Bug 202197

Summary: net/dante 1.4.1 fails to build on AMD64
Product: Ports & Packages Reporter: Thomas Scholten <thomas.scholten>
Component: Individual Port(s)Assignee: Jason Unovitch <junovitch>
Status: Closed Overcome By Events    
Severity: Affects Some People CC: junovitch, thomas.scholten
Priority: ---    
Version: Latest   
Hardware: amd64   
OS: Any   
Attachments:
Description Flags
unoffical patch from inet.no none

Description Thomas Scholten 2015-08-09 11:16:31 UTC 
Hello,

it's quite being a while for me filing bugs here. So i wasn't sure how to reopen BUG#192295 again. So obey me for opening this one as an semi duplicate.

Bug 192295 was closed as fixed, but the same build error with loss of network and connectivity and filling mbufs appears still on my system: 

sysctl kern.version
kern.version: FreeBSD 9.3-RELEASE-p21 #0 Tue Jul 28 09:02:40 UTC 2015

My Build of the ports dies with the same symptoms and should be fixed by commit r269489 for the 10.x/i386 series. As it seems, also reported as last entry by <yoshin-t@jp.fujitsu.com> in Bug 192295, this commit didn't made it by now to the 9.3-RELEASE-p21.

Build Log:

[..]
checking for strlcpy... (cached) yes
checking for backtrace... no
checking for incorrect inet_ntoa behaviour... no
checking for working pselect()... yes
checking for working sockatmark... yes
checking for __libc_enable_secure... no
checking read-side pipe system... ^C
^C
^C
[actually ^C doesn't work out]

If anymore debug info will help out i can post it here as an addition.

Cheers,

Thomas
Comment 1 Thomas Scholten 2015-08-11 15:37:02 UTC 
Created attachment 159771 [details]
unoffical patch from inet.no
Comment 2 Thomas Scholten 2015-08-11 15:38:03 UTC 
Here an Update for this case,

in parallel i triggerd the devs of dante for a possible fix and workaround and got a patch for the configure script mainly skipping the buffer test which leads to a successful build. When i get the information provided in #192295 right there is still a fix for the kernel source needed in the 9.3 RELEASE tree to enable the buffer test again an prevent the dying network interface without the patch.

Hope it helps to get this ticket any further,

Thomas
Comment 3 Jason Unovitch freebsd_committer freebsd_triage 2016-04-06 01:50:02 UTC 
Is this PR still valid?  This port does not have any issues building in the FreeBSD build cluster (http://portsmon.FreeBSD.org/portoverview.py?category=net&portname=dante)
Comment 4 Jason Unovitch freebsd_committer freebsd_triage 2016-06-26 01:40:38 UTC 
I'm not sure what is required here.  I just testbuilt and here is the result on 9.3 amd64.

...
checking for system V getpwnam... no
checking read/send-side pipe system... recvbased
checking for errorless select behavior with high timeouts... yes
...

I checked myself and built successfully on the following platforms:
9.3-RELEASE-p42      amd64
9.3-RELEASE-p42      i386
10.1-RELEASE-p34     amd64
10.2-RELEASE-p17     amd64
10.2-RELEASE-p17     i386
11.0-ALPHA3 r301815  amd64
11.0-ALPHA3 r301815  i386

There are still no reported issues at (http://portsmon.FreeBSD.org/portoverview.py?category=net&portname=dante).
Comment 5 Thomas Scholten 2016-06-26 08:39:21 UTC 
I think it can be closed now. The bug belonged to an kernel issue which was triggered with -p21. When the built now runs nicely  with -p42 it seems the kernel bug has being patch away and no more fixes are needed.

Possibly inet.no pushed there fix into the portstree. Either way this PR can be marked as fixed.

Thank you for your support.