http://blog.gmane.org/gmane.os.inferno.general hourly 1 1901-01-01T00:00+00:00 http://gmane.org/img/gmane-25t.png http://gmane.org http://comments.gmane.org/gmane.os.inferno.general/4996 <pre>Hi! I've implemented Limbo module for subj, available from usual place: http://code.google.com/p/inferno-contrib-tap/ There exists itslib(2), but there no applications which use it and I don't see how it can be used to make writing tests easy. In perl, there a lot of powerful modules which make writing tests really ease - they show differences between complex data structures, allow you to mock functions/modules, etc. - I suppose most of these things can't be implemented in Limbo, but probably there enough room for TAP module improvement anyway (for example, it's probably worth adding functions which assert exceptions will/will not be thrown by tested code). Sadly, but there some chaos with TAP specification and versions, but I think I implemented complete "TAP version 12". Now, as usually, I have some questions about this. :-) It should be possible to develop and run tests both in host OS and in Inferno. While it's straightforward to develop and run TAP-tests in Inferno, it still lack TAP consumer application like `prove` command from Perl module Test::Harness. This mean if you have a lot of tests it will be hard to quickly run them all and get nicely printed summary for all tests. I don't plan to implement it, both because TAP consumer is much more complex than TAP producer, and because I develop in Limbo from host OS and will run tests also from host OS. Running tests implemented in Limbo from host OS and making it possible to use `prove` command to run these tests require some ugly hacks. And I like to find more clean way to do this, if possible. Here is how I implemented it: - let's suppose your application live in /opt/your/cool/appl/cmd/ - tests will live in /opt/your/cool/appl/cmd/t/ - now, to make it possible to run `mk test` in /opt/your/cool/ you'll have to append this line to your mkfile's in /opt/your/cool/, /opt/your/cool/appl/ and /opt/your/cool/appl/cmd/ (if Charles accept issue 228 this will not be needed anymore): test:V: test-$SHELLTYPE - next, you'll need to include /opt/powerman/tap/appl/lib/t/mktest in your /opt/your/cool/appl/cmd/t/mkfile. the interesting part is: test:V: $TESTS.test-$OSTARG %.dis.test-Inferno:V: all $stem.dis %.dis.test-os:V: all %.t ./$stem.t %.t: ln -nsf runtest.sh $stem.t - and last, you'll have to copy into your t/ directory file /opt/powerman/tap/appl/lib/t/runtest.sh and update it for your configuration. here is the file: #!/bin/bash -i TESTDIR=/opt/powerman/tap/appl/lib/t/ DIS=$(basename $0 .t).dis emu-g sh -c "run /lib/sh/profile; $TESTDIR$DIS; shutdown -h" || : * you'll need to change TESTDIR, this is obvious * you may need to change `run /lib/sh/profile` command - this command expected to setup right environment to run your tests: bind host OS directories with source of your application to correct place in Inferno file system, load services needed for your test (ndb/cs, maybe mount remote ndb/registry, etc.) As result of this configuration, after `mk` you'll have something like this in your t/ directory: powerman&lt; at &gt;home ~/inferno/opt/powerman/tap/appl/lib/t $ ls -l total 24 -rw-r--r-- 1 powerman users 202 Mar 15 23:26 mkfile -rw-r--r-- 1 powerman users 422 Mar 15 14:22 mktest -rwxr-xr-x 1 powerman users 145 Mar 15 18:56 runtest.sh -rw-r--r-- 1 powerman users 98 Mar 15 19:02 smoke.b -rw-r--r-- 1 powerman users 335 Mar 15 23:26 smoke.dis -rw-r--r-- 1 powerman users 1707 Mar 15 23:26 smoke.sbl lrwxrwxrwx 1 powerman users 10 Mar 15 19:10 smoke.t -&gt; runtest.sh With this setup, you can run your test from Inferno as `./smoke.dis`, from host OS as `./smoke.t`, and also you can use host OS command `prove` in either this or upper directory, because it expect to find tests in files "t/*.t". So, we have several more or less ugly hacks here: * needs to create smoke.t symlink (only good thing is it created automatically by `mk`) * needs to copy runtest.sh into each your directory with tests and manually configure it (while it's possible to move TESTDIR out from this file - maybe to t/mkfile - tests for different applications still may need different setup after starting emu... maybe use some sort of t/profile file for this?) * there two hacks in runtest.sh itself: - it run bash with "-i" option (because without it after emu exists bash print ugly "Killed" message which clutter test output... but I'm not sure is any subtle side effects possible because of this) - it replace real emu exit code with 0 (because `prove` check exit code and think something is wrong with all tests - emu always exit with non-zero code) Also, this won't work on Windows, but I don't use it (for work) and so can't propose more portable solution. If there ideas how to run Limbo tests using host OS `prove` tool without so many hack - I'll be really glad to hear it! </pre> Alex Efros 2010-03-15T22:23:12 http://comments.gmane.org/gmane.os.inferno.general/4992 <pre>Hi! Is current limbo polymorphic syntax documented anywhere (man/pdf/url/post to maillist/irc logs)? Only documentation I was able to find is two posts: http://9fans.net/archive/2004/05/373 http://9fans.net/archive/2004/05/374 Is these posts meant to be used as main official comprehensive doc for current polymorphic syntax? Is current syntax stable, or it still may change without notice in any incompatible way? P.S. Maybe next one "Addendum to The Limbo Programming Language.pdf" should be released? </pre> Alex Efros 2010-03-14T04:49:12 http://comments.gmane.org/gmane.os.inferno.general/4991 <pre>Hi! The only way to deploy 3rd-party software in Inferno which I'm aware of is one used by mjl at http://www.ueber.net/code/. But it has some limitations: * if Inferno installed system-wide (not in some user's home directory) it will require root access to install such software using host OS mk or emu started by root to install using native mk * it's likely will be conflicts between different software providers in application/module names because everything is installed in same global directories (can be solved by using /dis/mjl/ and /module/mjl/) * it's hard to provide different versions of some module for different applications (he tried to work around this by adding version number to module name like "util0.m", but that's ugly) * it's hard to uninstall I wanna propose different way to do this, something like DJB's slashpackage: http://cr.yp.to/slashpackage/management.html * let's choose and create some top-level directory for all 3rd-party software like /contrib or /opt or anything else you like * all 3rd-party software will expect to find all it's files here: /contrib/SOFTNAME/dis/ /contrib/SOFTNAME/dis/lib/ /contrib/SOFTNAME/module/ /contrib/SOFTNAME/lib/ ... where SOFTNAME may be just package name like "httpd" while we've not so much software, but also can be "mjl/httpd" or "ueber.net/httpd" to guarantee uniqueness of names. * to install such package system-wide it's enough to do: # hg clone https://soft.name/ $INFERNO_ROOT/contrib/softname and there no need to additionally run mk (dis files in /contrib/softname/dis/ are already in place and ready to use) * to install such package locally for one of your project needs without having root access: $ cd $YOUR_PROJECT_DIR $ mkdir -p contrib/ $ hg clone https://soft.name/ contrib/softname $ emu ; memfs -b / ; mkdir /contrib ; cd $YOUR_PROJECT_DIR_INSIDE_EMU ; bind -b contrib /contrib * in same way (using single bind) application may override system-wide version of some package in /contrib/ by it's own version * if there some local modification needed for this package, it's enough to run mk in /contrib/softname/ directory after modifying files in /contrib/softname/appl/ and it will rebuild binaries in /contrib/softname/dis/ (both native and host OS's mk can be used) * update is ease (hg pull &amp;&amp; hg update) * uninstall is ease :) Packages implemented in this way should be used a little differently: * from limbo code: include "contrib/softname/module/softmodulename.m"; load SoftModuleName SoftModuleName-&gt;PATH; this will "just works" if contrib packages was installed locally (i.e. in contrib/ subdirectory of your project), but if it was installed system-wide you'll need to add -I/ for native limbo or -I$INFERNO_ROOT for host OS limbo. * from sh (you can define short alias using sh's functions): ; contrib/softname/dis/softapplicationname Implementation of such packages should be done in this way: * in contrib/softname/module/softmodulename.m: PATH: con "/contrib/softname/dis/lib/softmodulename.dis"; * in contrib/softname/appl/{lib,cmd}/softmodulename.b: include "../../module/softmodulename.m"; load SoftModuleName SoftModuleName-&gt;PATH; * in contrib/softname/*/mkfile all paths are relative The example of such module you can found here: http://code.google.com/p/inferno-contrib-hashtable/ </pre> Alex Efros 2010-03-13T08:01:09 http://comments.gmane.org/gmane.os.inferno.general/4988 <pre>When the Tk module was developed what was the reason for not including Tcl? Has anyone ported Python to Inferno/Limbo? Has anyone coded a TIL(threaded interpreter language, eg. Forth) in Limbo? Does anyone have a complete list of development and application software? This would help so we do not have to recode what has already been done! By the way, thank you for Inferno it makes using multiple processors/systems such a pleasure. Regards, Paul </pre> Paul Lycett 2010-03-07T14:59:31 http://comments.gmane.org/gmane.os.inferno.general/4983 <pre>Hi list, I am always looking around at different OSes and recently came across Inferno. I have read a lot of the overview docs and downloaded the source and got it running on my GNU/Linux box. I really like the concepts encapsulated in the model. My other love is building virtual reality worlds and it struck me that Inferno makes half of a really neat P2P VR system. I brain-dumped a page of my ideas here: http://home.exetel.com.au/impact/inferno.shtml and if anyone would like to make an opinion on my ideas (positive OR negative welcome as long as it is constructive ;-) I would very much appreciate it. I am not a real programmer, though I understand programming concepts and can do simple things in C, Java, SmallTalk, Python. At best I aspire to be a really good virtual world content producer, however if any real programmers thought this was a worthwhile and ultimately doable proposition I am able to drop one day paid work per week to allocate to learning how to contribute meaningfully to realising the goal outlined. Regards, LaeMing Wang DipIT B.Ed(IT) B.Teach(Primary) Australia </pre> LaeMing 2010-02-27T01:01:22 http://comments.gmane.org/gmane.os.inferno.general/4977 <pre>When starting auth/keyfs, it always (that I can tell) prompts for the key to use to decrypt the key database. Is there a recommended way of daemonizing this? I'd like to not have the keydb password sitting around unencrypted on disk, obviously. Anthony </pre> Anthony Sorace 2010-02-25T21:52:27 http://comments.gmane.org/gmane.os.inferno.general/4974 <pre>1. when comunicating with servers, do the clients use quid from styx r-messages, or the fid is enough? 2. how clients arranges so that fids are different? 3. how path paramerter from quid is generated? dorin </pre> dorin bumbu 2010-02-24T16:18:39 http://comments.gmane.org/gmane.os.inferno.general/4972 <pre>just checking: is there a script somewhere to recreate the missing empty directories, e.g. $SYSTARG/$OBJTYPE/lib ? </pre> roger peppe 2010-02-23T16:18:03 http://comments.gmane.org/gmane.os.inferno.general/4968 <pre>Hi! Is anybody knows why appl/svc/webget/webget.b use ∎ (U+220E) as delimiter between tokens instead of usual space? Space delimiter looks safe and natural for given tokens. And real delimiter doesn't even mentioned in doc. </pre> Alex Efros 2010-02-17T23:05:52 http://comments.gmane.org/gmane.os.inferno.general/4966 <pre>Folks: Can anyone point me in the direction of some "useful" (and perhaps complete?) examples of implementations of distributed file systems using Inferno? I don't mean obvious things such as using Styx to import/export resources from one machine to the other but actual, working examples of distributed file systems implementing crypto, redundancy and fail-over using Styx as the transport mechanism. I've seen various mentions of such systems (or the beginnings of them, at least) here and there in blogs and random PDFs but no workable project that I can easily download and fiddle with to indulge my personal curiosity. Thanks in advance, Matt </pre> Matt Adams 2010-02-16T08:01:58 http://comments.gmane.org/gmane.os.inferno.general/4961 <pre>Folks: Can anyone enlighten me as to the general "health" of Inferno vis-a-vis on-going development? I'm curious to know what, if anything, has been happening with Inferno since it's last major release in 2004. I see that the mailing list &amp; IRC channels have fairly low activity but I don't want to judge the goings-on of the Inferno community on these two items alone. Any insight from those who've followed Inferno's development over the past six years would be appreciated. I'm trying to gauge it's suitability for a personal project but would rather not dip my toes into an environment that appears to have little going on. (I see fairly regular commits to the official hg repository but those seem to originate from Vita Nova itself - not sure how much of this is maintenance and how much is continued development to push things forward - assuming that's a "need" as such) Thanks in advance, Matt </pre> Matt Adams 2010-02-14T03:17:19 http://comments.gmane.org/gmane.os.inferno.general/4959 <pre>Why does Inferno have both modules/libc.m and libc0.m? And why are they different but overlapping? </pre> Venkatesh Srinivas 2010-02-13T20:13:20 http://comments.gmane.org/gmane.os.inferno.general/4956 <pre>Hi! Just to explain what I'm working on in inferno-npe... I've added 3 calls to internally to hosted emu: ksd_key_create, ksd_getspecific, and ksd_setspecific; they're inspired by pthreads's key_create/getspecific/setspecific. ksd_key_create() returns an integer, which is a key that can be passed to getspecific/setspecific, to access a pointer local to that kproc. ksd_key_create takes a function called a 'destructor', which is called when the kproc exits via pexit(), which is called on that key. Unlike pthreads, there is no way to delete a key; I had no need for that. I added KSD (kproc-specific data) in support of new allocator work. I am implementing a Solaris/libumem-style magazine layer; an early snapshot is available in emu/port/objcache.c. At first, the objcache will not provide an malloc/free interface - it will only provide for creating 'caches' of objects of the same type. This looks like it will help - there are a good number of commonly allocated these objects, primarily Progq, Prog, ... structures. The KSD will provide for kproc-local magazines, so that the hot path for allocation/free avoids locks. Eventually, I hope to support allocations from the magazines. Why the allocator? In some timing results from dtrace on a Solaris/sparc system (without the JIT), Inferno running thread-ring, poolalloc + poolfree checked in third in total time spent for functions; (xec checked in first, as expected). I'm not working directly on the malloc/free/poolalloc/poolfree; while that would speed up all allocations (not just specific ones from Mainmem); perhaps I should be? I didn't, because I didn't understand the interactions between the Block headers (BHdr) and the garbage collector. Even otherwise, allocating related objects close together, particular with Solaris-style ctors/dtors is probably a good idea... Comments appreciated! </pre> Venkatesh Srinivas 2010-02-12T19:04:02 http://comments.gmane.org/gmane.os.inferno.general/4955 <pre>This issue comes up for me every year or so: What's the best, recommended, way to handle clean Inferno install and administration? I've read and re-read "Installing the Inferno Software" but continue to get messed up every once in a while. I think my biggest problem is that I tend to use Inferno as only a single-user system.. thus losing some of the theoretical benefits. The man pages are good, but has anyone come up with a base paper/article for getting people up to speed on administering a network of Inferno machines? -jas </pre> Jeff Sickel 2010-02-12T17:07:22 http://comments.gmane.org/gmane.os.inferno.general/4949 <pre>hi all, the jit on linux/arm does not work for me. i'm trying it on debian lenny (5.0, the latest stable) on a sheevaplug (ARMv5TE). an attempt to start it looks like this: $ emu-g -c1 [Emuinit] Broken: "Segmentation violation" Killed i've already had a look, but haven't been able to find a solution. (the mailing list archive at gmane does not let me search at the moment, perhaps it is a known problem). at least the dis interpreter works fine. so far, i am wondering how the jit (glue) code handles different abi's. if i understand correctly, libinterp/xec.c:/^xec calls the jit code through comvec(). i would expect the preamble to save some registers: it seems that linux with its "eabi" (as specified by arm) is callee-saves. i haven't read the full jit yet, so my suspicion about the abi is likely premature. and at least my first attempt to save the registers still results in the same crashes... i added instructions to the preamble to: 1. store callee-saves registers that the jit code modifies 2. call the real preamble just a few instructions forward 3. restore the registers 4. return the preamble then looks like this: 00144e30 e92d43f0 MOVM.DB.W R13,[R4-R9,R14] 00144e34 eb000001 BL 144e40 00144e38 e8bd43f0 MOVM.IA.W [R4-R9,R14],R13 00144e3c e28ef000 ADD $#0,R14,R15 00144e40 e59f500c MOVW $#11acd0,R5 00144e44 e585e020 MOVW R14,#20(R5) 00144e48 e5959008 MOVW #8(R5),R9 00144e4c e5958004 MOVW #4(R5),R8 00144e50 e595f000 MOVW #0(R5),R15 the first four instructions are new. only r4-r9 are saved. r0-r3 are scratch, and higher registers are not used in the jit (unless i've missed something, perhaps punted functions make it necessary to save the others too). also, i'm not 100% sure the movm's do what i intend to do. the segmentation fault handler mentions addr=0xc (after modifying the handler to take a siginfo_t* arg). perhaps that hints at the cause? i don't see it though. :) the assembler above is printed by das(), from libinterp/das-arm.c. shouldn't the parameters to the MOVM instructions be reversed? the first MOVM has bit 20 not set (STM, register list to addr of r13), the second MOVM does (LDM, from addr of r13 to register list). and aren't ()'s around R13 appropriate too? any hints/insights about causes,solutions,workarounds are appreciated! in the mean time, i'll be reading more of the arm jit. cheers, mjl </pre> Mechiel Lukkien 2010-02-09T13:30:57 http://comments.gmane.org/gmane.os.inferno.general/4944 <pre> After using wave lightly for a few weeks for keeping some ideas private and sharing some with friends, I find the planned "Save/Put each character" feature suggesting a - probably unplanned, but nevertheless - "warm and cosy" reliability feeling. (Yes, I also think, that trusting google your content is a totally different issue.) Although I can hardly imagine using it for collaboration (concurrently editing documents, aka chating) _as advertised_. It not only looks ridiculous in the video, we tried it and it is ridiculous. Although I have to admit, that most of the time this type of "collaboration" diverts into a fun war-typing (new word?) game. I think wave needs new conventions to be usable as advertised. On the other hand, I can imagine using wave as a front-end for some collaboration based applications. </pre> fgergo-Re5JQEeQqe8AvxtiuMwx3w< at >public.gmane.org 2010-02-02T14:51:07 http://comments.gmane.org/gmane.os.inferno.general/4925 <pre>I've put some updates to my laptop file system on Google code. I hesitate to claim it's ready for prime time, but I've started using it and I don't plan to do anything else to it unless I run into some problems or someone asks for it to do something. You can find it in the source and in the downloads on: http://inferno-bls.googlecode.com/ BLS </pre> blstuart-Bdlq13kUjeyLZ21kGMrzwg< at >public.gmane.org 2010-01-30T04:27:24 http://comments.gmane.org/gmane.os.inferno.general/4919 <pre>is the inferno .bit format equivalent to any of the plan9 image formats? thanks, r </pre> Robert Ball 2010-01-20T22:34:21 http://comments.gmane.org/gmane.os.inferno.general/4918 <pre>Hi there. As I see, the color depth is too low to show 24-bit pictures. Is it possible to increase the depth? I use a host version of Inferno (20090630). </pre> Pavel Nevidomsky 2010-01-20T13:07:46 http://comments.gmane.org/gmane.os.inferno.general/4917 <pre>Hi everybody, Over the past few days, I've run a few experiments with hosted Inferno on i386: First, I replaced the lock / add / unlock pattern in incref/decref with a locked XADD. This fairly dramatically sped up the thread-ring benchmark; on a 2.66GHz Core 2 Duo, from around 11 seconds to somewhere in the 9 sec range. (a similar improvement was seen on a Core i7 system as well). This patch is commited to inferno-npe, if people are interested; it involved teaching 8a and 8l about XADD as well. Flush off some success, I attempted to tinker with taslock. I first tried to replace the XCHG underlying taslock with CMPXCHG; the reasoning was the CMPXCHG doesn't involve a write to the lockword in the contended case, which means that the cacheline containing the lockword will not be brutally stolen from the first successful writer. Sadly, I noticed no performance differences on the thread-ring benchmark, on Roger Peppe's chamenos-redux program, or on an I/O intensive tar-like program I'm working on (the theory was that I/O intensive programs will involve a high ratio of syscalls, so lots of threads releasing() and acquiring()). I also tried adding an unlocked read before the XCHG; I've often seen this described as an optimization for i386 spinlocks. No difference seen here either, all the times are very close together. Perhaps the locks just aren't contended enough for these tricks to matter. Just thought people might be interested, </pre> Venkatesh Srinivas 2010-01-18T04:52:56 http://comments.gmane.org/gmane.os.inferno.general/4913 <pre>Hi, is there a version of Inferno for x86-64 systems? If so, could someone please point me in the direction of an installation CD/DVD? Thanks! </pre> David Fockens 2010-01-15T11:15:30 Search the mailing list at Gmane query http://search.gmane.org/?group=$group=gmane.os.inferno.general