This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH] record_latest_value: Call release_value_or_incref instead of release_value
- From: Doug Evans <xdje42 at gmail dot com>
- To: Ludovic CourtÃs <ludo at gnu dot org>
- Cc: "gdb-patches\ at sourceware dot org" <gdb-patches at sourceware dot org>
- Date: Thu, 13 Mar 2014 09:33:51 -0700
- Subject: Re: [PATCH] record_latest_value: Call release_value_or_incref instead of release_value
- Authentication-results: sourceware.org; auth=none
- References: <m3zjl7u8ik dot fsf at sspiff dot org> <87wqgbsktg dot fsf at gnu dot org> <CAP9bCMQFR5DHAZmyN8MmJe=8uRLWuQAXEO_+nw4W1KaoJTy1jw at mail dot gmail dot com>
Doug Evans <xdje42@gmail.com> writes:
> On Mon, Mar 3, 2014 at 1:17 AM, Ludovic CourtÃs <ludo@gnu.org> wrote:
>> Doug Evans <xdje42@gmail.com> skribis:
>>
>>> I think this is because its reference counting is wrong.
>>> Upon return from record_latest_value, its reference count is still one.
>>> However it was one upon entry. It should be two, right?
>>> One for the Scheme wrapper and one for the history entry.
>>>
>>> (gdb) guile (define histnum (history-append! (make-value 42)))
>>
>> Indeed, good catch. Here's how I reproduced it:
>>
>> --8<---------------cut here---------------start------------->8---
>> (gdb) guile (use-modules (gdb))
>> (gdb) guile (history-append! (make-value 42))
>> 1
>> (gdb) p $1
>> $2 = 42
>> (gdb) guile (gc)
>> (gdb) p $1
>> Segmentation fault
>> --8<---------------cut here---------------end--------------->8---
>>
>> What about adding this to the patch as a test case?
>
> Shall do. Thanks.
Here's what I committed.
2014-03-13 Doug Evans <xdje42@gmail.com>
* value.c (record_latest_value): Call release_value_or_incref
instead of release_value.
testsuite/
2014-03-13 Ludovic CourtÃs <ludo@gnu.org>
Doug Evans <xdje42@gmail.com>
* gdb.guile/scm-value.exp (test_value_in_inferior): Verify value added
to history survives a gc.
diff --git a/gdb/value.c b/gdb/value.c
index 4e8d1fe..27043ee 100644
--- a/gdb/value.c
+++ b/gdb/value.c
@@ -1659,7 +1659,11 @@ record_latest_value (struct value *val)
from. This is a bit dubious, because then *&$1 does not just return $1
but the current contents of that location. c'est la vie... */
val->modifiable = 0;
- release_value (val);
+
+ /* The value may have already been released, in which case we're adding a
+ new reference for its entry in the history. That is why we call
+ release_value_or_incref here instead of release_value. */
+ release_value_or_incref (val);
/* Here we treat value_history_count as origin-zero
and applying to the value being stored now. */
diff --git a/gdb/testsuite/gdb.guile/scm-value.exp b/gdb/testsuite/gdb.guile/scm-value.exp
index 89f0ff1..a85d5bd 100644
--- a/gdb/testsuite/gdb.guile/scm-value.exp
+++ b/gdb/testsuite/gdb.guile/scm-value.exp
@@ -67,6 +67,10 @@ proc test_value_in_inferior {} {
gdb_test "gu (history-ref i)" "#<gdb:value 42>"
gdb_test "p \$" "= 42"
+ # Verify the recorded history value survives a gc.
+ gdb_test_no_output "guile (gc)"
+ gdb_test "p \$\$" "= 42"
+
# Test dereferencing the argv pointer.
# Just get inferior variable argv the value history, available to guile.