Re: Inconsistent -DNO_CLEAN build failure r293913 -> r294086

On 1/15/16 6:00 AM, David Wolfskill wrote:
> OK, first: I realize that using "-DNO_CLEAN" is treading on somewhat
> unstable ground.  I'm testing; the results I obtained this morning were
> a bit unexpected, so I'm reporting them in case said results might be
> of use or interest to others.
> 
> As indicated in the Subject, I started a couple of machines (a dedicated
> "build machine" and my laptop) at r293913, updated source to r294086,
> then proceeded to perform "make -j16 -DNO_CLEAN buildworld" on each.
> 
> Please note that in each case, I have "WITH_FAST_DEPEND=1" in
> /etc/src.conf.
> 
> The laptop chugged along and completed normally; the build machine
> encountered:
> 
>>>> World build started on Fri Jan 15 04:28:48 PST 2016
> ...
>>>> stage 5.1: building 32 bit shim libraries
> ...
> --- lib/libldns__L ---
> --- libprivateldns.so.5 ---
> /usr/obj/usr/src/tmp/usr/bin/ld: skipping incompatible /usr/obj/usr/src/tmp/usr/lib/libgcc_s.so when searching for -lgcc_s

This is odd...

> --- secure/lib/libssl__L ---
> --- libssl.a ---
> --- kerberos5/lib/libhx509__L ---
> --- keyset.po ---
> --- secure/lib/libssl__L ---
> building static ssl library
> --- lib/libldns__L ---
> /usr/obj/usr/src/tmp/usr/bin/ld: cannot find -lgcc_s

By this point (lib/libldns__L) libgcc should be built and installed. I
assume it says 'cannot find' since it is incompatible.

I don't think this is a build race or FAST_DEPEND issue.

It could be a -DNO_CLEAN issue. Nothing is standing out as an issue in
the commit range given, except if you were building for arm.


> --- secure/lib/libssl__L ---
> ar -crD libssl.a `NM='nm' NMFLAGS='' lorder bio_ssl.o d1_both.o d1_clnt.o d1_lib.o d1_meth.o d1_pkt.o d1_srtp.o d1_srvr.o s23_clnt.o s23_lib.o s23_meth.o s23_pkt.o s23_srvr.o s3_both.o s3_cbc.o s3_clnt.o s3_enc.o s3_lib.o s3_meth.o s3_pkt.o s3_srvr.o ssl_algs.o ssl_asn1.o ssl_cert.o ssl_ciph.o ssl_conf.o ssl_err.o ssl_err2.o ssl_lib.o ssl_rsa.o ssl_sess.o ssl_stat.o ssl_txt.o t1_clnt.o t1_enc.o t1_ext.o t1_lib.o t1_meth.o t1_reneg.o t1_srvr.o tls_srp.o  | tsort -q` 
> --- lib/libldns__L ---
> cc: error: linker command failed with exit code 1 (use -v to see invocation)
> *** [libprivateldns.so.5] Error code 1
> 
> make[4]: stopped in /usr/src/lib/libldns
> 
> 
> from which that process didn't recover at all well... though a
> re-start without "-DNO_CLEAN" completed successfully.
> 
> Here are the relevant "uname -a" outputs:
> 
> FreeBSD g1-252.catwhisker.org 11.0-CURRENT FreeBSD 11.0-CURRENT #305  r293913M/293913:1100093: Thu Jan 14 04:55:39 PST 2016     root_at_g1-252.catwhisker.org:/common/S4/obj/usr/src/sys/CANARY  amd64
> 
> FreeBSD g1-252.catwhisker.org 11.0-CURRENT FreeBSD 11.0-CURRENT #306  r294086M/294086:1100094: Fri Jan 15 05:03:24 PST 2016     root_at_g1-252.catwhisker.org:/common/S4/obj/usr/src/sys/CANARY  amd64
> 
> 
> FreeBSD freebeast.catwhisker.org 11.0-CURRENT FreeBSD 11.0-CURRENT #1961  r293913M/293913:1100093: Thu Jan 14 04:49:29 PST 2016     root_at_freebeast.catwhisker.org:/common/S4/obj/usr/src/sys/GENERIC  amd64
> 
> FreeBSD freebeast.catwhisker.org 11.0-CURRENT FreeBSD 11.0-CURRENT #1962  r294086M/294086:1100094: Fri Jan 15 05:33:30 PST 2016     root_at_freebeast.catwhisker.org:/common/S4/obj/usr/src/sys/GENERIC  amd64
> 
> 
> One other (possibly-salient) point is that the build machine's disk
> drive is an SSD.  (And yes, that does tend to help make it faster.)
> 
> Peace,
> david
> 


-- 
Regards,
Bryan Drewery