Forum OpenACS Q&A: OpenACS on Win2K

Okay, I spent some time today playing with Win2k (my day job). Thanks to Sam Snow, who pointed to me to Cygwin, I had a perfectly good excuse for doing nothing very useful all day :)

Cygwin (the latest available download anyway) comes with PG7.1 pre installed - very hoopy. You need to download some extra things to be able to start the postmaster (took me all of two minutes to fetch and install), after which the postmaster kindly starts, without fuss (Although it didn't have fizz either, it was quite slow).

I downloaded OpenACS (from CVS) and loaded the datamodel, without any problems. All you need to change is the .../var/lib/something/plpgsql.so in postgres.sql to /lib/plpgsql.dll.

The next part was AOLServer. You need to do some editing to get the config files working, but AOLServer will start and answer queries, no problem.

Next I needed the Postgres 7.1 source to be able to compile the nsd-postgres driver-2.3. Compiling the driver on Win2K sucks, but after figuring out which headers to use, which .lib to use (DO NOT link against the static libpq.lib, it won't work - use lippqdll.lib) I got it to compile without errors.

That's how far I got. It took me plenty of time, 6 hours in all, most of it trying to compile AOLServer-pg-driver against the postgres libs. I started AOLServer once, but the postgres driver wouldn't load (windows error 128, it said (or was it 126?) ). It may well be that the postmaster just wasn't listening on 127.0.0.1:5432, I'll figure that out on friday - which might well be my last day at with my current job, we'll probably go belly-up. Anyone got a nice (OpenACS) job for me?

<i>By the looks of it, running OpenACS on Windows 2000 is probably useful for academic purposes only... </i>
<p>
How come--just because it hasn't been tested extensively yet?  Would it be stable enough for development maybe?
</p>
<p>One of the big selling points for the ACS for me originally was that I could load all the stuff I needed on my Win32 laptop (AOLserver, Oracle, etc) and do development on the road.  I could check our website out of CVS and work on it, commit changes and check them back out on our Linux servers without any changes at all.
</p>

Oh, it would definitely be stable enough for development work. I would not do that, though. Cygwin provides a reasonably large number of the standard unix tools (haven't tried em all), but I'd rather have a full Unix environment. Also, as soon as you start interacting with the 'outside world' you'll run into platform dependant things. Like filenames, file handling (try rotating a log on win32) a lot of things are subtly different.

I would go for VMWare. I use it a lot already, and for text mode apps you don't need very much CPU horse power. Just memory. Lots of memory, I have 384MB at work. My 256MB at home usually doesn't cut it, although my home machine is a lot faster (700Athlon/Asus compared to 533PIII/Dell). I would do it the otherway round, though: A linux guest on Win2k host. I have experienced some rather large performance differences, between the two, mainly because Windows sucks. It's memory allocation mechanisms don't play nice with VMWare, Linux runs much, much smoother in VMWare than Win2k (even when using X).

Keep your VMWare guest OS lean and mean, and even Postgres will run like lightning (give it some RAM..). Another extremely nice feature of VMWare is 'undoable disks'. Just run your Postgres /DATA in such a disk and you can restore to a base platform by simply discarding a session. Very nice for setup/upgrade testing.

Comming back to ACS on Win2k. I wanted to see if it could be done. It can. If we are to support this (waaaay too early for me), it should be properly documented first. Again I am fairly certain there are all kinds of nasty Win32 snares.

As I have had more-recent-than-the-windows release hardware since, oh, 1996, I have (unfortunately) always had to stay with a rather recent version of Windows. Especially running M$ Visual C or Borland C Builder manages to crash a lot more often on older windows versions on newer hardware. Running '98 in the VMWare just hadn't crossed my mind yet (mind you, at work we can't as we are using it to test server applications with permissions and all), but VMWare would not have those problems.

I wasn't talking about disk performance, though. It seems that the SysMalloc()/SysFree() calls of [D]COM[+] are really inefficient within the VMWare client. That means that each time a VisualBasic app is splicing and merging strings CPU usage goes to 100%. 's probably got to do with allocating 'cleaned' memory pages, where the content is guaranteed to be all zeroes.

You need to get the PG sources and follow the compile-on-win32 instructions, cygwin alone will not work. You'll end up with client libraries for pg, both dll and static. I could only link against the DLL version (libpqdll.lib or something).

I compiled with the cygwin pg-include files and the win32 library files.

Hope this helps.

I have compiled without reconfiguring or anything, I don't remember
exactly how (this was only a test for me, nothing serious). I have
used VC++ 6 for compiling, I have never tried GCC on cygwin.