1 |
On Monday 28 April 2014 10:30:58 Michael Haubenwallner wrote: |
2 |
> IIRC, aixdll is needed by portage itself for these tasks: |
3 |
> *) extract the binary dependencies for preserve-libs feature |
4 |
> *) merge library archives containing shared objects - even with |
5 |
> "aix-soname". |
6 |
|
7 |
Then I think it needs to be installed before gcc which builds aix-soname |
8 |
libraries, but not necessarily any earlier than that. |
9 |
|
10 |
> > I don't have a good idea of what texinfo might possibly be used for, |
11 |
> > except for makeinfo for which there is now a more general workaround in |
12 |
> > place. Is it used for anything else? |
13 |
> |
14 |
> IIRC nope, makeinfo might have been the only reason. |
15 |
|
16 |
I'll remove it, then. |
17 |
|
18 |
> |
19 |
> > sys-apps/diffutils is apparently needed to work around |
20 |
> > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14251 , but I don't see how |
21 |
> > -- that bug needs gnu coreutils, not diffutils. Do you know what this is |
22 |
> > for? |
23 |
> |
24 |
> AFAICT, the comparison stage in gcc uses cmp. IIRC I've seen failures when |
25 |
> that is not the GNU diffutils' cmp. |
26 |
|
27 |
Ah yes, good point. |
28 |
|
29 |
> > Moreover, do you know whether this workaround is still necessary for the |
30 |
> > aix versions we support? |
31 |
> |
32 |
> Comment it out or drop it, and the buildbots here whill tell. |
33 |
|
34 |
Alright, I'll give it a test later. |
35 |
|
36 |
> > If I understand correctly, dev-libs/libiconv is a dependency for a lot of |
37 |
> > programs, including gcc, and as such needs to be installed before gcc. |
38 |
> > This is clear. Are there any other packages in the early bootstrap that |
39 |
> > depend on it? Could it be installed immediately before gcc? I see that it |
40 |
> > was recently moved from immediately before gcc, to one of the first |
41 |
> > things in stage3. Why is this? What hell are you avoiding this way that I |
42 |
> > should retain? |
43 |
> |
44 |
> Okay, you've asked: The "hell with shared libiconv.a" is this one: |
45 |
> |
46 |
> [snip] |
47 |
|
48 |
Hellish indeed. |
49 |
|
50 |
libiconv used to be built immediately before gcc, but recently you moved it to |
51 |
very early stage3. Why is that? As far as I can see, gcc is the earliest |
52 |
stage3 package that uses iconv. Or is there a hidden dependency somewhere? |
53 |
Evidently you changed it in response to a bug, but I don't see it. |
54 |
|
55 |
-- Ruud |