Dear mips porters,

I am maintaining the gnuplot in the Debian and there is a segfault on the mipsel, what I reported to upstream [1].

There is a question from the gnuplot author, which I am not able to response. Could you please have a look?

It segfaults somewhere in the static global structure. And the problem appeared recently, maybe due to the GCC-9 migration. There were not problems with GCC-8. Some regressions?

[1] https://sourceforge.net/p/gnuplot/bugs/2195/

Thank you

Anton

Am Di., 3. Sept. 2019 um 21:27 Uhr schrieb Ethan Merritt <sfeam@users.sourceforge.net>:

This trace does not make sense to me.
The code lines highlighted are:

} else if (r >= R_AXIS.min) {
/* We store internally as if plotting r(theta) - rmin */
r = r - R_AXIS.min;

R_AXIS is a static global structure that has been accessed repeated in the immediately preceding code, so I don't see any way it could cause a segfault here. r is a parameter to this routine so I would expect it to be on the stack (but I don't know

mipsel compiler calling conventions so maybe not). Could the compiler somehow be setting a read-only flag that prevents modifying a local parameter?

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On 2019-09-03 21:59, Anton Gladky wrote:

Dear mips porters,

I am maintaining the gnuplot in the Debian and there is a segfault on the mipsel, what I reported to upstream [1].

There is a question from the gnuplot author, which I am not able to response. Could you please have a look?

It segfaults somewhere in the static global structure. And the problem appeared
recently, maybe due to the GCC-9 migration. There were not problems with GCC-8.
Some regressions?

I confirm it's a regression due introduced by GCC 9. I have reported the
bug upstream: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91769


Until this bug is fixed, you can use the following workaround:

--- gnuplot-5.2.6+dfsg1.orig/src/axis.c
+++ gnuplot-5.2.6+dfsg1/src/axis.c
@@ -2698,6 +2698,9 @@ map_y(double value)
* If update is TRUE then check and update rrange autoscaling
*/
coord_type
+#ifdef __mips
+__attribute__((optimize("-O1")))
+#endif
polar_to_xy( double theta, double r, double *x, double *y, TBOOLEAN update)
{
coord_type status = INRANGE;

--
Aurelien Jarno GPG: 4096R/1DDD8C9B aurelien@aurel32.net http://www.aurel32.net

--
Aurelien Jarno GPG: 4096R/1DDD8C9B aurelien@aurel32.net http://www.aurel32.net

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hello Aurelien,

thanks for the note. It really helped!

@Ethan, could you please close the issue in gnuplot-tracker? It should
be fixed in GCC.

Regards

Anton

Am Sa., 14. Sept. 2019 um 18:38 Uhr schrieb Aurelien Jarno <aurelien@aurel32.net>:

On 2019-09-03 21:59, Anton Gladky wrote:

Dear mips porters,

I am maintaining the gnuplot in the Debian and there is a segfault on the mipsel, what I reported to upstream [1].

There is a question from the gnuplot author, which I am not able to response.
Could you please have a look?

It segfaults somewhere in the static global structure. And the problem appeared
recently, maybe due to the GCC-9 migration. There were not problems with GCC-8.
Some regressions?

I confirm it's a regression due introduced by GCC 9. I have reported the
bug upstream: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91769

Until this bug is fixed, you can use the following workaround:

--- gnuplot-5.2.6+dfsg1.orig/src/axis.c
+++ gnuplot-5.2.6+dfsg1/src/axis.c
@@ -2698,6 +2698,9 @@ map_y(double value)
* If update is TRUE then check and update rrange autoscaling
*/
coord_type
+#ifdef __mips
+__attribute__((optimize("-O1")))
+#endif
polar_to_xy( double theta, double r, double *x, double *y, TBOOLEAN update)
{
coord_type status = INRANGE;

--
Aurelien Jarno GPG: 4096R/1DDD8C9B aurelien@aurel32.net http://www.aurel32.net

--
Aurelien Jarno GPG: 4096R/1DDD8C9B aurelien@aurel32.net http://www.aurel32.net

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hello Aurelien,

thanks for this information. I have dropped the patch.

Regards

Anton

Am Do., 17. Okt. 2019 um 20:22 Uhr schrieb Aurelien Jarno <aurelien@aurel32.net>:

Hi Anton,

On 2019-09-15 13:01, Anton Gladky wrote:

Hello Aurelien,

thanks for the note. It really helped!

Please note that the problem has now been fixed in the Debian gcc-9
package, and this fixed version is installed on the buildds. Therefore
you should be able to drop the workaround in next upload.

Aurelien

--
Aurelien Jarno GPG: 4096R/1DDD8C9B aurelien@aurel32.net http://www.aurel32.net

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi Anton,

On 2019-09-15 13:01, Anton Gladky wrote:

Hello Aurelien,

thanks for the note. It really helped!

Please note that the problem has now been fixed in the Debian gcc-9
package, and this fixed version is installed on the buildds. Therefore
you should be able to drop the workaround in next upload.

Aurelien

--
Aurelien Jarno GPG: 4096R/1DDD8C9B aurelien@aurel32.net http://www.aurel32.net

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)