Slashdot Mirror
Linux Foundation Promises LSB4
gbjbaanb writes "Ever thought it was difficult to write software for Linux? For multiple distros? InternetNews reports that the LSB is making a push for their next release (due out later this year) that should help make all that much easier. Although the LSB has not lived up to expectations, this time around Linux has a higher profile and ISVs are more interested. This is to help persuade them to develop applications that will run on any LSB-compliant Linux distribution. If it gets adopted, LSB 4 could bring a new wave of multidistribution Linux application development. 'It is critically important for Linux to have an easy way for software developers to write to distro "N," whether it's Red Hat, Ubuntu or Novell,' [said Jim Zemlin, executive director of the Linux Foundation.] 'The reason you need that is because we don't want what happened to Unix to happen to Linux in terms of fragmentation.' The LSB defines a core set of APIs and libraries, so ISVs can develop and port applications that will work on LSB-certified Linux distributions."
The Linux Standard Base, or LSB, is a joint project by several Linux distributions under the organizational structure of the Linux Foundation to standardize the internal structure of Linux-based operating systems. The LSB is based on the POSIX specification, the Single UNIX Specification, and several other open standards, but extends them in certain areas.
http://en.wikipedia.org/wiki/Linux_Standard_Base
The quote in the summary reads:
'It is critically important for Linux to have an easy way for software developers to write to distro 'N,' whether it's Red Hat, Ubuntu or Novell,"
Personally (as a Linux on the desktop user), I'm a lot more concerned about easily acquiring installing software, than whether it has problems with my distro. For the most part I can get software to run, but it can be a huge pain in the butt. I wish LSB would focus on extending and standardizing package formats and creating advanced standards for package managers to simplify that part of my workflow. I never wonder, "will this run on Ubuntu," so much as "which package format is this in, or how hard is it going to be to compile and update."
UNIX fractured into a large number of incompatible variants including BSD (fractured further into Open/Free/Net BSD), Solaris, IRIX, HP-UX, SCO Unix, etc., etc..
See this graph for more information.
"The reason you need that is because we don't want what happened to Unix to happen to Linux in terms of fragmentation."
What makes you think what happened to UNIX was bad? It's called evolution. Things change.
Let me remind you, my friend, that evolution means SUCCESSFUL ADAPTATION to an environment. What happens when a change (mutation) results in inadaptation? Extinction.
Evolution refers to a species. But in Linux what we have is not a single species, but a genus (a set of different species): Redhat, Debian, etc. "DNA" recombination is impossible in these. The resulting software would be inoperable.
LSB4, hopefully, will be a further step in the evolution of Linux: The convergence to a single species that will be able to share one single configuration.
In other words, yes, change is necessary, but there needs to be a period of stabilization. Just as stable/unstable releases in software. And LSB is providing this stability. LSB is, in fact, evolving.
maybe you mean something different, but I'm not sure how your statement relates to this issue. Afaik the LSB is about standardizing directory layouts and configuration files, and while sure under the GPL any linux distro CAN be made to follow those guidelines, almost none of them DO, so the difference between nonstandardized linux systems and nonstandardized UNIX systems is a philosophical one and not a practical one.
(Although on Linux it's a fair bit easier to remedy)
Yes, that's kind of the whole point of the LSB.
Customers choose OSes based on many criteria. One of them is how much of the software they need will run on each platform. Now, this is rarely actually determined by the quality of the platform -- it's mostly a question of which platforms were already popular enough to be targeted. In theory, LSB will make it easier for new Linux-based OSes to run existing software, and will make it easier for ISVs to write software for Linux-based OSes in general.
Those OSes can then compete on more interesting metrics like performance, stability, scalability, price, and quality of support. How is this not a good thing?
LSB4 is all very well, but if RHEL does not follow (does anybody really think they will?) it will not amount to a hill of beans.
The masses are the crack whores of religion.
It relates to his statement that I quoted.
That shows how clueless he is regarding the history of *nix.
It was the various PROPRIETARY licenses that caused the fragmentation because an improvement made by HP had to be specifically licensed by Sun to be included in Solaris.
But with the GPL, the improvements made in one fork are available to ALL forks.
Therefore, the fragmentation will not happen because if a feature is worth it, it will be ported to the other forks. Without the need to coordinate licenses with HP or Sun or anyone else.
The GPL rocks.
Other than eliminating conflicting directory structures, the most important standard for Linux distros to completely unify would be a single API to data protocols and MIME types. Like the one FreeDesktop.org has managed to sync (in principle) between GNOME and KDE Desktops, but for all distros (including servers).
A registry of which app to hand off a URL to given its protocol part, to retrieve the data. A registry of which app to hand off the data to once it's retrieved. Different data handler lists for displaying, editing or executing (the usual Linux RWX modes) the content, depending on the use case triggering the registry access. The registries could include prioritized lists of different apps, depending on user selection or settable default preference. And of course any single app could be registered to either registry, in any mode it will function properly.
Then the OS is performing its main task of connecting processes to the hardware and to each other. In a very simple and clear architecture. That every single app can use, without having to anticipate how the other apps will agree with it.
If LSB4 can pull that off, using the existing attempts as a starting point, it won't just make a unified Linux target for developers across distros. It will make LSB4 itself more quickly and completely adopted, because its benefits will be so compelling.
--
make install -not war
POSIX has multiple components -- kernel APIs, command line utilities, shell scripting, libraries, etc -- so there's more too it than just the linux kernel.
Do you even lift?
These aren't the 'roids you're looking for.
UNIX fragmentation wasn't caused by anything other than all the proprietary, incompatible licenses. Whenever Sun made an improvement to UNIX, HP couldn't simply adopt it like they can with the GPL. With the GPL, if I take an OSS program and fork it, and change it radically, the original creators of the software can always add my changes back into the main branch. And yes it would be bad, if you had to write a program, say an HTTP server, you had to test it on every Unix imaginable, today, just release the source, package an RPM and a DEB, and it will be ported to the rest soon enough.
Taxation is legalized theft, no more, no less.
But you see, they dont want to write for distros foo, bar, etc... they want to write an app for linux.
They dont want to "collaborate" with dozens of distros, all of which will tell them that "in our distro, the proper way of how to do this" is different than the other ones...
No we understand the GPL, but it has very little to do with the subject, namely that regardless of open vs closed, some distros remain incompatible with each other in small but significant ways.
If there is to be a stable platform to target with Linux, that crap has to stop. Simple being GPL software means very little toward that goal if distros continue to be arbitrarily different and the situation never really resolves itself.
Well, from what I've seen, /usr/local and /opt were reserved for the local sysadmin to manage, and the package management system generally stayed away from that. This meant that custom software and distro packages didn't have file conflicts.
Now, I like the way that works, a lot. But I don't have any objections against further partitioning of that scheme.
For large sets, this will be our guide even unto death, for the LORD will work for each type of data it is applied to...
Classic zealot response. Pretend the entire world is moving to GPL-only software and neglect to address the concerns of anyone who disagrees.
Wrong.
Any given distro will have to make a choice of what modules each program should support, this means even as a PHP programmer you have no guarantee your software will work with default installation of PHP under a specific distro.
I have been mentioning this for years
Yes, I've seen you post here often!
LSB isn't needed, Linux distros just need to implement 16-year-old standards. I think most do put env in the right place, although some also put it in /bin (which should only contain binaries needed to boot the system in single-user mode).
I am TheRaven on Soylent News
What makes you think what happened to UNIX was bad? It's called evolution.
I COMPLETELY DISAGREE WITH YOU! Linux, SunOS, Solaris (Not an evolved SunOS!), BSD, etc were magically created out of... erm.. random bits generated from some pseudo-random generator in DigiGod's image. Proof? DigiBless.com
</satire>
One piece of this was that simple commands had different names and different options depending on the variety of UNIX. I am talking things like lp vs. lpr, options for ps, and so on. Also, some things were hidden in really weird places (X on Sun is/was under an "openwin" directory, I think). Writing cross-platform scripts is a pain when you first have to test for the OS and then redefine command names, options, and paths accordingly. In theory, under the LSB you always know where commands and libraries are.
Every project needs a code name. For this, I propose "Bullwinkle", and their slogan can be "This time for sure!".
Instead of trying to make all the distributions the same, why don't they make a library that abstract away the difference?
Example: If my program need to link to a ssl library(Such as openssl), version 2.3 or newer, I should call a function .so file, or null if the file is not installed. There could then be a
findLibrary("ssl",2,3) which would return the path to the needed
function to also ask the os to install the needed library if it were not there.
Each linux distribution should then implement the library in a way, so that the Redhat version, might forward the call to rpm, while the debian version of the library would query the dep database insted.
And instead of the infinite debate on /opt vs /usr/local the program could just call getPathForUserInstalledSoftware();
And getDefaultCompilerPath() instead of the current autoconfig hack.
Then a linux standard base, would just be a specification of the needed functions in LinuxStandardBaseLibrary.
And we would newer have to use the autoconfig hack. (The library might ofcause also be implemented on Solaris, and maybe even cygwin/windows)
env should always be in /usr/bin. This will work on any POSIX.2-compliant system:
POSIX.2 doesn't even mention /usr. The location of env is not standardized.
COO: so, we've developed a new widget, brilliant. When can we ship it.
DEV: today, its all tested. I've run it myself on my Debian box.
COO: sure, it'll run on Customer B's Redhat environment won't it?
DEV: Ummmm.. well, not straight off.
COO: ok, so what do we need to do to make it work?
DEV: well, alter a couple of paths, recompile, change the dependancy for package Z and build a rpm
COO: and how long will that take?
DEV: a week, maybe 2 with testing.
COO: so, that's 2 weeks development costs on top of what we've already spent. We wouldn't have this kind of problems with Windows!
and I'm not trolling either - standards are good, making extra work for yourself for no good reason is bad. Its not as if a common directory layout, ABI or programming libraries need to affect open source linux in any way. The Kernel is a standard, and no-one complains they don't have a choice of kernels to develop against.
The first way is with a command like 'sh foo.sh'. sh will then read foo.sh and execute each command in it in order (if it's not a shell script, it will hopefully read the magic number and run it).
The second way is to just exec() it. The loader then reads the first few bytes of the file. This tells it what type of file it is. For ELF files, they will be ".ELF". For Mach-O binaries, they will be 0xFEEDFACE or 0xCAFEBABE, depending on the architecture. For scripts the first two bytes will be "#!".
If the loader encounters "#!" then it will read the rest of the line execute the specified command (and arguments) and pass the file to it as the last argument. You can see this in operation with the following script:
If you have a shell script that needs to run with the standard POSIX shell, then there's no problem. You just specify /bin/sh after the #! and it's fine. But what happens if you want to use some bash-specific features? On GNU platforms, bash is the default shell and /bin/sh is a hard link to bash, so it's in /bin. On other platforms it's a third-party thing and so will be in /usr/local/bin or /opt/something. This is where env comes in.
When you specify "#!/usr/local/env bash" you can safely hard-code the path of env, because POSIX defines where it is. Env then looks up where bash is and execs it with whatever command line arguments it was given. You can see this in action again like this:
If you just run 'env' from a command line to inspect the environment variables, you are most likely just calling a shell built-in command, which lists the things passed in to the third argument to main() and any set since the shell started. Env, however, can be used when you are not launching from a shell. If your program wants to run a shell script, you can just vfork() and exec() it, and the loader will find the correct interpreter. You could always inspect the environment yourself, but having every app do that whenever it needs to run a script is quite silly (especially since it means that the app also has to know the difference between a binary and a script and so on).
Some people still have a habit of hard-coding /bin/bash, and this is probably the kind of crap LSB will encourage ('bash is always in /bin on Linux, and I don't care about other platforms!') but the correct thing to do is use env.
I am TheRaven on Soylent News