Bug#674006: libwvstreams4.6-base 4.6.1-4 ----- wvdial stopped working
Dear Gregor Herrmann, As you were the kind DD who fixed the gcc 4.7 failure to build for libwvstreaams4.6-base, I am emailing you regarding this bug since you might be able to do a QA upload. This bug relates to further problems with libwvstreams4.6-base when building with gcc 4.7 rather than gcc 4.6. The compiler aggressively optimizes causing a failure when trying to use the wvdial package. The fedora project has also found a problem with gcc 4.7 and wvstreams package. The assert error wvdial: utils/wvtask.cc:409: static void WvTaskMan::_stackmaster(): Assertion `magic_number == -0x123678' failed. thrown by wvdial is caused by a failure in libwvstreams4.6-base. In particular the variable magicnumber should be marked volatile to prevent compiler optimizations. I attach the Fedora version of the patch. It may need to be refreshed for use in Debian. The original discussion at Fedora for this bug is found at: http://permalink.gmane.org/gmane.linux.redhat.fedora.extras.cvs/779635 Please note wvdial is a way of dialing a 3G cell phone modem for internet connection. It is thus important for accessing the internet for some users. Would you be able to refresh this patch and upload to Debian? I am unable to do this since I am a lowly uploader for an unrelated packages festival and speech-tools. I have no archive upload privileges to Debian. If you are unable to address this, do you know the correct DD to assist since it appears to me the package is an orphan? with the very best regards, Peter Peter Drysdale (Uploader for festival and speech-tools) magic.patch Description: Binary data
Bug#674006: libwvstreams4.6-base 4.6.1-4 ----- wvdial stopped working
Dear Gregor, Thank you for your email. It forced me to look further into Debian procedures and policy. Between your email and details of Debian policy I think I have found a way forward. To avoid noise to your account I will not copy you to subsequent emails regarding this bug where I can avoid it. I, of course, welcome any feedback if you choose to give it. With the very kindest regards, Peter
Bug#674006: libwvstreams4.6-base 4.6.1-4 ----- wvdial stopped working
The purpose of this email is to set forth the reasons for raising the severity of this bug and describe my intentions. 1) I intend to increase the severity to release critical. The Debian document regarding severity states: critical makes unrelated software on the system (or the whole system) break, or causes serious data loss, or introduces a security hole on systems where you install the package. I encountered this bug when a fellow Debian user did a "apt-get dist-upgrade" from squeeze to wheezy. (The machine has had dist-upgrades going back as far as etch) That machine has internet access only through a 3G modem at the site it is located. After the apt-get dist-upgade the machine was restarted and failed to connect to the internet. This made the machine completely unusable for the purposes of network access. Since no other forms of network access are available at that site the machine would not have been able to receive security updates either. The machine was transported to a wired network site [my place :-) ] to allow downgrading of this package from snapshots since this bug completely prevents network access if wvdial is used as the only 3G dialer. Subsequently network-manager has been installed on this machine to provide additional 3G modem dialer support. (Network-manager did not provide a dialer when 3G access was originally installed on this machine). I believe these facts constitute that this bug made unrelated software (browser, ssh , mail etc.) broken. Since the only network access was via this package - the completely unavailable network seriously compromised this machine. If left it in this state it would also have led to security holes accumulating. On reading of Debian guidelines I conclude this constituted a release critical bug for this machine. 2) Subject to my ability to gain access to hardware which I do not own, I intend to test the fedora patch and make it available on BTS followed by preparing for a NMU to be made available on Debian QA mailing list. The package has been abandoned upstream (homepage links are now non-existent). It is also orphaned in Debian. My intention is only to prevent users upgrading to wheezy getting a "surprise". I urge other users reading this to move to network-manager if possible before a dist-upgrade since network-manager appears better supported. I apologize for my intention not to assist in subsequent maintaining of this package after working on this bug. My decision is significantly influenced by the pejorative comments made about this software by one of the original authors at the following URL:- http://apenwarr.ca/log/?m=200912 with best regards, Peter
Bug#674006:
Please find attached a draft patch refreshed in the Debian style. This patch has passed build testing only on a Debian system. It has passed testing on Fedora. Run-time testing this patch on Debian will have to await either till I get access to the hardware effected or another person affected by this bug test compiles on their hardware, it works and they notify me so I can hustle for a NMU. best regards, Peter 07_magic_number.diff Description: Binary data
Bug#674006: wvdial further discussion
Dear Chris, Thanks for your testing and feedback. Apologies about the patch being insufficient. I have not had access to the affected hardware since my previous post. I note work continues at Fedora on this bug which can be found at: https://bugzilla.redhat.com/show_bug.cgi?id=812651 Instead of the insufficient patch a possible test which could be tried is altering line 37 of the debian/rules file to include CFLAGS += -O2 -fno-tree-dce -fno-optimize-sibling-calls This would follow from the information suggested by Fabrice Bellet on the Fedora bug list. This is both a bit of a hack and shot in the dark since I don't own the hardware to test. I might get access to the hardware this weekend though. best regards, Peter
Bug#674006: runtime tested solution found
Please find attached a new debian/rules file for wvstreams which appears to fix this bug. It is altered in two places to switch off some compiler optimizations. It is obviously a hack. This has been runtime tested on borrowed hardware. This follows Fabrice's suggestions. Proper diff and packaging will follow. It would be great if someone else also tests this in runtime context. very best regards, Peter rules Description: Binary data
Bug#674006: bug #674006 wvstreams diff
Attached is diff for debian/rules for wvstreams to fix this bug. It has been runtime tested on two different machines. Wvstreams has received an ITA (intent to adopt) since I first started looking into this bug. I presume the new maintainer (Bart Martens) will upload for this bug in combination with his adoption upload so I leave this bug in patch available state. best regards, Peter rules.diff Description: Binary data
