1 |
Il giorno lun, 29/11/2010 alle 22.57 +0100, Andreas K. Huettel ha |
2 |
scritto: |
3 |
> |
4 |
> I have a personal suspicion, namely that libcblas.so is (on my |
5 |
> system) |
6 |
> actually (symlinked to) libgslcblas.so.0, and that maybe cmake reacts |
7 |
> because |
8 |
> the library filename is not equal its SONAME. Does this make sense? |
9 |
> And if |
10 |
> yes, what can I do about it? |
11 |
|
12 |
No. As I have written already [1] & [2], there is no need for the |
13 |
filename and the soname to be related at all |
14 |
|
15 |
What I think the problem could be here is that libblas and |
16 |
libcblas/libgslcblas provide two different interfaces; clapack finds |
17 |
itself fine with libblas, but it doesn't find the required symbols in |
18 |
libcblas, so --as-needed discards the linking. |
19 |
|
20 |
I don't think I have the libraries around, maybe the tinderbox. |
21 |
Otherwise prod me on IRC/Jabber, it's faster, then we can post the |
22 |
results. |
23 |
|
24 |
|
25 |
[1] http://blog.flameeyes.eu/2010/10/08/linkers-and-names |
26 |
[2] |
27 |
http://blog.flameeyes.eu/2009/10/27/a-shared-library-by-any-other-name |
28 |
|
29 |
-- |
30 |
Diego Elio Pettenò — “Flameeyes” |
31 |
http://blog.flameeyes.eu/ |
32 |
|
33 |
If you found a .asc file in this mail and know not what it is, |
34 |
it's a GnuPG digital signature: http://www.gnupg.org/ |