Bug #20946 for Sys-Syslog: Sys::Syslog problems

Fri Aug 11 10:26:44 2006 t.d.lee [...] durham.ac.uk - Ticket created

CC:	Julian Field <MailScanner [...] ECS.SOTON.AC.UK>
Subject:	Sys::Syslog problems
Date:	Fri, 11 Aug 2006 15:26:13 +0100 (BST)
To:	bug-sys-syslog [...] rt.cpan.org
From:	David Lee <t.d.lee [...] durham.ac.uk>

Bug report for Sys::Syslog 0.17. Disclaimer: I'm merely piggy-in-the-middle, trying to get a bug recognised and fixed. I'm not actually subject to the bug myself. The MailScanner project (www.mailscanner.info) has for many years used Sys::Syslog extensively. In early July, a few MailScanner folk noticed problems with 0.16, which were apparently OK if they backtracked to 1.15. See: http://lists.mailscanner.info/pipermail/mailscanner-beta/2006-July/ thread: Sys::Syslog 0.16 problems Then later in July 0.17 appeared, and apparently seemed OK for some folk. (So I guess that the new 0.16 problems were quickly recognised and fixed: thank you!) See same URL, thread "Sys::Syslog 0.17". But note the reports about "make test" failing in that thread. Since then (during August) there have apparently been more failures (in "make test", I understand). Are you aware of any problems in 0.17, particularly in "make test"? Are there any plans to resolve this? If you need more detail, the best authority is Julian Field, the developer of MailScanner, to whom this is Cc:d. (I could attempt to handle queries about this issue if necessary if you wish.) -- : David Lee I.T. Service : : Senior Systems Programmer Computer Centre : : Durham University : : http://www.dur.ac.uk/t.d.lee/ South Road : : Durham DH1 3LE : : Phone: +44 191 334 2752 U.K. :

Fri Aug 11 11:26:05 2006 SAPER [...] cpan.org - Requestor mailscanner@ecs.soton.ac.uk added

Fri Aug 11 12:40:49 2006 SAPER [...] cpan.org - Correspondence added

Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Fri, 11 Aug 2006 18:40:14 +0200
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Sébastien Aperghis-Tramoni <saper [...] cpan.org>

Show quoted text

> Disclaimer: I'm merely piggy-in-the-middle, trying to get a bug > recognised and fixed. I'm not actually subject to the bug myself. > > The MailScanner project (www.mailscanner.info) has for many years used > Sys::Syslog extensively. In early July, a few MailScanner folk noticed > problems with 0.16, which were apparently OK if they backtracked to 1.15.

Apologises for creating this problem. Show quoted text

> Then later in July 0.17 appeared, and apparently seemed OK for some folk. > (So I guess that the new 0.16 problems were quickly recognised and fixed: > thank you!) See same URL, thread "Sys::Syslog 0.17".

A bug was reported but I admit I didn't see the urgency of it until someone opened another ticket, with the precision that the bug was breaking Spamassassin (and after I released 0.17, someone also reported problems in RT). Had I been told that it broke Spamassassin, I would have certainly tried to fix it earlier. Show quoted text

> But note the reports about "make test" failing in that thread.

Are you referring to this mail[1]? There is no failure here, only skipped tests, which is perfectly normal. The tests are skipped because some test specific modules are not installed, but as they are not needed to check the core functionality of the module, they're not prerequisites. [1]http://lists.mailscanner.info/pipermail/mailscanner-beta/2006-July/000246.html Show quoted text

> Since then (during August) there have apparently been more failures (in > "make test", I understand). > > Are you aware of any problems in 0.17, particularly in "make test"? > Are there any plans to resolve this?

If you or other people have failure cases, don't hesitate to fill a bug with the full output, and also tell me what software is broken by the bug. Currently, as reported by CPAN Testers, Sys::Syslog 0.17 seems to pass all its tests in a few systems: » http://cpantesters.perl.org/show/Sys-Syslog.html#Sys-Syslog-0.17 And there is no recent failure caused by Sys::Syslog reported by Perl smokers, which cover a larger set of operating systems: » http://www.nntp.perl.org/group/perl.daily-build.reports/ Someone has also confirmed the existence of a heisenbug related to the UDP mechanism on Darwin. I'll give it a look this week-end. There's also the ticket #20635 about tests failure on Cygwin: » http://rt.cpan.org/Ticket/Display.html?id=20635 Of course, one of the problem is that the test suite only checks the API, but has currently no way to check that the data are correctly sent or received. I've begun working on a data validation test script, but have not yet finished it. Show quoted text

> If you need more detail, the best authority is Julian Field, the developer > of MailScanner, to whom this is Cc:d. (I could attempt to handle queries > about this issue if necessary if you wish.)

I've added Julian in the requestors of this ticket so he should automatically receive updates to this ticket. I haven't look at the source of MailScanner yet, but one of the problems with Sys::Syslog is that it didn't provide the "native" mechanism until very recently (version 0.15) and unfortunately kind of recommended the "unix" mechanism even though it is quite fragile. The native mechanism, as it uses the real C functions, is what the module should have provided since day one as it make Perl behaves like any other program. Therefore, I'd now suggest to avoid changing the mechanism with setlogsock() unless necessary. I'll modify the documentation to make that statement more clear. -- Sébastien Aperghis-Tramoni Close the world, txEn eht nepO.

Fri Aug 11 12:40:49 2006 The RT System itself - Status changed from 'new' to 'open'

Fri Aug 11 13:18:50 2006 t.d.lee [...] durham.ac.uk - Correspondence added

CC:	mailscanner [...] ecs.soton.ac.uk
Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Fri, 11 Aug 2006 18:18:22 +0100 (BST)
To:	Sébastien Aperghis-Tramoni via RT <bug-Sys-Syslog [...] rt.cpan.org>
From:	David Lee <t.d.lee [...] durham.ac.uk>

On Fri, 11 Aug 2006, [UTF-8] SÃ©bastien Aperghis-Tramoni via RT wrote: Show quoted text

> <URL: http://rt.cpan.org/Ticket/Display.html?id=20946 > > > [...]

> > But note the reports about "make test" failing in that thread.

> > Are you referring to this mail[1]? There is no failure here, only skipped > tests, which is perfectly normal. The tests are skipped because some > test specific modules are not installed, but as they are not needed > to check the core functionality of the module, they're not prerequisites. > > [1]http://lists.mailscanner.info/pipermail/mailscanner-beta/2006-July/000246.html

Many thanks for your swift reply! Much appreciated. Sadly, I'm just the middle-man, with no first-hand knowledge of the problem. I had spotted a criticism of the recent module development on the main MailScanner list, because of claimed failures in recent revisions, including (apparently) a residual "make test" hanging problem in 0.17. I simply wanted to check that the developer (you, Sebastian) were either aware of the problem or willing to accept fault reports. (It seems that the problem is new to you (fair enough), and that you are open to reports (excellent!). Show quoted text

> > Since then (during August) there have apparently been more failures (in > > "make test", I understand). > > > > Are you aware of any problems in 0.17, particularly in "make test"? > > Are there any plans to resolve this?

> > If you or other people have failure cases, don't hesitate to fill a bug > with the full output, and also tell me what software is broken by the > bug. [...]

Thanks. Earlier this afternoon (UK time) I asked (on the main MailScanner email list) for a more detailed report of the alleged problem. It apparently claims a hang (presumably very, very long wait) during the "make test" on some platforms. No report forthcoming as yet. I've also searched the archives of the main list, but was unable to track down a detailed report. Sorry. Show quoted text

> I've added Julian in the requestors of this ticket so he should automatically > receive updates to this ticket.

Thanks. Julian actively reads his email! The chances are that he can dig out the exact report of the alleged "make test" hang and send it to you. The responsibility now fairly and squarely lies with us (the MailScanner community) to provide you (Sebastian) with decent information. Best wishes! Have a good weekend. (Time for me to go home...) -- : David Lee I.T. Service : : Senior Systems Programmer Computer Centre : : Durham University : : http://www.dur.ac.uk/t.d.lee/ South Road : : Durham DH1 3LE : : Phone: +44 191 334 2752 U.K. :

Fri Aug 11 17:12:26 2006 SAPER [...] cpan.org - Correspondence added

Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Fri, 11 Aug 2006 23:12:08 +0200
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Sébastien Aperghis-Tramoni <saper [...] cpan.org>

Quoting David Lee via RT <bug-Sys-Syslog@rt.cpan.org>: Show quoted text

> Many thanks for your swift reply! Much appreciated.

You're welcome :) Show quoted text

> Sadly, I'm just the middle-man, with no first-hand knowledge of the > problem.

No problem. People can contribute to this ticket from the RT interface. Show quoted text

> I had spotted a criticism of the recent module development on the main > MailScanner list, because of claimed failures in recent revisions, > including (apparently) a residual "make test" hanging problem in 0.17. > > I simply wanted to check that the developer (you, Sebastian) were either > aware of the problem or willing to accept fault reports. (It seems that > the problem is new to you (fair enough), and that you are open to reports > (excellent!).

I'm more the maintainer than the developper. Sys::Syslog is a piece of old code, with roots dating back from Perl 4, and has evolved in a strange and organic way. I became the maintainer last year when it was decided to released it on the CPAN, independently of the main Perl distrbution, in order to address CVE-2005-3962 (the sprintf() buffer overflow problem). Since then, I've tried to address the tickets that were opened and accept the patches that were waiting, also trying to tidy up the code and add new features. Show quoted text

> Earlier this afternoon (UK time) I asked (on the main MailScanner email > list) for a more detailed report of the alleged problem. > > It apparently claims a hang (presumably very, very long wait) during the > "make test" on some platforms.

I've seen this occurs on a Debian system, but have not yet investigated this case. Show quoted text

> No report forthcoming as yet. > > I've also searched the archives of the main list, but was unable to track > down a detailed report. Sorry.

No problem, I can wait. Show quoted text

> Thanks. Julian actively reads his email! The chances are that he can dig > out the exact report of the alleged "make test" hang and send it to you. > > The responsibility now fairly and squarely lies with us (the MailScanner > community) to provide you (Sebastian) with decent information.

I took a quick glance at MailScanner::Log, and I think the code can be simplified by using the features from the recent Sys::Syslog versions. Namely, the eval{} can be avoided by using the "nofatal" option of openlog(). » http://search.cpan.org/dist/Sys-Syslog/Syslog.pm#openlog Also, I really think that you should avoid calling setlogsock() to change the mechanism and just keep with the native one. Especially on systems like Solaris, where Alan Burlison, Solaris developper and the main Perl contact for this system, expressed his clear opinion about this subject and explained that he prefers the native mechanism over the others. Show quoted text

> Best wishes! Have a good weekend. (Time for me to go home...)

Thanks. Although I'll have to spend part of it on the Sys::Syslog problems. -- Sébastien Aperghis-Tramoni Close the world, txEn eht nepO.

Sat Aug 12 09:32:48 2006 mailscanner [...] ecs.soton.ac.uk - Correspondence added

CC:	t.d.lee [...] durham.ac.uk
Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Sat, 12 Aug 2006 14:32:11 +0100
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Julian Field <mailscanner [...] ecs.soton.ac.uk>

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Sébastien Aperghis-Tramoni via RT wrote: Show quoted text

> <URL: http://rt.cpan.org/Ticket/Display.html?id=20946 > > > Quoting David Lee via RT <bug-Sys-Syslog@rt.cpan.org>: >

>> Many thanks for your swift reply! Much appreciated.

> > You're welcome :) >

>> Sadly, I'm just the middle-man, with no first-hand knowledge of the >> problem.

> > No problem. People can contribute to this ticket from the RT interface. >

>> I had spotted a criticism of the recent module development on the main >> MailScanner list, because of claimed failures in recent revisions, >> including (apparently) a residual "make test" hanging problem in 0.17. >> >> I simply wanted to check that the developer (you, Sebastian) were either >> aware of the problem or willing to accept fault reports. (It seems that >> the problem is new to you (fair enough), and that you are open to reports >> (excellent!).

> > I'm more the maintainer than the developper. Sys::Syslog is a piece of > old code, with roots dating back from Perl 4, and has evolved in a strange > and organic way. I became the maintainer last year when it was decided to > released it on the CPAN, independently of the main Perl distrbution, in > order to address CVE-2005-3962 (the sprintf() buffer overflow problem). > > Since then, I've tried to address the tickets that were opened and accept > the patches that were waiting, also trying to tidy up the code and add > new features. >

>> Earlier this afternoon (UK time) I asked (on the main MailScanner email >> list) for a more detailed report of the alleged problem. >> >> It apparently claims a hang (presumably very, very long wait) during the >> "make test" on some platforms.

> > I've seen this occurs on a Debian system, but have not yet investigated > this case.

The hang that was reported to me was on a CentOS 4 system. But another CentOS 4 system worked fine, so it may be a kernel revision problem or something subtle like that. Is there a better Syslog module that I could use instead? The problems with 0.16 and 0.17 have caused me quite a lot of grief, forcing me to issue more than 1 revision after I published my latest stable release, which damages my (and MailScanner's) reputatio as a reliable package. Show quoted text

>> No report forthcoming as yet. >> >> I've also searched the archives of the main list, but was unable to track >> down a detailed report. Sorry.

> > No problem, I can wait. >

>> Thanks. Julian actively reads his email! The chances are that he can dig >> out the exact report of the alleged "make test" hang and send it to you. >> >> The responsibility now fairly and squarely lies with us (the MailScanner >> community) to provide you (Sebastian) with decent information.

> > I took a quick glance at MailScanner::Log, and I think the code can be > simplified by using the features from the recent Sys::Syslog versions. > Namely, the eval{} can be avoided by using the "nofatal" option of > openlog().

I will look into that, many thanks. Show quoted text

> > » http://search.cpan.org/dist/Sys-Syslog/Syslog.pm#openlog > > Also, I really think that you should avoid calling setlogsock() to change > the mechanism and just keep with the native one. Especially on systems > like Solaris, where Alan Burlison, Solaris developper and the main Perl > contact for this system, expressed his clear opinion about this subject > and explained that he prefers the native mechanism over the others.

I will check that out too. I have used Sys::Syslog without exactly the same code for several years now, it's just all been broken by upgrading to a newer release. I don't usually keep absolutely up to date with the latest module releases, as I prefer to stick with the same set that I know work; but the Sys::Syslog module version I was using was getting very old so I upgraded it. Slightly wish I hadn't now :-) Show quoted text

>> Best wishes! Have a good weekend. (Time for me to go home...)

> > Thanks. Although I'll have to spend part of it on the Sys::Syslog problems.

What would we all do at the weekend without a good pile of work to do! - -- Julian Field www.MailScanner.info Buy the MailScanner book at www.MailScanner.info/store MailScanner customisation, or any advanced system administration help? Contact me at Jules@MailScanner.biz PGP footprint: EE81 D763 3DB0 0BFD E1DC 7222 11F6 5947 1415 B654 Get your PCs and servers from Transtec.de, very well built and reliable! -----BEGIN PGP SIGNATURE----- Version: PGP SDK 3.7.0 Charset: UTF-8 wj8DBQFE3dheEfZZRxQVtlQRAoSwAJ9BxxnP0cLfrtyQgI3A8UVY/wnEFwCgjiAi KiIomYnmLJiU4dQkKJSLvIk= =S2AI -----END PGP SIGNATURE----- -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean. For all your IT requirements visit www.transtec.co.uk

Sun Aug 13 21:11:50 2006 SAPER [...] cpan.org - Correspondence added

Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Mon, 14 Aug 2006 03:11:35 +0200
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Sébastien Aperghis-Tramoni <saper [...] cpan.org>

Show quoted text

> The hang that was reported to me was on a CentOS 4 system. But another > CentOS 4 system worked fine, so it may be a kernel revision problem or > something subtle like that.

Once again, I am unable to reproduce the bug on my systems. This is totally frustrating. I'll try tomorrow on the Debian system where I once saw the problem. Show quoted text

> Is there a better Syslog module that I could use instead? The problems > with 0.16 and 0.17 have caused me quite a lot of grief, forcing me to > issue more than 1 revision after I published my latest stable release, > which damages my (and MailScanner's) reputatio as a reliable package.

AFAICT, Sys::Syslog is the only maintained Perl module for sending syslog messages. There are old modules like Net::Syslog or Unix::Syslog, but they are unmaintained and have less features than Sys::Syslog. Show quoted text

> I will check that out too. I have used Sys::Syslog without exactly the > same code for several years now, it's just all been broken by upgrading > to a newer release. I don't usually keep absolutely up to date with the > latest module releases, as I prefer to stick with the same set that I > know work; but the Sys::Syslog module version I was using was getting > very old so I upgraded it. Slightly wish I hadn't now :-)

Sorry for the problems caused by Sys::Syslog. OTOH, the positive side of the upgrade is to find this sort of regressions, which can't be detected by the current test suite, which I'm also working on to improve it. On this matter, I'd say that if the test suite hangs or is longer to execute, but that the module works correctly on normal runs, people should see the module as operational, because the test suite is not perfect and it may try some mechanisms that just won't never work on their systems. Here, I think that my error was to actually connect to the syslog facility of the system. But I'm afraid that several of the functions used by Sys::Syslog can't be replaced or mocked, making data validation tests (which could catch more regression problems) more difficult. -- Sébastien Aperghis-Tramoni Close the world, txEn eht nepO.

Mon Aug 14 12:43:47 2006 SAPER [...] cpan.org - Correspondence added

Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Mon, 14 Aug 2006 18:43:25 +0200
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Sébastien Aperghis-Tramoni <saper [...] cpan.org>

Show quoted text

>> The hang that was reported to me was on a CentOS 4 system. But another >> CentOS 4 system worked fine, so it may be a kernel revision problem or >> something subtle like that.

> > Once again, I am unable to reproduce the bug on my systems. This is > totally frustrating. I'll try tomorrow on the Debian system where I > once saw the problem.

Ok, I gave it a try on a Debian system I have access to, and I can reproduce the problem there: grive:/tmp/Sys-Syslog-0.17# prove -bv t/syslog.t ... ok 135 - setlogsock() called with ['console'] ok 136 - setlogsock() should return true: '1' ok 137 - setlogsock() called with 'console' ok 138 - setlogsock() should return true: '1' ok 139 - openlog() called with facility 'local0' and without option 'ndelay' ok 140 - openlog() should return true: 'Sys::Syslog::SYSLOG' ok 141 - openlog() called with facility 'local0' with option 'ndelay' ok 142 - openlog() should return true: 'Sys::Syslog::SYSLOG' ok 143 - syslog() called with level -1 ok 144 - syslog() should return false: '0' ok 145 - syslog() called with level 'info,notice' ok 146 - syslog() should return false: '0' ok 147 - syslog() called with level 'info,notice' ok 148 - syslog() should return false: '0' And indeed, it hangs at this points, before test 149, at line 140 of t/syslog.t, while trying to write the message to the console: grive:/tmp/Sys-Syslog-0.17# strace -p 15891 Process 15891 attached - interrupt to quit write(4, "<134>Aug 14 18:07:24 perl: uid 0"..., 124 <unfinished ...> Process 15891 detached grive:/tmp/Sys-Syslog-0.17# ls -lF /proc/15891/fd/4 l-wx------ 1 root root 64 Aug 14 18:14 /proc/15891/fd/4 -> /dev/console Adding the "nowait" option to the openlog() avoids this hangs, but then, two processes are left trying to write to the console. So the best solution is probably to just forget this test for the console. As these tests pass on both Mandriva Linux systems I have access to (x86-32 and amd64), I guess it's a packager choice. I note that it also works without hanging on my OSX.4 system. Time to write a rant about OSes that offer a writeable /dev/console but indefinitely hang the processes that ever try to write there. -- Sébastien Aperghis-Tramoni Close the world, txEn eht nepO.

Mon Aug 14 14:23:37 2006 mailscanner [...] ecs.soton.ac.uk - Correspondence added

Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Mon, 14 Aug 2006 19:23:11 +0100
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Julian Field <mailscanner [...] ecs.soton.ac.uk>

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Sébastien Aperghis-Tramoni via RT wrote: Show quoted text

> <URL: http://rt.cpan.org/Ticket/Display.html?id=20946 > > >

>>> The hang that was reported to me was on a CentOS 4 system. But another >>> CentOS 4 system worked fine, so it may be a kernel revision problem or >>> something subtle like that.

>> Once again, I am unable to reproduce the bug on my systems. This is >> totally frustrating. I'll try tomorrow on the Debian system where I >> once saw the problem.

> > Ok, I gave it a try on a Debian system I have access to, and I can reproduce > the problem there: > > grive:/tmp/Sys-Syslog-0.17# prove -bv t/syslog.t > ... > ok 135 - setlogsock() called with ['console'] > ok 136 - setlogsock() should return true: '1' > ok 137 - setlogsock() called with 'console' > ok 138 - setlogsock() should return true: '1' > ok 139 - openlog() called with facility 'local0' and without option 'ndelay' > ok 140 - openlog() should return true: 'Sys::Syslog::SYSLOG' > ok 141 - openlog() called with facility 'local0' with option 'ndelay' > ok 142 - openlog() should return true: 'Sys::Syslog::SYSLOG' > ok 143 - syslog() called with level -1 > ok 144 - syslog() should return false: '0' > ok 145 - syslog() called with level 'info,notice' > ok 146 - syslog() should return false: '0' > ok 147 - syslog() called with level 'info,notice' > ok 148 - syslog() should return false: '0' > > And indeed, it hangs at this points, before test 149, at line 140 of > t/syslog.t, while trying to write the message to the console: > > grive:/tmp/Sys-Syslog-0.17# strace -p 15891 > Process 15891 attached - interrupt to quit > write(4, "<134>Aug 14 18:07:24 perl: uid 0"..., 124 <unfinished ...> > Process 15891 detached > > grive:/tmp/Sys-Syslog-0.17# ls -lF /proc/15891/fd/4 > l-wx------ 1 root root 64 Aug 14 18:14 /proc/15891/fd/4 -> /dev/console > > Adding the "nowait" option to the openlog() avoids this hangs, but then, > two processes are left trying to write to the console. So the best solution > is probably to just forget this test for the console. > > As these tests pass on both Mandriva Linux systems I have access to > (x86-32 and amd64), I guess it's a packager choice. I note that it also > works without hanging on my OSX.4 system.

Thanks for hunting for it, well found. I couldn't reproduce it myself. Show quoted text

> > Time to write a rant about OSes that offer a writeable /dev/console > but indefinitely hang the processes that ever try to write there.

Indeed! Does anyone explicitly only write to the console any more these days? We're a bit beyond the days of attaching line-printers to the console serial port. I still have a VT100 for connecting to Sun's serial ports though :-) Show quoted text

> > -- > Sébastien Aperghis-Tramoni > > Close the world, txEn eht nepO. >

- -- Julian Field www.MailScanner.info Buy the MailScanner book at www.MailScanner.info/store MailScanner customisation, or any advanced system administration help? Contact me at Jules@MailScanner.biz PGP footprint: EE81 D763 3DB0 0BFD E1DC 7222 11F6 5947 1415 B654 For all your IT requirements visit www.transtec.co.uk -----BEGIN PGP SIGNATURE----- Version: PGP SDK 3.7.0 Charset: UTF-8 wj8DBQFE4L+TEfZZRxQVtlQRAl2vAKCIw54UOLRw/zYjs9+YrgSV/uyFPgCfdZ6k J6W7Lb74HEJi41g+MeYjpSY= =4ONk -----END PGP SIGNATURE----- -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean. For all your IT requirements visit www.transtec.co.uk

Mon Aug 14 15:40:36 2006 SAPER [...] cpan.org - Correspondence added

Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Mon, 14 Aug 2006 21:40:16 +0200
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Sébastien Aperghis-Tramoni <saper [...] cpan.org>

Quoting "mailscanner@ecs.soton.ac.uk via RT": Show quoted text

>> And indeed, it hangs at this points, before test 149, at line 140 of >> t/syslog.t, while trying to write the message to the console: [...] >> >> Adding the "nowait" option to the openlog() avoids this hangs, but then, >> two processes are left trying to write to the console. So the best solution >> is probably to just forget this test for the console.

> > Thanks for hunting for it, well found. I couldn't reproduce it myself.

It was a 10 minutes tracking once I had access to a system where the problem occured. Show quoted text

>> Time to write a rant about OSes that offer a writeable /dev/console >> but indefinitely hang the processes that ever try to write there.

> > Does anyone explicitly only write to the console any more these days? > We're a bit beyond the days of attaching line-printers to the console > serial port. I still have a VT100 for connecting to Sun's serial ports > though :-)

Well, if a system offer a writable /dev/console, I'd say one can expect to write to it. Anyway, I now wonder whether the people who added this mechanism really tested it because I'm not totally sure it works as they expected it to work. -- Sébastien Aperghis-Tramoni Close the world, txEn eht nepO.

Mon Aug 14 21:38:12 2006 SAPER [...] cpan.org - Correspondence added

Subject:	Re: [rt.cpan.org #20946] Sys::Syslog problems
Date:	Tue, 15 Aug 2006 03:37:55 +0200
To:	bug-Sys-Syslog [...] rt.cpan.org
From:	Sébastien Aperghis-Tramoni <saper [...] cpan.org>

Here is a beta of the upcoming Sys::Syslog 0.18. » http://www.maddingue.net/Sys-Syslog-0.18_01.tar.gz It passes all tests on my systems: - Perl 5.6.2, 5.8.5-threads, 5.8.8 / Mandrake Linux / x86-32 - Perl 5.8.6-thread / Mac OS X.4 / PowerPC - Perl 5.8.8 / FreeBSD 6.0 / x86 Can you and other people on the MailScanner list test it on the systems where previous versions hanged and confirm that this version doesn't present the same problem? (I can't send a mail to the mailing list myself because I can't connect to the MailScanner site since a few hours now.) Thanks in advance -- Sébastien Aperghis-Tramoni Close the world, txEn eht nepO.

Tue Sep 05 17:13:02 2006 SAPER [...] cpan.org - Correspondence added

The test hanging issue has been resolved with version 0.18 -- Close the world, txEn eht nepO.

Tue Sep 05 17:13:07 2006 SAPER [...] cpan.org - Status changed from 'open' to 'resolved'

Tue Sep 05 17:13:20 2006 SAPER [...] cpan.org - Fixed in 0.18 added