Check the attached build log, this happens with -j14.
Created attachment 171812 [details]
The problem is that I'm not able to reproduce the problem. Did you have mit-krb5 installed on this machine before emerging heimdal? Do you use distcc - I mean -j14 are really a lot of compile units (btw - I have a test installation x86 in a vbox using distcc and -j10 for compilation - no issues here).
distcc does not influence installs, you need a truly multi-way system to test parallel install. Here it's running -j14 on an 8-way.
What do you mean by "installs" - Your compilation is dying during compiling "/var/tmp/portage/app-crypt/heimdal-1.2.1-r1/work/heimdal-1.2.1/lib/krb5".
./krb5.h:43:22: error: krb5_err.h: No such file or directory
tastes a bit like you have a krb5.h on your system left from a mit-krb5 install. Therefore my question once again: "Did you have mit-krb5 installed on this machine before emerging heimdal?" An krb5.h in any global include directory wins over the local krb5.h of heimdal during compile. Please check the situation. I am pretty sure that the issue isn't related to parallel build but I might be wrong of course.
Oh sorry I confused this bug with another, my fault :/
It's possible mit-krb5 was in the system, I'm building every package in tree so I can't easily tell what was in before, I'll check the emerge log and let you know.
Ok, thx. But if mit-krb5 was in system before and the emerge fails it's still a bug I have to fix ... seams really strange to me. Would be nice if we can investigate a procedure for reproduction of the problem so that I can test and fix it myself.
i got the same problem some time ago, but wasn't able to reproduce it. the problem is the use of libcom_err and compile_et to create some of the header files including krb5_err.h. i will try to write a patch for that as soon as possible. for now, please use -j1 to avoid the bug.
The problem is in cf/Makefile.am.common
This causes double generation of everything, instead the .h should depend on the .c file, and the .c on .et, so that only one call is present.
(In reply to comment #8)
> The problem is in cf/Makefile.am.common
> $(COMPILE_ET) $<
> $(COMPILE_ET) $<
> This causes double generation of everything, instead the .h should depend on
> the .c file, and the .c on .et, so that only one call is present.
Aren't the compile_et binaries gone now with heimdal-1.3? Remember the file collisions with sys-libs/e2fsprogs-libs?