1 |
Michael Haubenwallner wrote: |
2 |
> Hi Dag, |
3 |
> |
4 |
> Dag Sverre Seljebotn wrote: |
5 |
> |
6 |
>> Just stumbled over Gentoo prefix and it seems very exciting (see my |
7 |
>> usecase below). My question: |
8 |
>> |
9 |
> |
10 |
> You're welcome! |
11 |
> |
12 |
> |
13 |
>> Is it possible to use the host system's bash/gcc/... when having a |
14 |
>> Gentoo prefix on Mac or Linux? I.e. a system for setting up software |
15 |
>> which doesn't have *too* much redundancy with the host system. One would |
16 |
>> then set PATH, LD_LIBRARY_PATH, MANPATH etc. to use it as an overlay |
17 |
>> over the host system. |
18 |
>> |
19 |
> |
20 |
> While it should be "possible" to some degree to do this, we can not recommend |
21 |
> and of course not really support this - although you might get hints for specific |
22 |
> problems here and there. |
23 |
> |
24 |
Thanks for the input, it's very helpful. |
25 |
|
26 |
> First hint: Do _not_ use LD_LIBRARY_PATH (on a regular basis) for various reasons. |
27 |
> |
28 |
I keep seeing this statement repeated on the web, but somehow always |
29 |
feel the reasons given don't apply in my usecase (I might of course be |
30 |
wrong..). And using LD_LIBRARY_PATH is incredibly convenient for a |
31 |
number of reasons: It allows moving directories around, moving programs |
32 |
to our cluster for execution without recompiling, ship a binary tarball |
33 |
to others with the same system. |
34 |
|
35 |
I realize people have strong opinions about this. Please, I don't want |
36 |
to start any flamewars. (In particular I don't feel strongly for arguing |
37 |
about one vs. the other, just saying that I'm currently a happy |
38 |
LD_LIBRARY_PATH user.) |
39 |
|
40 |
> <snip> |
41 |
> |
42 |
>> - Need to distribute software fast and iteratively to coworkers on the |
43 |
>> other side of the world with a totally different setup |
44 |
>> - Linux, Mac, Solaris, Windows... |
45 |
>> |
46 |
> |
47 |
> Third hint: |
48 |
> Having that many different systems and users feels like becoming a maintenance |
49 |
> nightmare when trying to use as much as possible packages from the host system. |
50 |
> |
51 |
Something that might help is the simplicity of the packages I want to |
52 |
deploy, but I do see your point. Others have already taken this approach |
53 |
though, such as Sage (http://sagemath.org) and it works pretty well. |
54 |
(Well, there's no Windows.) |
55 |
|
56 |
It is a weakness shared with the *current* approach used by pretty much |
57 |
everybody, which is to provide my package, list the 15 dependencies, and |
58 |
have people sort it out by themselves -- which is a major waste of time |
59 |
(to make an understatement, not all scientists are experts in building |
60 |
software). To the point that people rewrite complicated stuff themselves |
61 |
just to avoid the dependencies... |
62 |
|
63 |
(I live in a world where people just don't have root access to their |
64 |
boxes. Using package managers to install software only works if the |
65 |
software is stable and one can ask sysadmins to do it and wait some |
66 |
days; which drastically lengthens the turnaround time.) |
67 |
|
68 |
Another alternative is writing packages for Sagemath -- |
69 |
http://sagemath.org. I already know that the Gentoo community don't |
70 |
particularly care for the Sagemath approach though :-) : |
71 |
|
72 |
http://bugs.gentoo.org/show_bug.cgi?id=240622 |
73 |
|
74 |
Thanks for the input, I'll just have to see where this takes me... |
75 |
|
76 |
Dag Sverre |