Slashdot Mirror
A Windows-Based Packaging Mechanism
FishWithAHammer writes "As part of my Google Summer of Code project, I'm working with WinLibre to develop a Debian-like software download system for free/open source software on the Windows platform. My reasoning is that open source software suffers from poor presentation. Most computer laymen, even those aware of open source software, often don't have any idea how to go about looking for it, but would use it if it were easier to access. What I have proposed is both a Debian-style packaging mechanism (capable of using Windows Installer MSIs or not, as the user wishes) and a software 'catalog' that takes the best aspects of Synaptic and Linspire's Click-N-Run system. Seamless, simple installation and removal of programs in as straightforward a way as apt-get (there will be a command-line tool as well). I'm posting to Slashdot to get the ideas of you lot who, while you may not be the target audience, can certainly provide insights that can be of value." Read on for more of this reader's ideas and questions.
There are areas that I'm personally not familiar with, and while I have done some research I would like the opinions of Slashdotters on some others. While at first I intend to set it up so that WinLibre (and I) run only one repository, I am curious as to how this sort of tool could be most useful to network administrators. Customizable repositories will be available; the code will be under the GPL, after all, so it'd be a little hard for them not to be available.
I'm also interested in the ideas of those who might be in a position to roll together packages. I intend to package a number of open-source language interpreters with the core software to allow special pre- and post-install scripts, as well as removal scripts. C#Script, Perl, and Python are definites, as is a Cygwin sh interpreter. We will have some program requirements — chief among them that no registry changes may be made by the program — but some of them, I fear, will require some flexibility; some programs really do require a way to edit the registry, for example, and I am considering offering some sort of tracked way to make registry changes so they can be rolled back on uninstallation of the program.
I'd love to hear what Slashdotters think of this. Think of it as a wishlist, but you don't get any damn ponies.
Ed Ropple (FishWithAHammer)"
There are areas that I'm personally not familiar with, and while I have done some research I would like the opinions of Slashdotters on some others. While at first I intend to set it up so that WinLibre (and I) run only one repository, I am curious as to how this sort of tool could be most useful to network administrators. Customizable repositories will be available; the code will be under the GPL, after all, so it'd be a little hard for them not to be available.
I'm also interested in the ideas of those who might be in a position to roll together packages. I intend to package a number of open-source language interpreters with the core software to allow special pre- and post-install scripts, as well as removal scripts. C#Script, Perl, and Python are definites, as is a Cygwin sh interpreter. We will have some program requirements — chief among them that no registry changes may be made by the program — but some of them, I fear, will require some flexibility; some programs really do require a way to edit the registry, for example, and I am considering offering some sort of tracked way to make registry changes so they can be rolled back on uninstallation of the program.
I'd love to hear what Slashdotters think of this. Think of it as a wishlist, but you don't get any damn ponies.
Ed Ropple (FishWithAHammer)"
C:\>apt-get install bsod
That's great of course, but it's the community and a selection of packages with mutually consistent packaging metadata which make systems like Debian and their derivatives so popular. The packaging system itself is an enabling technology.
Do not let this become a new attack vector.
--- These are not words: wierd, genious, rediculous
I would say the big thing that I would look for in such a product would be a consistent (or even better, non-existent) use/removal of registry entries. I have dealt with so many so-called "professionally" done software pieces that upon uninstallation would leave several dozen registry entries. This seems terribly unnecessary, and if the so-called apt-get method could circumvent the registry (much like the run from USB flash drive programs) altogether, or at least make it a sure-fire thing to remove, instead of wipe-and-pray.
Good on you for trying to better the system man, I wish you the best of luck!
For user-specified (or multiple fallback) repositories, you need nothing more complex than reading your base path(s) from a config file. Prepend that address to every file you download, and it will all go well.
For the bigger project, basically you just need a set of per-package install/uninstall scripts that check for dependancies (or no-longer-needed dependancies on uninstall), do their thing, and write themselves to a standardized catalog of installed software. Whether or not you can adapt Windows' list of such software, and the MSI interface in general, to your needs, I can't say offhand. I would think you can at least list the package therein, but I don't think that handles dependancy information quite as elegantly as you would want.
I see the biggest problem you'll have as coming from the poor regression testing done for Windows ports of FOSS - You may well need multiple (version-specific) instances of some dependancies installed at the same time, for different packages that use "working until version 2.8.10.4" features (or more of a nightmare, "working until KB935356").
Overall, I wish you luck with this. I think the Windows world has needed something like apt-get (with a mind-numbingly simple GUI) for a loooooong time.
That packages provide functionality. This is already done in the form of virtual packages like web-browser, but I'd like to go further.
.pcx file, and have no clue what is it, what can I open it with?
.s3m files?
.s3m files) and "fully implements" (can create .s3m files).
For example, the current system is that OO Writer and KWord are in the "word processor" category. But what if I want something that can open AmiPro documents? What options do I have there? That's generally not included anywhere in the package's description.
I found this weird
Or, what music player has the ability of playing
What mail clients can I choose from if I'd like both NNTP and IMAP support?
What programs are available that do some function that is related to an HP nx5000 laptop? (this would match programs controlling LCD brightness, support for the onboard bluetooth, etc)
A nice thing would having these capabilities roughly grouped as "can access" (can play
So, in theory this should work with ReactOS when they are both finished, right?
I hope you're planning on making it interoperate with the cygwin packaging system. Cygwin's a great piece of software which is, IMO, let down by its obscure and difficult-to-use setup program. A new, friendlier way of installing and updating cygwin components would be a great asset. And if it worked with other OSS stuff as well, that would be a huge asset.
.exe that doesn't require your system, but which can interoperate with your system easily -- perhaps by having a version of your system that can wrap up a package with a copy of the relevant parts of itself in a .exe file.
One thing I would suggest is that you make it easy for somebody to package a standalone
It is not only the programmers' fault, though. Far too few users bother to suggest interface simplification,or even know how to advocate it. Merely complaining will not work - developers need to be shown that it can be done, and how, by means of mock-ups or illustrations. A few innovative user interface interested users could do wonders for many projects simply by drawing new user interfaces and submitting them to various free software projects, asking if they are interested in going a few rounds of design iterations with them. Often an outside eye, and interest in doing some adapting from both sides, is all that is needed.
You may want to look at wpkg (http://wpkg.org/)
It is a windows package management system based on dpkg.
We use it at work and it appears to work fairly well. Although I don't know for sure, as I'm not the PC admin and I don't run a Windows desktop :)
I just get to hear him saying how much easier it is to manage the PCs with it.
Ever stop to think
They all handle updates as well, so you have a central place to keep all your entire system patched. For example, when a vulnerability is discovered in a core library (libz, or linpng have been recent examples), you need to go through your system checkiing that every application which uses one of these libraries is updated. This is almost impossible on Windows, but automatic on systems with package managers.
The first issue that occurs to me immediately is that Windows has no single suitable native package management system that you can hook onto. Because of this, program installations tend either to (i) include whatever prerequisites they need and check whether their installation is necessary; or (ii) list the prerequisites in the installation instructions and leave it up to the user to ensure they are satisfied. Now, you might say that the whole point of the project is to resolve this, but I think you are going to run into licensing problems when you try. Let's say a particular open source product relies on .NET Framework 2. Are you then going to include .NET Framework 2 in your repository? Are you going to download it from Microsoft, using Microsoft's Download Center as a kind of adjunct repository? Are you going to talk to Microsoft to see if they will cooperate in working out a solution? This seems hard.
I do think that a single starting point for finding quality open source solutions on Windows has merit. Right now there is a bewildering mass of products out there, and no easy way of sifting the gems from the dross. If nothing else, you might be able to provide a good menu of open source products that are deemed worthy of consideration.
Good luck!
One thing I think shopuld be considered from the beginning is how to handle multiple archives, which may be independently maintained. Sure, the basic operation is simple: You add a new URL to the list of archives to search, and then you can see the contents of those archives. However that's not all there is to archives:
1. How do you find additional repositories?
2. How do you find out if a given repository is trustworthy?
3. What to do if several repositories contain packages for the same application or library?
4. What about version inconsistencies?
Points 1 and 2 can IMHO be (mostly) solved together through a "repository web": Repositories not only contain packages, but also links to other repositories. Those links should also be rated, so you get a web of trust for repositories: You can mark several "root repositories" as trusted or untrusted (those settings should, of course, be user-changeable). Then trust would "propagate" through links marked as trusted, or "anti-propagate" through mistrust-links. One could even imagine "repository hubs", repositories which don't contain files, but only links to other repositories together with trust ratings. It might also be a good idea to have several trust ratings for the contained files, and for the contained links (after all, you can well imagine an excellent file repository where the maintainer isn't able to accurately rank the trust on inter-repository links).
For points 3 and 4 I don't have a suggestion right now, but they definitely should be considered (note that separately maintained repositories will almost certainly cause inconsistencies at some point).
Of course you can just pretend that there will always be only one repository, or that all repository providers will work together to avoid inconsistencies, but I think that's not really a good idea. Additional independent repositories will eventually come (assuming the project is a success), and therefore the problems caused by those should definitively be anticipated, even if originally there's only one repository.
The Tao of math: The numbers you can count are not the real numbers.
I have been thinking about this recently.
I have lots of applications, both OSS and commercial, that have some kind of update system built in - the application checks for an update when you start it, for instance, or when you select the option from the help menu. In fact it is getting to the stage where practically every app. has this.
What I would like to see is a single open method of doing this which could work for all applications (so even commercial software providers could opt into it if they wanted), which would be simple and secure. It would be great to have a single application open that ran at start-up that said: "The following applications have updates available:" and then lists the applications, and two buttons "Update all" and "Advanced" which would allow you to see details about the updates and select just the ones you want.
For instance on my Mac I have:
1) The Official Apple "Software update" that updates OSX and Apple Apps.
2) The Adobe updater for Photoshop, Dreamweaver etc.
3) The Firefox/Thunderbird updater
4) Dozens of updaters for individual apps like TextMate and OSS software
5) Updaters for OSS packages (Fink/darwinports)
(Yes, I know about the App Update widget but that only addresses part of the problem, and it does not provide a technical solution that can be used across platforms and projects).
And on Windows, I have the same kind of mess of updaters.
I'm sure there could be a simple, elegant technical solution for this, a kind of RSS-type standard for application updates - you could then choose your prefered updater just as you can now choose your preferred RSS reader.
Your complaint boils down to "some people make bad packages", which occurs on Linux as well, and is just the nature of software to be imperfect. I cannot count the number of bugs or non-working setups I've tracked down to bad packages, and even better, in the Linux world fixing such a bug once doesn't make it go away - it'll be repeated in 3 months time by a different distribution.
That would be nice, yes.
I thought this too.
But MSI doesn't do what the Linux/BSD packagers do. These packagers work by tracking every single file or update done to the entire system. Then they track dependencies between files and packages. They store all this in a database format, which allows you to ask questions like "what is every package that uses MSVCRT71.DLL? And "what will break if I update package GIF_VIEWER from version 1.0 to version 1.1?" They also manage side-by-side installs, provide a central repository for searching for packages and upgrades, and provide a safe digitally signed repository for applications.
This is one of the killer features of Linux that I miss on Windows. But I suspect it won't work for the same reasons it doesn't work on Linux. It's only useful if 100% of the applications use it. If any one of them doesn't, then the whole system can come crumbling down. But basically, it is a fix to DLL hell, so it can't make things on Windows any worse.
On a note of MSI, MSI may seem to do the above, but it doesn't. It's a packaging format, and it allows for install and rollback much like the Linux packaging systems do. But most of the time it is unrealistic to expect the repair/rollback/uninstall features to actually work. I've worked at a few companies who have made MSIs, and generally you take some other EXE or script-based installer, then you wrap it in an MSI and say you are done. You rarely use the actual MSI features because they are too complicated and the tools don't generall support them. And Windows installs are full of kluges like editing a registry key here, adding a shell extension there, etc. Things generally don't fit into the nicely packaged mentality.