Cool. My suggestion would not have solved the issue, but looking at Damjan's .am files gives me a hint why the years ago project had issues, for example:
libsvm_la_LIBADD = libvppinfra.la -lrt -lpthread libsvm_la_DEPENDENCIES = libvppinfra.la I'll bet we did not have the second line in the years ago project. (This is just an example; the years ago problems were not VPP related.) If anyone is interested in whether I could build after removing all the LIBADD and related DEPENDENCIES in my experiment, the answer is: ALMOST. I had to add this if WITH_IPV6SR bin_vpp_LDADD += -lcrypto endif in vpp.am. But I was able to remove more lines than I had to add back. Burt On Thu, Jan 12, 2017 at 7:46 PM, Jon Loeliger <j...@netgate.com> wrote: > > > On Thu, Jan 12, 2017 at 5:40 PM, Damjan Marion <dmarion.li...@gmail.com> > wrote: > >> >> A few more details: this is race condition in parallel execution of “make >> install” targets. >> I am aware of the problem, but for me it was hard to troubleshoot as it >> is not happening >> in my system, and so far I noticed it only occasionaly in verify jobs. >> >> I think it was caused by: >> >> libvppinfradir = ${libdir} >> libvppinfra_LTLIBRARIES = libvppinfra.la >> >> And is now fixed with: >> >> lib_LTLIBRARIES += libvppinfra.la >> >> >> Basically by having 2 LTLIBRARIES groups, we were allowing Make to >> install them in parallel. >> As we are relinking libs during install, it was happening that that lib >> was not there when >> it was needed. >> >> Let me know if it works now, if not I have some other ideas what we can >> do… >> >> Thanks, >> >> Damjan > > > That does appear to have fixed it! > > Thank you! > > jdl > > PS -- Yes, I see it in top-of-tree as well! > > > _______________________________________________ > vpp-dev mailing list > vpp-dev@lists.fd.io > https://lists.fd.io/mailman/listinfo/vpp-dev >
_______________________________________________ vpp-dev mailing list vpp-dev@lists.fd.io https://lists.fd.io/mailman/listinfo/vpp-dev