1 |
Hi Benda, |
2 |
|
3 |
During May 9 - May 27, I created a demonstration overlay |
4 |
for our initial implimentation [3] (eselect + symlink). |
5 |
The C++ demo program shipped by that git repo suggests |
6 |
the draft implementation is working correctly. |
7 |
|
8 |
During May 27 - June 2, I posted about the proposed |
9 |
solution on Gentoo-dev to wait for comments. Unfortunately, |
10 |
it seems that simply porting Debian's solution to Gentoo |
11 |
is not the best choice, and there are people against it. |
12 |
|
13 |
I redesigned the solution based on ld.so.conf: |
14 |
|
15 |
1. sci-libs/lapack could expose all its shared objects |
16 |
and headers in the public dirs (i.e. /usr/include, |
17 |
/usr/$(get_libdir)/ ). |
18 |
|
19 |
2. optimized implementations build additional libblas.so |
20 |
and liblapack.so objects with patched build systems. |
21 |
We hint ld.so with ld.so.conf via eselect, so any |
22 |
program linked against libblas.so and liblapack.so |
23 |
could make use of this mechanism. |
24 |
|
25 |
The second point is different from our private discussion |
26 |
in telegram. And different from my draft implementation, |
27 |
the redesigned solution will introduce a new USE flag for |
28 |
BLAS/LAPACK reverse dependencies to decide whether the |
29 |
program should link against switchable BLAS/LAPACK or |
30 |
specified BLAS/LAPACK. |
31 |
|
32 |
Although there are negative responses against the first |
33 |
proposal, we are not far from the original schedule -- |
34 |
this week I'm going to implement the redesigned solution |
35 |
for sci-libs/lapack and ask Gentoo-dev for comments again. |
36 |
|
37 |
Thanks, |
38 |
Mo. |
39 |
|
40 |
Reference: |
41 |
|
42 |
1. Drawback of the eselect+symlink solution: |
43 |
https://bugs.gentoo.org/632618 |
44 |
2. The eselect+ld.so.conf solution: https://bugs.gentoo.org/531842 |
45 |
3. My Overlay: https://github.com/cdluminate/my-overlay |