Logical short circuiting with Fortran argument lists
Commit Message
Logical short circuiting with argument lists
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 proposed here is to hoist the skip from the function call
case to apply to all types of argument list.
While this patch allows a wider range of expressions to be evaluated, it
should be noted that this patch does not allow the skipping of arrays
which use Fortran array slicing, due to the inability of the debugger
to skip OP_RANGE.
This patch has been tested with GCC 7.3 on x86_64, aarch64 and ppc64le.
2018-06-06 Richard Bunt <richard.bunt@arm.com>
Chris January <chris.january@arm.com>
* eval.c (evaluate_subexp_standard): Skip evaluation of Fortran
argument lists when requested.
gdb/testsuite/ChangeLog:
2018-06-06 Richard Bunt <richard.bunt@arm.com>
Chris January <chris.january@arm.com>
* gdb.fortran/short-circuit-argument-list.exp: New file.
* gdb.fortran/short-circuit-argument-list.f90: New test.
---
gdb/eval.c | 13 +++-
.../gdb.fortran/short-circuit-argument-list.exp | 80
++++++++++++++++++++++
.../gdb.fortran/short-circuit-argument-list.f90 | 66 ++++++++++++++++++
3 files changed, 157 insertions(+), 2 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
Comments
On 2018-08-03 05:32, Richard Bunt wrote:
> Logical short circuiting with argument lists
>
> 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 proposed here is to hoist the skip from the function call
> case to apply to all types of argument list.
>
> While this patch allows a wider range of expressions to be evaluated,
> it
> should be noted that this patch does not allow the skipping of arrays
> which use Fortran array slicing, due to the inability of the debugger
> to skip OP_RANGE.
>
> This patch has been tested with GCC 7.3 on x86_64, aarch64 and ppc64le.
Hi Richard,
Thanks for the patch, hat sounds reasonnable. Just one question:
> diff --git a/gdb/eval.c b/gdb/eval.c
> index
> 9db6e7c69dad9e674f66991e2aee7dd8d66d80c7..2350c7faaf6221a5990cf5264fe7fe93b4f4be4c
> 100644
> --- a/gdb/eval.c
> +++ b/gdb/eval.c
> @@ -1902,6 +1902,17 @@ evaluate_subexp_standard (struct type
> *expect_type,
>
> /* First determine the type code we are dealing with. */
> arg1 = evaluate_subexp (NULL_TYPE, exp, pos, noside);
> +
> + /* Allow the short-circuiting of array accesses, function calls
> + and substring operations in logical expressions. */
> + if (noside == EVAL_SKIP)
> + {
> + /* Skip all the array subscripts. */
> + for (int i = 0; i < nargs; ++i)
> + evaluate_subexp (NULL_TYPE, exp, pos, noside);
> + return eval_skip_value (exp);
> + }
> +
> type = check_typedef (value_type (arg1));
> code = TYPE_CODE (type);
>
> @@ -1952,8 +1963,6 @@ evaluate_subexp_standard (struct type
> *expect_type,
> for (; tem <= nargs; tem++)
> argvec[tem] = evaluate_subexp_with_coercion (exp, pos, noside);
> argvec[tem] = 0; /* signal end of arglist */
> - if (noside == EVAL_SKIP)
> - return eval_skip_value (exp);
Is this change needed? Your test passes even when leaving that part
as-is.
Thanks,
Simon
On 08/03/2018 08:24 PM, Simon Marchi wrote:
> On 2018-08-03 05:32, Richard Bunt wrote:
>> Logical short circuiting with argument lists
>>
>> 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 proposed here is to hoist the skip from the function call
>> case to apply to all types of argument list.
>>
>> While this patch allows a wider range of expressions to be evaluated,
>> it
>> should be noted that this patch does not allow the skipping of arrays
>> which use Fortran array slicing, due to the inability of the debugger
>> to skip OP_RANGE.
>>
>> This patch has been tested with GCC 7.3 on x86_64, aarch64 and ppc64le.
>
> Hi Richard,
>
> Thanks for the patch, hat sounds reasonnable. Just one question:
>
>> diff --git a/gdb/eval.c b/gdb/eval.c
>> index
>> 9db6e7c69dad9e674f66991e2aee7dd8d66d80c7..2350c7faaf6221a5990cf5264fe7fe93b4f4be4c
>> 100644
>> --- a/gdb/eval.c
>> +++ b/gdb/eval.c
>> @@ -1902,6 +1902,17 @@ evaluate_subexp_standard (struct type
>> *expect_type,
>>
>> /* First determine the type code we are dealing with. */
>> arg1 = evaluate_subexp (NULL_TYPE, exp, pos, noside);
>> +
>> + /* Allow the short-circuiting of array accesses, function calls
>> + and substring operations in logical expressions. */
>> + if (noside == EVAL_SKIP)
>> + {
>> + /* Skip all the array subscripts. */
>> + for (int i = 0; i < nargs; ++i)
>> + evaluate_subexp (NULL_TYPE, exp, pos, noside);
>> + return eval_skip_value (exp);
>> + }
>> +
>> type = check_typedef (value_type (arg1));
>> code = TYPE_CODE (type);
>>
>> @@ -1952,8 +1963,6 @@ evaluate_subexp_standard (struct type
>> *expect_type,
>> for (; tem <= nargs; tem++)
>> argvec[tem] = evaluate_subexp_with_coercion (exp, pos, noside);
>> argvec[tem] = 0; /* signal end of arglist */
>> - if (noside == EVAL_SKIP)
>> - return eval_skip_value (exp);
>
> Is this change needed? Your test passes even when leaving that part
> as-is.
Hi Simon,
Many thanks for the review.
To clarify, do you observe the test passing with no changes to eval.c or
just when the additions are applied?
If the former is the case, may I ask which compiler are you using? I
have retested with 953473375500 and I see 13 tests fail without the
additions. If it's the latter, the deletions are not strictly needed but
my analysis determined that this code was unreachable after this patch.
My analysis consisted of checking for regressions in the test suite (of
which there were none) and examining all uses of noside from the new
early termination to eval_call in TYPE_CODE_FUNC. There are no
assignments and it's passed to all functions by value.
I can separate this removal into a second patch if this is desirable.
Rich
>
> Thanks,
>
> Simon
>
On 2018-08-06 12:35, Richard Bunt wrote:
> Hi Simon,
>
> Many thanks for the review.
>
> To clarify, do you observe the test passing with no changes to eval.c
> or
> just when the additions are applied?
I mean that I saw the test passing only with the first hunk applied (the
one that adds code).
> If the former is the case, may I ask which compiler are you using? I
> have retested with 953473375500 and I see 13 tests fail without the
> additions. If it's the latter, the deletions are not strictly needed
> but
> my analysis determined that this code was unreachable after this patch.
>
> My analysis consisted of checking for regressions in the test suite (of
> which there were none) and examining all uses of noside from the new
> early termination to eval_call in TYPE_CODE_FUNC. There are no
> assignments and it's passed to all functions by value.
Ah indeed, I read it wrong the first time. I thought that the second
hunk was in a totally different case of the main switch (handling of a
different OP_*. But now I see that this is all under the
"OP_F77_UNDETERMINED_ARGLIST" case, and that we'll never reach the
bottom part with "noside == EVAL_SKIP".
It LGTM then.
Simon
>>>>> "Richard" == Richard Bunt <richard.bunt@arm.com> writes:
Richard> Investigation revealed that EVAL_SKIP was not being handled correctly
Richard> for all types serviced by the OP_F77_UNDETERMINED_ARGLIST case in
Richard> evaluate_subexp_standard. While skipping an undetermined argument
Richard> list the type is resolved to be an integer (as this is what
Richard> evaluate_subexp returns when skipping) and so it was not possible to
Richard> delegate to the appropriate case (e.g. array, function call).
While I agree with Simon that this patch is fine, I think the intended
design of eval.c is that skipped expressions should still try to return
a value of the correct type when possible. The reason for this is that
the type is still sometimes needed, for example to compute the correct
type of a ?: ternary operator, which in turn could be used for overload
resolution.
Richard> While this patch allows a wider range of expressions to be evaluated, it
Richard> should be noted that this patch does not allow the skipping of arrays
Richard> which use Fortran array slicing, due to the inability of the debugger
Richard> to skip OP_RANGE.
This sounds like a bug to me.
Tom
Hi Tom,
Many thanks for the feedback.
On 08/06/2018 08:06 PM, Tom Tromey wrote:
>>>>>> "Richard" == Richard Bunt <richard.bunt@arm.com> writes:
>
> Richard> Investigation revealed that EVAL_SKIP was not being handled correctly
> Richard> for all types serviced by the OP_F77_UNDETERMINED_ARGLIST case in
> Richard> evaluate_subexp_standard. While skipping an undetermined argument
> Richard> list the type is resolved to be an integer (as this is what
> Richard> evaluate_subexp returns when skipping) and so it was not possible to
> Richard> delegate to the appropriate case (e.g. array, function call).
>
> While I agree with Simon that this patch is fine, I think the intended
> design of eval.c is that skipped expressions should still try to return
> a value of the correct type when possible. The reason for this is that
> the type is still sometimes needed, for example to compute the correct
> type of a ?: ternary operator, which in turn could be used for overload
> resolution.
This makes sense, I'll develop a new patch taking this design into account.
>
> Richard> While this patch allows a wider range of expressions to be evaluated, it
> Richard> should be noted that this patch does not allow the skipping of arrays
> Richard> which use Fortran array slicing, due to the inability of the debugger
> Richard> to skip OP_RANGE.
>
> This sounds like a bug to me.
GDB fails with a graceful error message on this class of expression e.g.
p .TRUE. .OR. binary_string(1:2)
"GDB does not (yet) know how to evaluate that kind of expression"
So this seemed like a good place to stop to keep the patch as small and
scoped as possible.
To satisfy this review comment, I'll add another patch to extend the
short-circuiting to include arrays with slicing.
>
> Tom
>
Thanks,
Rich
>>>>> "Richard" == Richard Bunt <richard.bunt@arm.com> writes:
Richard> While this patch allows a wider range of expressions to be evaluated, it
Richard> should be noted that this patch does not allow the skipping of arrays
Richard> which use Fortran array slicing, due to the inability of the debugger
Richard> to skip OP_RANGE.
>> This sounds like a bug to me.
Richard> So this seemed like a good place to stop to keep the patch as small and
Richard> scoped as possible.
Makes sense to me.
Richard> To satisfy this review comment, I'll add another patch to extend the
Richard> short-circuiting to include arrays with slicing.
Sorry, I didn't meant to imply this was something you had to do.
In fact your patch was fine with me as-is.
If you're interested in this of course I would welcome a patch :)
Or if you wanted to just file it in bugzilla, that would be fine as well.
thanks,
Tom
On 08/07/2018 06:40 PM, Tom Tromey wrote:
>>>>>> "Richard" == Richard Bunt <richard.bunt@arm.com> writes:
>
> Richard> While this patch allows a wider range of expressions to be evaluated, it
> Richard> should be noted that this patch does not allow the skipping of arrays
> Richard> which use Fortran array slicing, due to the inability of the debugger
> Richard> to skip OP_RANGE.
>
>>> This sounds like a bug to me.
>
> Richard> So this seemed like a good place to stop to keep the patch as small and
> Richard> scoped as possible.
>
> Makes sense to me.
>
> Richard> To satisfy this review comment, I'll add another patch to extend the
> Richard> short-circuiting to include arrays with slicing.
>
> Sorry, I didn't meant to imply this was something you had to do.
> In fact your patch was fine with me as-is.
>
> If you're interested in this of course I would welcome a patch :)
> Or if you wanted to just file it in bugzilla, that would be fine as well.
>
I am interested in debugging Fortran codes so I will fix this while I am
here :). Fortunately, the re-designed fix which returns type information
even when under EVAL_SKIP fixes this issue as a side-effect. I'm just
waiting on the regression sweep to get back to me.
Many thanks,
Rich
> thanks,
> Tom
>
index
9db6e7c69dad9e674f66991e2aee7dd8d66d80c7..2350c7faaf6221a5990cf5264fe7fe93b4f4be4c
100644
@@ -1902,6 +1902,17 @@ evaluate_subexp_standard (struct type *expect_type,
/* First determine the type code we are dealing with. */
arg1 = evaluate_subexp (NULL_TYPE, exp, pos, noside);
+
+ /* Allow the short-circuiting of array accesses, function calls
+ and substring operations in logical expressions. */
+ if (noside == EVAL_SKIP)
+ {
+ /* Skip all the array subscripts. */
+ for (int i = 0; i < nargs; ++i)
+ evaluate_subexp (NULL_TYPE, exp, pos, noside);
+ return eval_skip_value (exp);
+ }
+
type = check_typedef (value_type (arg1));
code = TYPE_CODE (type);
@@ -1952,8 +1963,6 @@ evaluate_subexp_standard (struct type *expect_type,
for (; tem <= nargs; tem++)
argvec[tem] = evaluate_subexp_with_coercion (exp, pos, noside);
argvec[tem] = 0; /* signal end of arglist */
- if (noside == EVAL_SKIP)
- return eval_skip_value (exp);
return eval_call (exp, noside, nargs, argvec, NULL, expect_type);
default:
b/gdb/testsuite/gdb.fortran/short-circuit-argument-list.exp
new file mode 100644
index
0000000000000000000000000000000000000000..294d808d99e6debfde5b802dd34aa17e3a136cd0
@@ -0,0 +1,80 @@
+# 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."
+ gdb_test "p .TRUE. .OR. function_${arg_lower}_arg($trimmed_args)" \
+ "\\\$$decimal = .TRUE."
+ set argument_list "$argument_list .TRUE.,"
+}
+
+# Vary number of components in the expression to skip.
+set expression "p .TRUE."
+foreach_with_prefix expression_components {1 2 3 4} {
+ set expression "$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 evaluation of substring operations in logical expressions.
+gdb_test "p .FALSE. .OR. binary_string(1)" "\\\$$decimal = .FALSE."
+
+# 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())" \
+ " One, return true.\r\n\\\$$decimal = .TRUE."
b/gdb/testsuite/gdb.fortran/short-circuit-argument-list.f90
new file mode 100644
index
0000000000000000000000000000000000000000..28386d029c1db0afa503d77804fc2c4bb6881dad
@@ -0,0 +1,66 @@
+! 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
+
+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
+ end interface
+ logical, dimension (4,2) :: truth_table
+ logical :: a, b, c, d
+ 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)
+ print *, truth_table(:, 1), a, b
+ print *, truth_table(:, 2), c, d
+end program generate_truth_table