1 |
On 12/13/2010 03:32 PM, Perry Smith wrote: |
2 |
>>> make[4]: Entering directory `/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/p51.powerpc-ibm-aix5.3.0.0/libcpp' |
3 |
>>> /gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/p51.powerpc-ibm-aix5.3.0.0/./prev-gcc/xgcc -B/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/p51.powerpc-ibm-aix5.3.0.0/./prev-gcc/ -B/gsa/ausgsa/home/p/e/pedzan/prefix-launcher-2.1.0/powerpc-ibm-aix5.3.0.0/powerpc-ibm-aix5.3.0.0/bin/ -I/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/gcc-4.2.4/libcpp -I. -I/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/gcc-4.2.4/libcpp/../include -I/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/gcc-4.2.4/libcpp/include -g -O2 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition -Wmissing-format-attribute -pedantic -Wno-long-long -I/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/gcc-4.2.4/libcpp -I. -I/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/gcc-4.2.4/libcpp/../include -I/gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildro |
4 |
ot |
5 |
>> /gcc/gcc-4.2.4/libcpp/include -c -o charset.o -MT charset.o -MMD -MP -MF .deps/charset.Po /gsa/ausgsa-h2/01/pedzan/src/prefix-launcher-2.1.0/buildroot/gcc/gcc-4.2.4/libcpp/charset.c |
6 |
>>> xgcc: Internal error: Killed (program cc1) |
7 |
>> |
8 |
>> Interesting - this works since ages on the AIX 5.3 boxes here... |
9 |
> |
10 |
> Yes. As a curiosity, I pulled down 4.2.4 and compiled it on my own sandbox and it did fine. |
11 |
> |
12 |
>> Are you able to compile 'charset.o' with '-save-temps' compiler flag? |
13 |
> |
14 |
> Its been a long weekend with many misadventures. |
15 |
|
16 |
This is how it works, unfortunately ;) |
17 |
|
18 |
> I did not know about the INTERACTIVE method and I copied and pasted the compile command (above) |
19 |
|
20 |
The INTERACTIVE feature is an internal detail and for debugging purpose only. |
21 |
However, there is some doc in prefix-launcher-*/BUILDING. |
22 |
|
23 |
> and added -save-temps and it compiled fine. So I then just did "make" in that directory -- which worked. |
24 |
|
25 |
WTF...? |
26 |
|
27 |
> Moved up and did "make" again and that eventually failed with arg list too long. |
28 |
|
29 |
"arg list too long" rings the "ncargs defaults to 4" bell: |
30 |
$ lsattr -E -l sys0 -a ncargs |
31 |
|
32 |
We do have it set to 40 here, but on some AIX6.1 I've seen this set to 256: |
33 |
$ chdev -l sys0 -a ncargs=40 |
34 |
|
35 |
> So, I wasn't using *exactly* the right environment. |
36 |
|
37 |
This might be the reason: Inside prefix-launcher the environment is a little |
38 |
larger than outside, which may trigger the arg list too long problem. |
39 |
Thing is that both the commandline argument /plus/ the environment variables |
40 |
are counted in here for 'ncargs'. |
41 |
|
42 |
There is another thing: You could 'source' the installed environment-file of prefix-launcher: |
43 |
$ . /prefix/for/launcher/chost/bin/prefix-launcher-env.sh |
44 |
|
45 |
Then you should have a similar environment than within the builds, |
46 |
except for a package's local build-variables set within pkg/pkg.build. |
47 |
|
48 |
<snip> |
49 |
> First, my build scripts for gcc always do: |
50 |
> |
51 |
> ulimit -S -d unlimited |
52 |
> |
53 |
> I can't remember why I've added that over the years. But build script that worked did this: |
54 |
|
55 |
Seems like you've hit the arg list too long error before. |
56 |
|
57 |
> As I mentioned, the "make" above ended with arg list too long. |
58 |
> I hit that again doing some other path. So, I restarted your process with these changes: |
59 |
> |
60 |
> 1) I *forgot* to do the CONFIG_SHELL (and I'm paying the price for that. |
61 |
> I plan to open a defect on ksh. |
62 |
|
63 |
Good to know you have resources to do so. |
64 |
|
65 |
<snip> |
66 |
> 4) Used a different gcc. Before I was using my newly built gcc 4.5.0 and this last trip I'm using 4.3.1. |
67 |
|
68 |
It should work with any Ansi-C compiler that is able to bootstrap gcc-4.2.4. |
69 |
For what it's worth, on AIX 6.1 I've recently used gcc-4.2.0 from the |
70 |
"AIX Toolbox for Linux Applications". |
71 |
|
72 |
> Can we set CONFIG_SHELL inside the top level Makefile? I'm going to just add it to my .bash_profile too. |
73 |
|
74 |
It isn't that easy. The only thing I could do is to set |
75 |
CONFIG_SHELL = $(SHELL) |
76 |
as I cannot have an idea where to find some bash in that Makefile. |
77 |
Additionally, bash isn't a technical requirement for prefix-launcher on any platform. |
78 |
|
79 |
> One other suggestion, I assumed that the install of the prefix-launcher was going to be a half minute adventure |
80 |
> and the big adventure was going to come while using it. |
81 |
> My suggestion is mention that installing it will take a long time. |
82 |
|
83 |
Added, thanks! |
84 |
|
85 |
> Also, give some help (like you just did) about doing things piecemeal if things go wrong. |
86 |
|
87 |
You're welcome to draft up something you'd like to read (in the MediaWiki), |
88 |
and I'll be happy to review and add it. |
89 |
|
90 |
/haubi/ |
91 |
-- |
92 |
Michael Haubenwallner |
93 |
Gentoo on a different level |