Package Managers : my utopia specifications

[lalebarde]

Hi all,

These last years, I read many litteratures on portage and alternatives, and configure, scons and co. All that is hard stuff for me. So, I would like to make here my wishes for a new kind of installer.

EDIT : Following further remarks, I put in small olive fonts irrelevant things you may not consider.

First of all, what I see is:

- Every open source project is proposed with build instructions, based on configure, scons, or other. This involves manual install.
- Linux distributions "packetize" these projects with precompiled binaries or provides scripts to build them from sources. I will use "packet" and derivates for both in the following. With all pros and cons, anyway, the maintener have perpetual work to keep it up to date.
- Every distribution has their set of projects "packetized". If a project is not in the distribution set, the average user won't use it. If he can do it, he will either install it by hand or make the packet. But he seldom have time for it and would have appreciated to have it available. Most users are only users.

My naive spec (I can afford writting stupid things since I am not a software guy, just an amator with little time for it) :

- Have a project builder from project sources with no need for package nor script nor config file - so no ebuild.
- When a project of version n has been build on an arch, it is registered in a tracker server and available pear to pear or on servers for other users.
- The installer checks the tracker before building the project, and fetches the arch optimized binaries when available.
- Options to authorize to be a pear server for projects builds on one's machine.
- Options to check available binaries for one's arch or full build anyway.
- All good stuff from portage : sandbox, useflags, features.
- Dependency full resolver, including circular ones, by building intermediate versions, and/or partial builds.
- A distro specific config file to put files in right paths.

The design would take into account :

- A source parser and algorithms/euristics that makes itself the makefile, ignoring the one of the project.
- Other intelligence to detect optional sources, and/or providing usefull patches to make them optional, in order to be able to fit USE flags philisophy.
- In case of multiple choices, build in different sandboxes the different builds and test them. Keep the ones that works with the tests. If several candidates remain, make a report and output select keys. User can then eselect the build to use, keep only the selected build or some of them.

Advantages :

- All projects available for any distro.
- All projects optimized for one's arch and use.
- Quick install thanks to arch/version binary builds exchanges and/or repositories, and compatibility with all distros. EDIT : As explained by Ant_P and john_r_graham, it is seldom you will find somebody with the same arch and use flags. But we may also consider corporations.

Artificial intelligence would be usefull of course, but I think this is not utopia, and that we have the technology to do it.


Your thoughts.....

[Ant P.]

Um, you're proposing precompiled binaries for everyone that allow useflag-like functionality?

Not possible. A package with only 10 flags would need 1024 compiles per arch. GCC has 24 of them, not counting what seem to be arch-dependent flags - 16 million.

[John R. Graham]

16 million times the number of target architectures times the reasonable optimization choices times...well, you get the idea.

- John
_________________
I can confirm that I have received between 0 and 499 National Security Letters.

[lalebarde]

ok, so let's abandon the "pre-compiled" things, and let's consider the other stuff.....
I have edited the former post to put in small olive fonts the pre-compiled ideas.

EDIT : I have turned my mind back : "pre-compiled" projects may be usefull for corporations.

[lalebarde]

I asked tho Moderator to put this thread in OTG instead of US. Sorry for the unconveniance, but that was a mistake from mine.

[Earthwings]