1 |
Hi Benda, |
2 |
|
3 |
On Mon, Nov 04, 2019 at 12:04:25AM +0800, Benda Xu wrote: |
4 |
> Dear All, |
5 |
> |
6 |
> I am going to give an oral presentation at the Computing in High Energy |
7 |
> Physics Conference at Adelaide Australia on Tuesday: |
8 |
> |
9 |
> https://indico.cern.ch/event/773049/sessions/323858/#20191105 |
10 |
> |
11 |
> I have uploaded my slides at, |
12 |
> |
13 |
> http://hep.tsinghua.edu.cn/~orv/pd/chep2019.pdf |
14 |
> |
15 |
> |
16 |
> Note that there will be talks on Spack and Conda after mine. A |
17 |
> head-to-head comparison in the session will be likely to happen. |
18 |
|
19 |
You can see their slides ahead of time. I think they are all posted |
20 |
already, in case you want to make a comparison in your talk. |
21 |
|
22 |
> |
23 |
> Comments welcome! |
24 |
|
25 |
Here are my comments for the slides: |
26 |
|
27 |
For the theme, I like the darker purple. See here an example from when |
28 |
I first presented Portage in a group meeting here: |
29 |
https://docs.google.com/presentation/d/1dYyYL4L9i4g6WibYpsJqL__dqD3w3KZmQKFkw40SanU/edit?usp=sharing |
30 |
This is just a suggestion, though. Your theme is totally fine. |
31 |
|
32 |
Slide 3: |
33 |
|
34 |
There is a typo: maintenability -> maintainability |
35 |
|
36 |
Slide 4: |
37 |
|
38 |
from daily to ... -> from daily use to ... |
39 |
|
40 |
Slide 5: |
41 |
|
42 |
I think this is a key slide. It shows the main difference between |
43 |
prefix and others, namely, that the INTERP points into EPREFIX and |
44 |
makes it independent. Important points are that it doesn't require |
45 |
sourcing any setup scripts to use, neither setting any variables. |
46 |
You should point that out, and say that we can only do this because |
47 |
of the special glibc that we build. No other package manager does |
48 |
that as far as I can tell. Also, you can use the same prefix on |
49 |
CVMFS for Ubuntu, CentOS, etc, instead of having one prefix per host |
50 |
system. That's another great advantage that prefix has over everything |
51 |
else. |
52 |
|
53 |
Slide 6: |
54 |
|
55 |
For CMake, it should be cmake -DCMAKE_INSTALL_PREFIX="$EPREFIX" ... |
56 |
|
57 |
Slide 7: |
58 |
|
59 |
Rust Carge -> Rust Cargo |
60 |
|
61 |
I'd avoid bundling the stuff that doesn't exist yet with the ones that |
62 |
already exist. Maybe it's better to put them only in the next slide |
63 |
and say that it would be possible to have converters for some things. |
64 |
Spack and conda seem quite unfeasible, though, so I wouldn't list those. |
65 |
Spack uses python for recipes, and shares the same recipe for all versions |
66 |
of a given package. See the recipe for ROOT, which was based on my ebuild: |
67 |
https://github.com/spack/spack/blob/6e78730a9776e310e375fa4697aae828665af85f/var/spack/repos/builtin/packages/root/package.py |
68 |
Not sure about conda, but it's more like apt/rpm, so converters are |
69 |
probably not an option, or at least very hard not to make a mess, since |
70 |
packages will be duplicated in both worlds. |
71 |
|
72 |
Another thing is that spack and spack-dev are somewhat different things. |
73 |
The spack-dev thing is supposed to be something to help work on package |
74 |
development by creating a dev environment where you can work on the |
75 |
source and use spack to install. (Think like using ebuild configure, |
76 |
then work on the source, then ebuild compile/install/qmerge by hand.) |
77 |
|
78 |
Non-exist, ... -> Do not exist yet, ... |
79 |
|
80 |
Slide 8: |
81 |
|
82 |
Please point to packages.gentoo.org :-) It's a great resource. It lists |
83 |
the number of packages in the front page, so you could just put a |
84 |
screenshot in your slide and make it a link to the site. Same for the |
85 |
categories, you can link to https://packages.gentoo.org/categories and |
86 |
use a screenshot. |
87 |
|
88 |
Not sure about what you mean with the red box (Our manifesto). It can |
89 |
probably be removed. |
90 |
|
91 |
Slide 10: |
92 |
|
93 |
Maybe try using minted for getting syntax hilighting, and put the code |
94 |
in listing boxes. I can send you examples if you want. It will look a |
95 |
lot nicer! |
96 |
|
97 |
Slide 11: |
98 |
|
99 |
Coexistance -> Coexistence |
100 |
|
101 |
This is also an important slide. Here there is a combinatorial build |
102 |
with many different versions of everything and then "views" into the |
103 |
builds are created by symlinking into the real directories, but that |
104 |
has its issues, obviously. Gentoo puts all of userspace into a single |
105 |
directory and stacked prefix can be used to share the base and change |
106 |
the top while still keeping everything consistent when things are |
107 |
updated, which is the important point to make. |
108 |
|
109 |
Slide 12: |
110 |
|
111 |
This is also a crucial slide, as it explains the bootstrapping process. |
112 |
You can probably move it earlier, when you start explaining about |
113 |
prefix and put a lot of emphasis on the fact that only the host kernel |
114 |
matters, and that we can have one thing that works on any Linux distro, |
115 |
as long as the kernel is recent enough. This is something only prefix |
116 |
can provide at the moment, and with CVMFS, it's easy to take advantage |
117 |
of this benefit. |
118 |
|
119 |
Slide 13: |
120 |
|
121 |
Maybe use some screenshots from Android? That would be really cool. |
122 |
|
123 |
Slide 14: |
124 |
|
125 |
I think that you don't need to mention other package managers in the |
126 |
conclusion. Just focus on the strengths of Portage and prefix in |
127 |
combination with a means of distribution like CVMFS (or any distributed |
128 |
or networked filesystem). I think that it would be great to see more |
129 |
adoption of Gentoo in the scientific community! |
130 |
|
131 |
The presentation is very good overall, and I'm sure you'll do great |
132 |
there presenting Gentoo. |
133 |
|
134 |
Good luck! |
135 |
|
136 |
Cheers, |
137 |
-Guilherme |