Arx Libertatis Bug Tracker ~ [CLOSED] #245 - [1CE529DF] __kernel

Arx Libertatis Bug Tracker

CLOSED Crash report #245 - [1CE529DF] __kernel_vsyscall ()

Posted Apr 24, 2012 - updated Jul 05, 2013 Shortlink: http://arx.vg/245

Votes

This issue has been closed with status "Fixed" and resolution "RESOLVED".

Issue details

Type of issue

Crash report Unknown issue type
Status

Fixed

Status not determined
Assigned to

Not assigned to anyone
Progress
Type of bug

Not triaged
Likelihood

Not triaged
Effect

Not triaged
Posted by

CrashBot

CrashBot
Offline

This user was last seen online at Nov 02, 2026 - 03:06

Show user details

Close this menu
Owned by

Not owned by anyone
Estimated time

No time Not estimated
Time spent

2 hours No time spent
Category

Not determined
Resolution

RESOLVED Not determined
Priority

Not determined
Reproducability

Not determined
Severity

Not determined
Targetted for

Not determined
OS

Not determined
Architecture

Not determined
Fixed in

Arx Libertatis 1.0.2Not determined

Issue description

Abnormal termination

GDB stack trace:

Steps to reproduce this issue

Compile debug version, complete demo and save in the last section, where goblin asks for a license. Compile release version, load save and place cursor on torch item in the inventory.

Show system-generated comments

Comment posted by

CrashBot

Apr 24, 15:29

A file was uploaded. arx.log

This comment was attached:

arxcrash-10860-2816373975

Comment posted by

CrashBot

Apr 24, 15:29

A file was uploaded. cfg.ini

This comment was attached:

arxcrash-10860-2816373975

Comment posted by

CrashBot

Apr 24, 15:29

A file was uploaded. crash.xml

This comment was attached:

arxcrash-10860-2816373975

Comment posted by

Daniel Scharrer

Apr 24, 15:42

If this is reproducible, could you please attach the save that causes this?

Does it still happen if you switch back to the debug version?

Comment posted by

Guest user

Apr 24, 15:53

Output is:

terminate called after throwing an instance of 'std::length_error'
what(): basic_string::_S_create

It crashes the same way if I point cursor on the first bone lying in first stage.

Does it still happen if you switch back to the debug version?

No. Only release version is affected.

Comment posted by

Guest user

Apr 24, 17:22

Works fine when building without unity-build.

Comment posted on behalf of

Daniel Scharrer

May 24, 23:15

This issue has been updated with the latest changes from the code repository.

Comment posted by

Daniel Scharrer

May 24, 23:17

I believe this is now fixed in the latest git version and the unity build should not be a problem anymore. The fix will be part of the next patch release (1.0.2)

The issue was updated with the following change(s):

This issue has been closed
The status has been updated, from New to Fixed.
This issue's progression has been updated to 100 percent completed.
The resolution has been updated, from Not determined to RESOLVED.
Time spent on this issue, from No time spent to 2 hours.

Comment posted on behalf of

Daniel Scharrer

Jun 30, 05:17

This issue has been updated with the latest changes from the code repository.

#10

Comment posted on behalf of

Daniel Scharrer

Jul 05, 19:44

The issue was updated with the following change(s):

The custom field Fixed in has been updated, from Unknown to Arx Libertatis 1.0.2.

64c2352

master Revision 64c2352cda48e90934aad41c5de74624ef0b5177 by

Daniel Scharrer

May 24, 22:01 - Preceeded by c979ef0861ba07c598f9946c1e90488c2ad5fc7a

Show more details

Log entry

Don't use the -fwhole-program compiler flag

-fwhole-program can be used to tell the gcc that everything that
makes up the final program is included in the current command, which
allows it to use more aggressive optimizations. This can result in
smaller and faster code.

However, cmake always uses at least two commands to compile each
executable, even with the unity build: one for building the user code
and one for linking all libraries.

Using the -fwhole-program option here can cause the compiler to not
properly internalize some C++ objects (such as typeids) that are used
in both the user code and external libraries, resulting in unexpected
behavior.

From my tests, there is no measurable performance gain and the file size
reduction for the arx executable is 'only' around 18%. I think the
potential problems outweigh those benefits, so -fwhole-program must go.

This only affects unity builds as -fwhole-program was only used for
those. Unity builds are still possible and recommended.

In the future we can look into using link-time optimizations, but those
are not stable enough in current gcc versions to be enabled by default.
Also, quick tests show no real benefits.

See crash report #245
See also commit 58aa3 and crash report #243

64c2352

renderer_stack Revision 64c2352cda48e90934aad41c5de74624ef0b5177 by

Daniel Scharrer

May 24, 22:01 - Preceeded by c979ef0861ba07c598f9946c1e90488c2ad5fc7a

Show more details

Log entry

Don't use the -fwhole-program compiler flag

-fwhole-program can be used to tell the gcc that everything that
makes up the final program is included in the current command, which
allows it to use more aggressive optimizations. This can result in
smaller and faster code.

However, cmake always uses at least two commands to compile each
executable, even with the unity build: one for building the user code
and one for linking all libraries.

Using the -fwhole-program option here can cause the compiler to not
properly internalize some C++ objects (such as typeids) that are used
in both the user code and external libraries, resulting in unexpected
behavior.

From my tests, there is no measurable performance gain and the file size
reduction for the arx executable is 'only' around 18%. I think the
potential problems outweigh those benefits, so -fwhole-program must go.

This only affects unity builds as -fwhole-program was only used for
those. Unity builds are still possible and recommended.

In the future we can look into using link-time optimizations, but those
are not stable enough in current gcc versions to be enabled by default.
Also, quick tests show no real benefits.

See crash report #245
See also commit 58aa3 and crash report #243

* Username
* Display name
The "display name" is the name shown to others
Real name
* E-mail address
* Confirm e-mail

Welcome to The Bug Genie

Really delete this comment?

Really delete this comment?

Really delete this comment?

Really delete this comment?

Really delete this comment?

Really delete this comment?

Really delete this comment?

Really delete this comment?

Really delete this comment?

Really delete this comment?

Username
Password