Older blog entries for dan (starting at number 67)

A cry for help: anyone successfully using Prism II (SMC/D-Link/Compaq/Eumitcom/etc) 802.11b wireless ethernet cards in Ad Hoc mode, with a free OS (i.e. has a driver with source I can look at)

I have two computers with wireless cards in them: one's the SMC 2632W PCMCIA card, and the other's the 2602W, which is basically the same thing stuck into a PCI adaptor. The orinoco_cs driver in Linux 2.4 talks happily to the PCMCIA card, and with the help of various web resources I was able to write the necessary PCI glue for the other (it's not a PCI->PCMCIA bridge in any normal sense; it's a PLX 9052)

Problem is, I have nothing to test either of them against except the other (the desktop box is intended to substitute for a real access point) - and at least one of them's not working. They both flash lights and so forth. The driver idents them as "Firmware ID 1F vendor 0x6 (LinkSys/D-Link) version 0.7". I can run iwconfig to set ESSID, put the cards into IBSS mode, set the channel, and so forth, then run ifconfig to set addresses and bring them up. At which point I find that they don't actually see each other on the network ...

I've tried both the old Lucent port3 and the newer ibss mode. Firmware version is claimed to be 0.7 - maybe there's something newer somewhere, I don't know. Anyone any clues?

Oh, I'll send my PCI glue stuff to anyone who asks for it and can can tell me whether it works ...

ksandstr:

Not that tearing down and reimplementing (aka refactoring, like the OO-correct like to call it) isn't a good thing some of the time -- I'm just not sure if the language design should enforce it.

It's a good thing some of the time, yes, but it's not refactoring. Refactoring is about incremental changes to an existing code base, not about throwing the existing code away. wnewman is on record as saying that more programmers should read Martin Fowler's book; I concur. ISBN 0-201-48567-2, if you want to go and look it up.

It sounds to me as if ocaml is making it harder for you to refactor rather than easier. I don't recall having had that experience with it, but it was a long long time ago that I used it and I've probably forgotten the scary bits.

Long time etc blah (sorry, dria). Hacking

• Lots of time recently spent working on CCLAN policies and packaging.

• learnt an awful lot about logical pathnames.

• detachtty is a silly little utility that lets you run a Lisp "backgrounded" and connect to it over the network on an as-needed basis.

• CLiki got an automatically updated Recent Changes page

• db-sockets CCLAN package uploaded. Araneida and CLiki now work on SBCL, so they're next

• There are SBCL packages in CCLAN (x86 and alpha), but they're old. It would be cool to update to a newer snapshot one day soon.

and non-hacking

• Regular cycling

• Babylon 5 videos

• Reading "The Shockwave Rider" and re-reading "Cryptonomicon"; time assigned to each book according to whether it or the other is nearest at hand

• Probably ought to remember to eat regular meals.

wnewman integrated my SBCL patches, and it looks like a clean job - a fresh checkout from SBCL CVS will now build and pass tests on Alpha. If you're intending to try, you need to edit make-config.sh (to set $sbcl_arch), customize-target-features.lisp (to make sure :GENCGC isn't in the features list; by default it is) and then to make alpha-linux-types.lisp (see http://www.geocrawler.com/lists/3/SourceForge/1663/0/5705651/)). Also the sun is shining. So I'm happy.

Old college friend staying at the weekend, and parents showed up for lunch on Monday. Once again, sorting out my expense claim has taken a back seat.

Your mail to 'Sbcl-devel' with the subject

     Alpha patches!

 Is being held until the list moderator can review it for
approval.

 The reason it is being held:

     Message body is too big: 167203 bytes but there's a
limit of 40 KB

 Either the message will get posted to the list, or you will
receive
notification of the moderator's decision.

 

Whee. Sent. Now I have absolutely no excuse to put off sorting out my expenses claim any longer.

For some obscure reason I decided I wanted to get up in time to hear the singing from the tower at Magdalen College. So I did

Now it's 6:50am and I'd actually quite like to go back to bed.

Happy May Day.

Back from Turkey, and busy.

• The machine hosting cliki was being ill again, so I moved it all to the new Rackspace box.

• "Play Nice With Unix" got a new name - see link to cirCLe above

• Moves are afoot to create CCLAN (working title) - a CL equivalent for CPAN. looking for a name, but that's not the problem right now. Decided to start by creating RPMs of cmucl with common-lisp-controller (the necessary package infrastructure to make it work) and put them up for ftp. Hoping Bryce can come through with the disk space

• Someday I hope to tell you all about the hosting provider I was using for the customer project before I went to rackspace. I'd better not until I've calmed down slightly, but, well, they didn't like me using 20Mb of RAM (the size of two mod_perlized apache children; what's the fuss?), I didn't like them double-billing me for bandwidth. I think we were about evenly matched in the moral outrage stakes.

  Still, if anyone tells you that FreeVSD is "just like having a dedicated server", they're wrong, OK?

• Every time I get it together to start working on sbcl again, something somewhere else goes wrong

tbmoore: CMUCL GC work? Neat! Is this the x86 generational gc, or the stop/copy one used on the other ports?

Not a lot of SBCL has happened lately as I've been occupied with moving house, settling in and so on. Latest update: it can rebuild itself and pass all tests but two (dynamic loading and "check that the fasl file extension is x86f" - I don't intend to fix the latter, either ;-) but it won't compile the latest CVS sources. Seems there are still VOPs (Virtual OPerations - short instruction sequences) in there that leave badly formed data in the descriptor registers - if a GC happens during one of them, blammo. With my two-month-old snapshot it happens not to tickle the bug, but on the current source tree it does. I think.

More SBCL, then. After updating to CVS current I find that four of five times it won't build in a 64Mb heap - unfortunately heap size is hardwired into CMUCL at compile time. So I reverted to my backup, and now I'm trying the "rebuild SBCL with itself" exercise - the rebuilt image will have 128Mb heap, too. (There's no longer any particularly convincing reason the heap size has to be compiled in at all, as far as I can tell, but one step at a time, hey?). When it's built itself, and the second stage can pass all the regression tests, it might be time to start figuring out how to do CVS

And while it builds I'm packing books into boxes. I found my Oxford A-Z, which is a bonus

Note to self: fix PURIFY

frabjous, dude

I found my SBCL/Alpha bug. And it was the last showstopper remaining. That's right, I have working SBCL for the Digital^WCompaq^API^Wwhoever-they-are-now Alpha

I still need to resync against current CVS, because my working copy is a couple of months old. And, as noted in that email, it's not exactly tested yet. But. Still. So cool. I was so happy when it worked, I even did the washing up.

5 days until I move from this place. That makes me very happy too.