1 |
Hans-Gunther Borrmann posted |
2 |
<200603081035.15013.hans-gunther.borrmann@×××××××××××××××.de>, excerpted |
3 |
below, on Wed, 08 Mar 2006 10:35:14 +0100: |
4 |
|
5 |
> I tried to "emerge --oneshot glibc" on a dual opteron system with |
6 |
> "MAKEOPTS="-j3" in make.conf. The emerge began several times to loop at |
7 |
> different places. The looping program always was "sh". I could successfully |
8 |
> emerge glibc after commenting out my MAKEOPTS setting in make.conf. What are |
9 |
> the experiences with setting MAKEOPTS on smp servers? |
10 |
|
11 |
I'm running a dual Opteron (242) here, and routinely run MAKEOPTS="-j6 |
12 |
-l6" here, plus sometimes run multiple parallel emerges (using emerge -tp |
13 |
first to see what the dependencies are, then emerge -a to verify that the |
14 |
same dependency isn't in both). |
15 |
|
16 |
As I'm running ~amd64, plus running way out front with still-masked gccs |
17 |
(now gcc-4.0.2-rX and gcc-4.1.0), which sometime last year (with |
18 |
gcc-4.0.0) required still experimental glibcs if I wanted to compile themm |
19 |
with the new gcc, I tend to do more merging of everything, including |
20 |
glibc, than most, and IDR ever having that sort of issue. |
21 |
|
22 |
Occasionally, there's an ebuild that has parallel job issues, but they |
23 |
tend to be pretty rare, because as they are found and bugged, the ebuild |
24 |
will be adjusted to filter out the -jX and force serial compilation jobs. |
25 |
In fact, a quick: "grep MAKEOPTS /p/sys-libs/glibc/*|grep j1" returns a |
26 |
huge number of hits, multiple per ebuild, where serial compilation is |
27 |
already forced by the various glibc ebuilds. |
28 |
|
29 |
Assuming you didn't manually filter all those forced -j1 calls, all I can |
30 |
say is perhaps another one or more need added. They apparently haven't |
31 |
been triggered for most of us or it'd already be forced, but maybe your |
32 |
machine just happens to trigger it in a different area. <shrug> Either |
33 |
that or there's something strange going on with your bash or make or the |
34 |
like and it's filtering the wrong thing. |
35 |
|
36 |
What glibc are you attempting to merge, what are your bash, portage, gcc, |
37 |
and make versions, and where exactly is the error? If you know bash |
38 |
scripting well enough to follow what the ebuild is doing, it shouldn't be |
39 |
hard to verify whether the problem make calls are forcing -j1 or not (some |
40 |
in the glibc ebuilds do, some don't). |
41 |
|
42 |
Also note that MAKEOPTS and MAKE_OPTS are two entirely different vars. |
43 |
There's a note in the xorg ebuilds mentioning that some users had set the |
44 |
_ version instead of the run-together version, and it messed things up. |
45 |
The xorg ebuilds have a preemptive fix for the problem that the glibc |
46 |
ebuilds are missing. If you mistakenly set MAKE_OPTS instead of MAKEOPTS... |
47 |
|
48 |
-- |
49 |
Duncan - List replies preferred. No HTML msgs. |
50 |
"Every nonfree program has a lord, a master -- |
51 |
and if you use the program, he is your master." Richard Stallman in |
52 |
http://www.linuxdevcenter.com/pub/a/linux/2004/12/22/rms_interview.html |
53 |
|
54 |
|
55 |
-- |
56 |
gentoo-amd64@g.o mailing list |