1 |
Mike Frysinger wrote: |
2 |
> On Wednesday 19 August 2009 17:01:27 Thomas Kahle wrote: |
3 |
>> Peter Volkov wrote: |
4 |
>>> В Срд, 12/08/2009 в 00:03 +0200, Thomas Kahle пишет: |
5 |
>>>> the following check in configure.ac fails with --as-needed but works |
6 |
>>>> without it. Any hints what is wrong ? Thanks / Thomas |
7 |
>>> Common reason for failure with --as-needed is the order of libraries. In |
8 |
>>> you previous mail you wrote that with |
9 |
>>> |
10 |
>>> sed "s/\"-lntl\"/\"-lntl\" \"-lgf2x -lntl\"/" -i "${S}/configure.ac" |
11 |
>>> |
12 |
>>> it works but here: |
13 |
>>>> for lib in "" "-lntl" "-lntl -lgf2x" |
14 |
>>> I see different order. Try to change order of -lgf2x and -lntl inside |
15 |
>>> ebuild. Also |
16 |
>> This change of order is actually since I tried what you suggested. I |
17 |
>> find that this check fails just because the compile fails. It has |
18 |
>> nothing to do with configure. So here is my problem stripped down: |
19 |
>> |
20 |
>> $cat ntlcheck.c |
21 |
>> #include <NTL/tools.h> |
22 |
>> int main () { _ntl_GetTime(); } |
23 |
>> |
24 |
>> g++ ntlcheck.c -lntl -lgf2x |
25 |
>> works |
26 |
> |
27 |
> this is a bug in libntl.so. it uses gf2x symbols but doesnt link against it |
28 |
> in libntl.so. you shouldnt have to also link against libgf2x.so. |
29 |
> -mike |
30 |
|
31 |
Thanks => bug #282174 |
32 |
|
33 |
Thomas |
34 |
|
35 |
|
36 |
-- |
37 |
Thomas Kahle |
38 |
|
39 |
The fundamental theorem of algebra is open source. Like any other |
40 |
mathematical theorem it can be applied free of charge and everybody |
41 |
has access to its proof and can convince himself how it works. Why |
42 |
should software be any different? |