subexp (NULL_TYPE, exp, pos, noside);
+}
+
struct value *
evaluate_subexp_standard (struct type *expect_type,
struct expression *exp, int *pos,
@@ -1282,16 +1295,21 @@ evaluate_subexp_standard (struct type *expect_type,

case OP_ADL_FUNC:
case OP_VAR_VALUE:
- (*pos) += 3;
- if (noside == EVAL_SKIP)
- return eval_skip_value (exp);
-
{
+ (*pos) += 3;
symbol *var = exp->elts[pc + 2].symbol;
if (TYPE_CODE (SYMBOL_TYPE (var)) == TYPE_CODE_ERROR)
error_unknown_type (SYMBOL_PRINT_NAME (var));
-
- return evaluate_var_value (noside, exp->elts[pc + 1].block, var);
+ if (noside != EVAL_SKIP)
+ {
+ return evaluate_var_value (noside, exp->elts[pc + 1].block, var);
+ }
+ else
+ {
+ /* Return a dummy value of the correct type when skipping, so
+ that parent functions know what is to be skipped. */
+ return allocate_value (SYMBOL_TYPE (var));
+ }
}

case OP_VAR_MSYM_VALUE:
@@ -1929,13 +1947,27 @@ evaluate_subexp_standard (struct type *expect_type,
if (exp->elts[*pos].opcode == OP_RANGE)
return value_f90_subarray (arg1, exp, pos, noside);
else
- goto multi_f77_subscript;
+ {
+ if (noside == EVAL_SKIP)
+ {
+ skip_undetermined_arglist (nargs, exp, pos, noside);
+ /* Return the dummy value with the correct type. */
+ return arg1;
+ }
+ goto multi_f77_subscript;
+ }

case TYPE_CODE_STRING:
if (exp->elts[*pos].opcode == OP_RANGE)
return value_f90_subarray (arg1, exp, pos, noside);
else
{
+ if (noside == EVAL_SKIP)
+ {
+ skip_undetermined_arglist (nargs, exp, pos, noside);
+ /* Return the dummy value with the correct type. */
+ return arg1;
+ }
arg2 = evaluate_subexp_with_coercion (exp, pos, noside);
return value_subscript (arg1, value_as_long (arg2));
}
diff --git a/gdb/expression.h b/gdb/expression.h
index 9f26bb8d60ba6905abaf1afd4473dbf5946b958a..bc7625f98427fb3168c2e00053ab947ef8303d55 100644
--- a/gdb/expression.h
+++ b/gdb/expression.h
@@ -118,7 +118,9 @@ extern int parse_completion;
enum noside
{
EVAL_NORMAL,
- EVAL_SKIP, /* Only effect is to increment pos. */
+ EVAL_SKIP, /* Only effect is to increment pos.
+ Return type information where
+ possible. */
EVAL_AVOID_SIDE_EFFECTS /* Don't modify any variables or
call any functions. The value
returned will have the correct
diff --git a/gdb/testsuite/gdb.cp/koenig.exp b/gdb/testsuite/gdb.cp/koenig.exp
index b7dff90e6b4c5ed5b58eda00cbd0eb24681dcd16..0725c1bc91fe165d5819fbf5c0fdd77c09569dad 100644
--- a/gdb/testsuite/gdb.cp/koenig.exp
+++ b/gdb/testsuite/gdb.cp/koenig.exp
@@ -33,6 +33,9 @@ gdb_test "p first(c)" "= 11"
# the qualifying parameter
gdb_test "p second(0,0,c,0,0)" "= 33"

+# Test evaluating function under EVAL_SKIP
+gdb_test "p true || second(0,0,c,0,0)" "= true"
+
# Test the name "entry" being used for `***@entry' entry values.
gdb_test "p entry (c)" " = 44"

diff --git a/gdb/testsuite/gdb.fortran/short-circuit-argument-list.exp b/gdb/testsuite/gdb.fortran/short-circuit-argument-list.exp
new file mode 100644
index 0000000000000000000000000000000000000000..1b948bb19d289e5327cb2caa3bf665e33cd6d98d
--- /dev/null
+++ b/gdb/testsuite/gdb.fortran/short-circuit-argument-list.exp
@@ -0,0 +1,106 @@
+# Copyright 2018 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/> .
+
+# Test evaluating logical expressions that contain array references, function
+# calls and substring operations that are to be skipped due to short
+# circuiting.
+
+if {[skip_fortran_tests]} { return -1 }
+
+standard_testfile ".f90"
+
+if {[prepare_for_testing ${testfile}.exp ${testfile} ${srcfile} {debug f90}]} {
+ return -1
+}
+
+if {![runto [gdb_get_line_number "post_truth_table_init"]]} then {
+ perror "couldn't run to breakpoint post_truth_table_init"
+ continue
+}
+
+# Vary conditional and input over the standard truth table.
+# Test that the debugger can evaluate expressions of the form
+# a(x,y) .OR./.AND. a(a,b) correctly.
+foreach_with_prefix truth_table_index {1 2 3 4} {
+ gdb_test "p truth_table($truth_table_index, 1) .OR. truth_table($truth_table_index, 2)" \
+ "[expr $truth_table_index > 1 ? \".TRUE.\" : \".FALSE.\"]"
+}
+
+foreach_with_prefix truth_table_index {1 2 3 4} {
+ gdb_test "p truth_table($truth_table_index, 1) .AND. truth_table($truth_table_index, 2)" \
+ "[expr $truth_table_index > 3 ? \".TRUE.\" : \".FALSE.\"]"
+}
+
+# Vary number of function arguments to skip.
+set argument_list ""
+foreach_with_prefix arg {"No" "One" "Two"} {
+ set trimmed_args [string trimright $argument_list ,]
+ set arg_lower [string tolower $arg]
+ gdb_test "p function_no_arg_false() .OR. function_${arg_lower}_arg($trimmed_args)" \
+ " $arg, return true.\r\n\\\$$decimal = .TRUE."
+ # Check the skipped function has not printed anything by asserting the
+ # absence of the full stop from its message.
+ gdb_test "p .TRUE. .OR. function_${arg_lower}_arg($trimmed_args)" \
+ "\[^.\]\r\n\\\$$decimal = .TRUE."
+ append argument_list " .TRUE.,"
+}
+
+# Check nested calls
+gdb_test "p function_one_arg(.FALSE. .OR. function_no_arg())" \
+ " No, return true.\r\n One, return true.\r\n\\\$$decimal = .TRUE."
+
+gdb_test "p function_one_arg(.TRUE. .OR. function_no_arg())" \
+ "\[^.\]\r\n One, return true.\r\n\\\$$decimal = .TRUE."
+
+# Vary number of components in the expression to skip.
+set expression "p .TRUE."
+foreach_with_prefix expression_components {1 2 3 4} {
+ set expression "$expression .OR. function_one_arg(.TRUE.)"
+ gdb_test "$expression" \
+ "\\\$$decimal = .TRUE."
+}
+
+# Check parsing skipped substring operations.
+gdb_test "p .TRUE. .OR. binary_string(1)" "\\\$$decimal = .TRUE."
+
+# Check parsing skipped substring operations with ranges. These should all
+# return true as the result is > 0.
+# The second binary_string access is important as an incorrect pos update
+# will not be picked up by a single access.
+foreach_with_prefix range1 {"1:2" ":" ":2" "1:"} {
+ foreach_with_prefix range2 {"1:2" ":" ":2" "1:"} {
+ gdb_test "p .TRUE. .OR. binary_string($range1) .OR. binary_string($range2)" \
+ "\\\$$decimal = .TRUE."
+ }
+}
+
+# Skip multi-dimensional arrays with ranges.
+foreach_with_prefix range1 {"1:2" ":" ":2" "1:"} {
+ foreach_with_prefix range2 {"1:2" ":" ":2" "1:"} {
+ gdb_test "p .TRUE. .OR. binary_string($range1) .OR. truth_table($range2, 1)" \
+ "\\\$$decimal = .TRUE."
+ }
+}
+
+# Check evaluation of substring operations in logical expressions.
+gdb_test "p .FALSE. .OR. binary_string(1)" "\\\$$decimal = .FALSE."
+
+# Function call and substring skip.
+gdb_test "p .TRUE. .OR. function_one_arg(binary_string(1))" \
+ "\\\$$decimal = .TRUE."
+
+# Function call and array skip.
+gdb_test "p .TRUE. .OR. function_array(binary_string)" \
+ "\\\$$decimal = .TRUE."
diff --git a/gdb/testsuite/gdb.fortran/short-circuit-argument-list.f90 b/gdb/testsuite/gdb.fortran/short-circuit-argument-list.f90
new file mode 100644
index 0000000000000000000000000000000000000000..5d8b9c73a705c598b513fd85a8d710c7a7dabebf
--- /dev/null
+++ b/gdb/testsuite/gdb.fortran/short-circuit-argument-list.f90
@@ -0,0 +1,78 @@
+! Copyright 2018 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/> .
+
+! Source code for short-circuit-argument-list.exp.
+
+logical function function_no_arg()
+ print *, "No, return true."
+ function_no_arg = .TRUE.
+end function function_no_arg
+
+logical function function_no_arg_false()
+ function_no_arg_false = .FALSE.
+end function function_no_arg_false
+
+logical function function_one_arg(x)
+ logical, intent(in) :: x
+ print *, "One, return true."
+ function_one_arg = .TRUE.
+end function function_one_arg
+
+logical function function_two_arg(x, y)
+ logical, intent(in) :: x, y
+ print *, "Two, return true."
+ function_two_arg = .TRUE.
+end function function_two_arg
+
+logical function function_array(logical_array)
+ logical, dimension(4,2), target, intent(in) :: logical_array
+ logical, dimension(:,:), pointer :: p
+ p => logical_array
+ print *, "Array, return true.", p(1,1), logical_array(1,1)
+ function_array = .TRUE.
+end function function_array
+
+program generate_truth_table
+ implicit none
+ interface
+ logical function function_no_arg()
+ end function function_no_arg
+ logical function function_no_arg_false()
+ end function
+ logical function function_one_arg(x)
+ logical, intent(in) :: x
+ end function
+ logical function function_two_arg(x, y)
+ logical, intent(in) :: x, y
+ end function
+ logical function function_array(logical_array)
+ logical, dimension(4,2), target, intent(in) :: logical_array
+ end function function_array
+ end interface
+ logical, dimension (4,2) :: truth_table
+ logical :: a, b, c, d, e
+ character(2) :: binary_string
+ binary_string = char(0) // char(1)
+ truth_table = .FALSE.
+ truth_table(3:4,1) = .TRUE.
+ truth_table(2::2,2) = .TRUE.
+ a = function_no_arg() ! post_truth_table_init
+ b = function_no_arg_false()
+ c = function_one_arg(b)
+ d = function_two_arg(a, b)
+ e = function_array(truth_table)
+ print *, truth_table(:, 1), a, b, e
+ print *, truth_table(:, 2), c, d
+end program generate_truth_table
--
2.7.4
Tom Tromey
2018-09-14 21:00:37 UTC
Permalink
Richard> Version 3 addresses review comments:

Richard> * Removed forward declaration of skip_undetermined_arglist.
Richard> * Made definition of skip_undetermined_arglist static.
Richard> * Added introductory comment to skip_undetermined_arglist.
Richard> * Dropped (*pos)+=3 into the block.
Richard> * Adjusted Tcl bracing.
Richard> * Removed string splitting in the test case.

Thanks again for the patch.

Richard> + if (noside != EVAL_SKIP)
Richard> + {
Richard> + return evaluate_var_value (noside, exp->elts[pc + 1].block, var);
Richard> + }

The gdb style is no braces if there's just a single line.

Richard> + else
Richard> + {
Richard> + /* Return a dummy value of the correct type when skipping, so
Richard> + that parent functions know what is to be skipped. */
Richard> + return allocate_value (SYMBOL_TYPE (var));

... compared to here, where braces are mandatory.

This patch is ok with that first set of braces removed.
(You don't have to re-send it.)

thanks,
Tom
Richard Bunt
2018-09-18 18:39:29 UTC
Permalink
Post by Tom Tromey
Richard> * Removed forward declaration of skip_undetermined_arglist.
Richard> * Made definition of skip_undetermined_arglist static.
Richard> * Added introductory comment to skip_undetermined_arglist.
Richard> * Dropped (*pos)+=3 into the block.
Richard> * Adjusted Tcl bracing.
Richard> * Removed string splitting in the test case.
Thanks again for the patch.
Richard> + if (noside != EVAL_SKIP)
Richard> + {
Richard> + return evaluate_var_value (noside, exp->elts[pc + 1].block, var);
Richard> + }
The gdb style is no braces if there's just a single line.
Richard> + else
Richard> + {
Richard> + /* Return a dummy value of the correct type when skipping, so
Richard> + that parent functions know what is to be skipped. */
Richard> + return allocate_value (SYMBOL_TYPE (var));
... compared to here, where braces are mandatory.
This patch is ok with that first set of braces removed.
(You don't have to re-send it.)
thanks,
Tom
Many thanks for reviewing this change. I will fix up the braces and push.

Rich.
Sergio Durigan Junior
2018-10-07 05:14:18 UTC
Permalink
Post by Richard Bunt
Post by Tom Tromey
Richard> * Removed forward declaration of skip_undetermined_arglist.
Richard> * Made definition of skip_undetermined_arglist static.
Richard> * Added introductory comment to skip_undetermined_arglist.
Richard> * Dropped (*pos)+=3 into the block.
Richard> * Adjusted Tcl bracing.
Richard> * Removed string splitting in the test case.
Thanks again for the patch.
Richard> + if (noside != EVAL_SKIP)
Richard> + {
Richard> + return evaluate_var_value (noside, exp->elts[pc + 1].block, var);
Richard> + }
The gdb style is no braces if there's just a single line.
Richard> + else
Richard> + {
Richard> + /* Return a dummy value of the correct type when skipping, so
Richard> + that parent functions know what is to be skipped. */
Richard> + return allocate_value (SYMBOL_TYPE (var));
... compared to here, where braces are mandatory.
This patch is ok with that first set of braces removed.
(You don't have to re-send it.)
thanks,
Tom
Many thanks for reviewing this change. I will fix up the braces and push.
Hi Richard,

Just a heads up, I've noticed a few failures when running
gdb.fortran/short-circuit-argument-list.exp on a
native{,-extended}-gdbserver board:

FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=No: p function_no_arg_false() .OR. function_no_arg()
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=One: p function_no_arg_false() .OR. function_one_arg( .TRUE.)
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=Two: p function_no_arg_false() .OR. function_two_arg( .TRUE., .TRUE.)
FAIL: gdb.fortran/short-circuit-argument-list.exp: p function_one_arg(.FALSE. .OR. function_no_arg())
FAIL: gdb.fortran/short-circuit-argument-list.exp: p function_one_arg(.TRUE. .OR. function_no_arg())

The failures are easily reproducible, but let me know if you need a
gdb.log excerpt. I haven't had the time to investigate them further.

Thanks,
--
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF 31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/
Richard Bunt
2018-12-03 13:22:21 UTC
Permalink
Hi Sergio,

My apologies, I completely missed this reply. I'll investigate these failures.

Rich.
Post by Sergio Durigan Junior
Post by Richard Bunt
Post by Tom Tromey
Richard> * Removed forward declaration of skip_undetermined_arglist.
Richard> * Made definition of skip_undetermined_arglist static.
Richard> * Added introductory comment to skip_undetermined_arglist.
Richard> * Dropped (*pos)+=3 into the block.
Richard> * Adjusted Tcl bracing.
Richard> * Removed string splitting in the test case.
Thanks again for the patch.
Richard> + if (noside != EVAL_SKIP)
Richard> + {
Richard> + return evaluate_var_value (noside, exp->elts[pc + 1].block, var);
Richard> + }
The gdb style is no braces if there's just a single line.
Richard> + else
Richard> + {
Richard> + /* Return a dummy value of the correct type when skipping, so
Richard> + that parent functions know what is to be skipped. */
Richard> + return allocate_value (SYMBOL_TYPE (var));
... compared to here, where braces are mandatory.
This patch is ok with that first set of braces removed.
(You don't have to re-send it.)
thanks,
Tom
Many thanks for reviewing this change. I will fix up the braces and push.
Hi Richard,
Just a heads up, I've noticed a few failures when running
gdb.fortran/short-circuit-argument-list.exp on a
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=No: p function_no_arg_false() .OR. function_no_arg()
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=One: p function_no_arg_false() .OR. function_one_arg( .TRUE.)
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=Two: p function_no_arg_false() .OR. function_two_arg( .TRUE., .TRUE.)
FAIL: gdb.fortran/short-circuit-argument-list.exp: p function_one_arg(.FALSE. .OR. function_no_arg())
FAIL: gdb.fortran/short-circuit-argument-list.exp: p function_one_arg(.TRUE. .OR. function_no_arg())
The failures are easily reproducible, but let me know if you need a
gdb.log excerpt. I haven't had the time to investigate them further
Richard Bunt
2018-12-03 14:58:29 UTC
Permalink
On 12/3/18 1:22 PM, Richard Bunt wrote:

The assertions in the test check against the program's standard out which
appears in a different place when the test is run using gdb server. So the
functionality is fine, but the test case is flawed.

I will make a patch to fix this.

Many thanks,

Rich.
Post by Richard Bunt
Hi Sergio,
My apologies, I completely missed this reply. I'll investigate these failures.
Rich.
Post by Sergio Durigan Junior
Post by Richard Bunt
Post by Tom Tromey
Richard> * Removed forward declaration of skip_undetermined_arglist.
Richard> * Made definition of skip_undetermined_arglist static.
Richard> * Added introductory comment to skip_undetermined_arglist.
Richard> * Dropped (*pos)+=3 into the block.
Richard> * Adjusted Tcl bracing.
Richard> * Removed string splitting in the test case.
Thanks again for the patch.
Richard> + if (noside != EVAL_SKIP)
Richard> + {
Richard> + return evaluate_var_value (noside, exp->elts[pc + 1].block, var);
Richard> + }
The gdb style is no braces if there's just a single line.
Richard> + else
Richard> + {
Richard> + /* Return a dummy value of the correct type when skipping, so
Richard> + that parent functions know what is to be skipped. */
Richard> + return allocate_value (SYMBOL_TYPE (var));
... compared to here, where braces are mandatory.
This patch is ok with that first set of braces removed.
(You don't have to re-send it.)
thanks,
Tom
Many thanks for reviewing this change. I will fix up the braces and push.
Hi Richard,
Just a heads up, I've noticed a few failures when running
gdb.fortran/short-circuit-argument-list.exp on a
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=No: p function_no_arg_false() .OR. function_no_arg()
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=One: p function_no_arg_false() .OR. function_one_arg( .TRUE.)
FAIL: gdb.fortran/short-circuit-argument-list.exp: arg=Two: p function_no_arg_false() .OR. function_two_arg( .TRUE., .TRUE.)
FAIL: gdb.fortran/short-circuit-argument-list.exp: p function_one_arg(.FALSE. .OR. function_no_arg())
FAIL: gdb.fortran/short-circuit-argument-list.exp: p function_one_arg(.TRUE. .OR. function_no_arg())
The failures are easily reproducible, but let me know if you need a
gdb.log excerpt. I haven't had the time to investigate
Loading...