1 |
I'll give my final thoughts on this, because I knew it was political, and |
2 |
there's already tons of traction in *Spack*: |
3 |
|
4 |
On Thu, Apr 04, 2019 at 01:51:58AM +0000, Francois Bissey wrote: |
5 |
> Gentoo prefix is awesome but some areas are not as flexible as spack. |
6 |
> Mainly because it is designed like a gentoo distro as a single tree |
7 |
> install. Everything goes into one prefix. |
8 |
|
9 |
So, what *could* have happened was that the HPC community did due diligence, and |
10 |
said, "hmm, what's this portage thing? it's like BSD ports, etc." and extended |
11 |
it for the HPC use cases. What I see time and time again, is that the HPC |
12 |
community reinvents something that exists rather than joining an existing |
13 |
community and making it better. Why do they isolate themselves? I won't go |
14 |
into my suspected reasons, but it's stupid, because there's so much "not |
15 |
invented here" problems that go on in supercomputing. |
16 |
|
17 |
Also, I think portage *can* fill all of the use cases of *Spack*, but it's not |
18 |
done in the way *Spack* presents it. I personally think that providing build |
19 |
configuration on the command line is a bad way to do it, like *Spack* does, and |
20 |
it should be encoded in overlays and/or site specific portage configuration |
21 |
files. Though, I don't know if there's something equivalent to *make.conf*, et |
22 |
al., I don't know Spack well enough. |
23 |
|
24 |
Also, yes, I get that multi-compiler issues that we have in HPC is a strange |
25 |
requirement that we have, but again, *portage* could have solved it if the |
26 |
HPC community had contributed -- and I don't think it's impossible with |
27 |
current *portage*. I do realize you can't use *Prefix* out of the box with |
28 |
the existing *portage* tree. |
29 |
|
30 |
I haven't given it a ton of thought of how to solve it, but we ought to be |
31 |
doing static linking anwyays, because shared linking is hell in a supercomputer |
32 |
center for all of the aforementioned reasons of needing lots of different user |
33 |
requirements. |
34 |
|
35 |
So, I can't turn back time and tell the *Spack* team to go and use *portage*, |
36 |
because it exists now, and it has a lot of traction, but I am going to say it |
37 |
was a stupid decision. |