A Starkit , where "start" is derived from STandAlone Runtime, and which was first devised by jcw, bundles into a single file a directory containing Tcl scripts, platform-specific compiled code, and application data. A starkit can then be appended to a compatible basekit to form a single executable file, which is a modified tclsh called a kitsh, that automatically mounts the virtual filesystem in the accompanying embedded starkit. This filesystem may be writeable, allowing the starpack to be modified at runtime as data is saved by the program. A single starkit is bundled with a basekit into a separate starpack for each desired target platform.
A kitsh is the Tcl interpreter that automatically mounts the virtual filesystem provided by a starkit. There are three ways to run a starkit:
what we have here is simply a Tcl version of Java ".jar" files, with the runtime environment optionally bound in. ... As for "simply jar files": not quite, SD's are r/w with transaction-safe commit/rollback because there's a database engine underneath. That means they can also store config info, update themselves, and store extensive datasets. -jcw
Starkits that have been published by various parties:
A basekit is not strictly necessary to execute a starkit. To execute a starkit with any tclsh:
See also: Running starkits without Tclkit
Could someone list here the various starkit namespace procs that are available for use?
For instance, there is:
package ifneeded Mk4tcl 220.127.116.11 "package require starkit [list starkit::pload $dir Mk4tcl Mk4tcl]"
- starpack when run from the binary/exe - tclhttpd when run from a tclhttpd (proc Httpd_Server must exist) - plugin when variable ::embed_args exists - service when variable ::tcl_service exists (Windows only) - starkit when called via tclsh myapp.kit - unwrapped when called via tclsh myapp.vfs/lib/app-myapp/myapp.tcl - sourced otherwise
David Zolli mentioned on the starkit mailing list these steps for adding encodings (since tclkit only has a subset of the encodings that normal Tcl has - one of the (few) differences between Tcl and Tclkit...)
I've tried and tested this, and it's works:
Then, these encoding are available in you starkit/starpack and your starkit can be launched with a 'standard' tclkit.
HaO: I use the following way, which is more natural for me:
Starkits were originally called Scripted Documents
When storing Starkits on a MacOS machine, to what are people generally setting the type and creator values?
Recently while playing with kitten, I was seeing an error when I attemped to package require one of the extensions, but was getting an error. I mused to jcw that it would be nice if there were a way to cd into the directory where tclkit was reporting a problem to browse, and he replied that I should be able to do just that. After some experimenting (and a false trail due to my forgetfulness) I found this indeed works:
$ tclkit % source /home/lwv26/kitten.kit % cd /home/lwv26/kitten.kit/lib % glob * Class1.0 ClassyTk1.0 Extral2.0 Mpexpr10 Tktable2.7 app-kitten ascenc-0.11 autoproxy autoscroll blowfish-0.10 bwidget-1.3.0 cgi1.6 dyncall-0.11 efftcl-1.0 expect-5.31 gbutton gk5.1 hexdump-0.10 ihash-0.11 itcl-3.2 itk-3.2 itk-3.3 iwidgets3.0.1 iwidgets4.0.1 lzrw1-0.10 mathf-0.11 mclistbox-1.02 md5c-0.11 mentry2.4 mkWidgets1.3 mvec-0.12 narray-0.81 scratch-0.10 supertext-1.0.1 tablelist2.7 tcldom-2.0 tclexpat1.1 tcllib1.3 tclsoap1.6.5 tclx8.4 tclxml2.0 tdom0.7 tix8.2 tkhtml0.0 tls-1.4 wcb2.8 wikit %
This may help you when attempting to debug strange errors.
Also, note that when I tried doing ls or exec ls, the ls command did not see the directories, etc. within the lib. This seems to me to be bad news for someone wanting to merge using exec, open |, or other external calls with VFS mounted filesystems.
(See VFS, exec and command pipelines for discussion on this issue; if someone wants to contribute the time/code, at least some sort of functionality ought to be possible and useful there).
or any other starkit, and the starkit executes. No requirement for a Tclkit. The Tclkit, of course, is still a superior distribution mechanism. Hopefully there will be tools one day that one can use to generate a starkit based on the Tcl environment available and the script(s) being wrapped.
Dossy: Hey, is there a standalone Starkit package out there? One that you could 'package require starkit' within a vanilla tclsh, and then go sourcing your Starkits? I was hoping to start using Starkits under AOLserver, so "use Tclkit" is not an acceptable solution, here. So, is there a starkit.tcl and corresponding pkgIndex.tcl that one can drop in their $TCL_LIBRARY and just "package require starkit" with?
LV: Hmm - let's think this one through. what is the difference between a tclsh and a tclkit? A package that resulted in the difference being applied to a tclsh is what we are talking about. I presume that one difference is the set of packages that need to be package required. What else?
jcw 2003-09-17: That's what ActiveTcl does. There is a package "starkit" and you need things like Metakit, trf-or-zlib, memchan-or-rechan, and of course the tclvfs package (C and scripts).
LV: JCW, I am trying to locate this package. I downloaded the latest Solaris Activetcl - 18.104.22.168. I install it. I then type this:
$ gfind . -name '*tarki*' -print ./lib/vfs1.2/starkit.tcl lwv26awu (2044) $ pwd /tmp/ActiveTcl lwv26awu (2047) $ gfind . -type f -print | xargs grep -l -i starkit ./lib/ppm/log/activetcl/install.log ./lib/vfs1.2/starkit.tcl ./lib/vfs1.2/tclIndex ./lib/vfs1.2/pkgIndex.tcl ./MANIFEST $ /tmp/ActiveTcl/bin/tclsh % package require starkit 1.2 %
That starkit is the vfs 1.2 code. To end up with the same environment as a tclkit, I presume that I need to do some other package requires.
When I start a tclkit, and do a package names, I see:
rechan pwb zlib vfslib vfs::mk4 mk4vfs vfs Mk4tcl Tcl
When I start an ActiveTcl tclsh, and do a package names, I see: ActiveTcl Tcl
Inside the ActiveTcl tclsh, I see:
% package require rechan can't find package rechan % package require pwb can't find package pwb % package require zlib can't find package zlib % package require vfslib 1.3 % package require vfs::mk4 1.9 % package require mk4vfs 1.9 % package require vfs 1.2.1 % package require Mk4tcl 22.214.171.124 % package require Tcl 8.4
So, are rechan, pwb, and zlib not important?
Dossy: Is there an easy way to "reload" a Starkit at runtime? In other words, suppose I have a long-running (daemon) Starkit. I want to modify some code, re-wrap the Starkit, then signal the currently running process to reload itself (or at least, its code) from the Starkit. I suppose I could mount the VFS and just 'source' the necessary files (or 'package forget' and 'package require'), but I'm wondering if this functionality already exists. Thanks.
jcw 2003-09-17: There's no magic in starkits, it's simply a VFS-mounted area. You can umount, remount, etc. You can even - with care - leave mounted, close the MK db, and re-open the MK db (but there's some caching going on in the mk4vfs.tcl script). Personally, I'd not go through this trouble: hook to the running process, and have *it* write the modified files into the starkit it is holding open. Compare it to a Unix mount: to change a file on the disk, you don't unmount, change it some other way, and remount - you simply write while it's mounted. If more complex scenario's are needed - why not run unwrapped? I mean, what are you trying to accomplish here, in terms of a higher-level perspective?
LV: The impression I got from reading Dossy's description was a scenario like this. However, I'm certain Dossy can explain it better if I am off base.
I have a starkit that I launch on a regular basis. I discover it has need of some modifications. I modify the file that I launch - this way, next time the changes are in place. Now, I want the running process to reflect the new version as well. I'd rather not take the process down - if it is a web server, I'd possibly terminate a user's session. What I want to do is signal the running process in some way that it would update itself when convenient.
Dossy 2003-09-20: Jean-Claude, yes, I could run unwrapped. But, Starkits are convenient: after developing, I can 'sdx wrap' all the code into one file, and deploy it where it needs to run. One file to push around, instead of a tarball that needs to be untarred before it can be run. Larry, you had the right idea. Implementing some kind of service as a Starkit, such as a webserver, where restarting the process is a nuisance (in my case, the process has persistent TCP connections to other servers -- process restarts means re-establishing those network connections). So, it'd be nice to be able to just modify code, re-wrap a new .kit, copy the new .kit over the old .kit then send a HUP signal to the running Starkit to go and reload any changed files (using 'source', perhaps?) ... of course, it'd mean writing your code in such a way that it could tolerate a warm restart, but that's the price we'll have to pay to get this kind of functionality, I guess. Or, we'll just have to keep hard-restarting these Starkits to get changes to be picked up, which is annoying but acceptable. But, I think a dynamic update of a running Starkit is a neato feature. :-)
jcw: Fair enough... so why not unmount/remount to pick up the changes?
Dossy 2003-09-28: Is there an easy way? If I sdx wrap foo.vfs while tclkit is running with foo.kit, it causes a Bus Error. If I vfs::unmount foo.kit at runtime, then sdx wrap foo.vfs then try to starkit::remount, on the remount it gives an error along the lines of no such mount /path/to/foo.kit. I can manually package require vfs::mk4 then vfs::mk4::Mount foo.kit foo.kit ... but that seems klunky.
What would be nice is a starkit::rewrap that did the unmount, wrap, and remount all in one shot. Hmm ... in the meantime, is there a hook into sdx wrap from within tclkit or must I exec the 'sdx' Starkit? source sdx.kit expects args - other than clobbering $argv, is there a better way of doing this? vfs::mk4::Mount /path/to/sdx.kit sdx.kit -readonly then source sdx.kit/lib/app-sdx/wrap.tcl still requires clobbering $argv. Guess I'll just exec, for now.
Thanks for entertaining my foolish idea, though. Starkits are still immensely useful, even without a really easy on-the-fly code update mechanism. :-)
VL: As it is not mentioned on the sdarchive page I thought I would add that on windows, you can of course as usual "register" the extension .kit, as an alternative to writing bat files, hence enabling the windows-double click behaviour.
One would presume some a similar setup could be made on MacOS, or for that matter the X Window based platforms, in relation to the appropriate GUI file manager.
If someone is wondering about using starkits for tclblend applications - that's really going to take some work. Tclblend requires one to use a threaded tcl - and tclkit doesn't by default build with threads. Then, to make use of tclblend, there's a start up 'wrapper' that I seem to understand one should use to start up the threaded tclsh, that ensures that various environment variables, etc. are set appropriately.
One of the hardest things to wrap are applications which make use of Tcl extensions which are dependent on compiled libraries (regardless of whether they are written in C, C++, etc.). There are a variety of issues that need to be resolved by these things. Is this part of building starkits documented anywhere?
MDD: I've noticed a strange problem with tclkit 8.4.6 under QEMU.
I've got a Kanguru Zipper 4GB USB hard drive
that I've put a 2gb ext2 disk image onto. Using QEMU , I booted up an .iso image of Damn Small Linux (dsl083) and then installed it onto the ext2 disk image file. From any Windows machine, I can then launch into DSL, running in the QEMU emulator window. I've installed a bunch of apps, including Firefox and Open Office, with no problem. They all seem to work fine, even though the execution is slow due to the emulation overhead.
OK, so far everything's working great. I've got this really cool little Linux machine I can carry around in my pocket. I've even set up autorun.inf to automatically launch the whole thing whenever the drive is plugged into a usb port. Very cool. One small weirdness is that gzip errors out whenever I try to gunzip something, though.
My main problem is that, even though both tclkit and dqkit run, and load up Tk without a hitch, neither of them can run Starkits. They can't seem to find main.tcl. This happens with all Starkits, even the most simple ones.
Any idea what might be wrong? I'm guessing there must be something about the QEMU emulation that interferes with the decompression that must occur with both gzip and Tclkit's vfs code.