[PATCH] Fix GDBHISTSIZE test failure on i686

[PATCH] Fix GDBHISTSIZE test failure on i686

[Patrick Palka]

The test

  test_histsize_history_setting "99999999999999999999999999999999999" "unlimited"

was failing on i686 because the condition in init_history() for
determining whether to map a large GDBHISTSIZE value to infinity was

  long var = strtol (tmpenv);
  if (var > INT_MAX)
    history_size = unlimited;

but this condition is never true on i686 because INT_MAX == LONG_MAX.
So in order to properly map large out-of-range values of GDBHISTSIZE to
infinity on targets where LONG_MAX > INT_MAX as well as on i686, we have
to instead change the above condition to

  if (var > INT_MAX
      || (var == INT_MAX && errno == ERANGE))
    history_size = unlimited;

[ I did not test this patch on i686 because I don't have access to
  such a machine.  But the patch seems straightforward enough... ]

gdb/ChangeLog:

	* top.c (init_history): Look at the errno set by strtol to
	properly map large GDBHISTSIZE values to infinity.
---
 gdb/top.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/gdb/top.c b/gdb/top.c
index 5114c2e..afb14be 100644
--- a/gdb/top.c
+++ b/gdb/top.c
@@ -1734,10 +1734,12 @@ init_history (void)
   if (tmpenv)
     {
       long var;
+      int saved_errno;
       char *endptr;
 
       tmpenv = skip_spaces (tmpenv);
       var = strtol (tmpenv, &endptr, 10);
+      saved_errno = errno;
       endptr = skip_spaces (endptr);
 
       /* If GDBHISTSIZE is non-numeric then ignore it.  If GDBHISTSIZE is the
@@ -1749,7 +1751,12 @@ init_history (void)
 	;
       else if (*tmpenv == '\0'
 	       || var < 0
-	       || var > INT_MAX)
+	       || var > INT_MAX
+	       /* On targets where INT_MAX == LONG_MAX, we have to look at
+		  the errno set by strtol to distinguish between a value that
+		  is exactly INT_MAX and an overflowing value that was clamped
+		  to INT_MAX.  */
+	       || (var == INT_MAX && saved_errno == ERANGE))
 	history_size_setshow_var = -1;
       else
 	history_size_setshow_var = var;
-- 
2.4.4.410.g43ed522.dirty

Re: [PATCH] Fix GDBHISTSIZE test failure on i686

[Patrick Palka]

On Tue, Jun 23, 2015 at 10:30 AM, Patrick Palka <patrick@parcs.ath.cx> wrote:
> The test
>
>   test_histsize_history_setting "99999999999999999999999999999999999" "unlimited"
>
> was failing on i686 because the condition in init_history() for
> determining whether to map a large GDBHISTSIZE value to infinity was
>
>   long var = strtol (tmpenv);
>   if (var > INT_MAX)
>     history_size = unlimited;
>
> but this condition is never true on i686 because INT_MAX == LONG_MAX.
> So in order to properly map large out-of-range values of GDBHISTSIZE to
> infinity on targets where LONG_MAX > INT_MAX as well as on i686, we have
> to instead change the above condition to
>
>   if (var > INT_MAX
>       || (var == INT_MAX && errno == ERANGE))
>     history_size = unlimited;
>
> [ I did not test this patch on i686 because I don't have access to
>   such a machine.  But the patch seems straightforward enough... ]
>
> gdb/ChangeLog:
>
>         * top.c (init_history): Look at the errno set by strtol to
>         properly map large GDBHISTSIZE values to infinity.
> ---
>  gdb/top.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/gdb/top.c b/gdb/top.c
> index 5114c2e..afb14be 100644
> --- a/gdb/top.c
> +++ b/gdb/top.c
> @@ -1734,10 +1734,12 @@ init_history (void)
>    if (tmpenv)
>      {
>        long var;
> +      int saved_errno;
>        char *endptr;
>
>        tmpenv = skip_spaces (tmpenv);
>        var = strtol (tmpenv, &endptr, 10);
> +      saved_errno = errno;

errno should be set to 0 before the call to strtol.

>        endptr = skip_spaces (endptr);
>
>        /* If GDBHISTSIZE is non-numeric then ignore it.  If GDBHISTSIZE is the
> @@ -1749,7 +1751,12 @@ init_history (void)
>         ;
>        else if (*tmpenv == '\0'
>                || var < 0
> -              || var > INT_MAX)
> +              || var > INT_MAX
> +              /* On targets where INT_MAX == LONG_MAX, we have to look at
> +                 the errno set by strtol to distinguish between a value that
> +                 is exactly INT_MAX and an overflowing value that was clamped
> +                 to INT_MAX.  */
> +              || (var == INT_MAX && saved_errno == ERANGE))
>         history_size_setshow_var = -1;
>        else
>         history_size_setshow_var = var;
> --
> 2.4.4.410.g43ed522.dirty
>

Re: [PATCH] Fix GDBHISTSIZE test failure on i686

[Pedro Alves]

On 06/23/2015 03:30 PM, Patrick Palka wrote:
> The test
> 
>   test_histsize_history_setting "99999999999999999999999999999999999" "unlimited"
> 
> was failing on i686 because the condition in init_history() for
> determining whether to map a large GDBHISTSIZE value to infinity was
> 
>   long var = strtol (tmpenv);
>   if (var > INT_MAX)
>     history_size = unlimited;
> 
> but this condition is never true on i686 because INT_MAX == LONG_MAX.
> So in order to properly map large out-of-range values of GDBHISTSIZE to
> infinity on targets where LONG_MAX > INT_MAX as well as on i686, we have
> to instead change the above condition to
> 
>   if (var > INT_MAX
>       || (var == INT_MAX && errno == ERANGE))
>     history_size = unlimited;
> 
> [ I did not test this patch on i686 because I don't have access to
>   such a machine.  But the patch seems straightforward enough... ]

Looks fine to me, with the missing errno=0, but note that assuming you
install the 32-bit dependencies in your distro, you can easily build
a 32-bit gdb on a 64-bit host.  E.g., on x86-64 GNU/Linux, configure with:

CC="gcc -m32" /path/to/configure \
    --host=i686-pc-linux-gnu \
    --build=i686-pc-linux-gnu \
    --target=i686-pc-linux-gnu

Thanks,
Pedro Alves

Re: [PATCH] Fix GDBHISTSIZE test failure on i686

[Doug Evans]

On Tue, Jun 23, 2015 at 12:19 PM, Pedro Alves <palves@redhat.com> wrote:
>... note that assuming you
> install the 32-bit dependencies in your distro, you can easily build
> a 32-bit gdb on a 64-bit host.  E.g., on x86-64 GNU/Linux, configure with:
>
> CC="gcc -m32" /path/to/configure \
>     --host=i686-pc-linux-gnu \
>     --build=i686-pc-linux-gnu \
>     --target=i686-pc-linux-gnu

--host and --target can be elided here.

Re: [PATCH] Fix GDBHISTSIZE test failure on i686

[Patrick Palka]

On Tue, Jun 23, 2015 at 1:19 PM, Pedro Alves <palves@redhat.com> wrote:
> On 06/23/2015 03:30 PM, Patrick Palka wrote:
>> The test
>>
>>   test_histsize_history_setting "99999999999999999999999999999999999" "unlimited"
>>
>> was failing on i686 because the condition in init_history() for
>> determining whether to map a large GDBHISTSIZE value to infinity was
>>
>>   long var = strtol (tmpenv);
>>   if (var > INT_MAX)
>>     history_size = unlimited;
>>
>> but this condition is never true on i686 because INT_MAX == LONG_MAX.
>> So in order to properly map large out-of-range values of GDBHISTSIZE to
>> infinity on targets where LONG_MAX > INT_MAX as well as on i686, we have
>> to instead change the above condition to
>>
>>   if (var > INT_MAX
>>       || (var == INT_MAX && errno == ERANGE))
>>     history_size = unlimited;
>>
>> [ I did not test this patch on i686 because I don't have access to
>>   such a machine.  But the patch seems straightforward enough... ]
>
> Looks fine to me, with the missing errno=0, but note that assuming you
> install the 32-bit dependencies in your distro, you can easily build
> a 32-bit gdb on a 64-bit host.  E.g., on x86-64 GNU/Linux, configure with:
>
> CC="gcc -m32" /path/to/configure \
>     --host=i686-pc-linux-gnu \
>     --build=i686-pc-linux-gnu \
>     --target=i686-pc-linux-gnu

Cool.  I will commit the patch after confirming that it's fixed on
i686 this way.

>
> Thanks,
> Pedro Alves
>