I will fix installer.*.xml files when I get to CSDL today. The reason why sensorshell.jar isn't removed is because each sensor doesn't really "know" about the others. So determining if sensorshell.jar should be removed was a little tough.
Option (a) means that sensorshell will just continue to grow in size. This option makes me nervous because all the sensors using Ant will get installed and removed at once. To me it seems like users on the outside using Hackystat will be wondering why they have to download so many sensors at once when they maybe only want one. This doesn't seem as "nice" for the users. Option (b) sounds better but could cause more problems. This would require more testing on our part to make sure everything is functioning correctly after being stripped down. However, the clients would still have control over all sensor installations individually. I think it really depends on whether we would like to make it easier for the developers to build the system or if the user's should have more "control" with their installation and removal of sensors. ~Julie ----- Original Message ----- From: Philip Johnson <[EMAIL PROTECTED]> Date: Friday, October 7, 2005 10:44 am Subject: Re: [HACKYSTAT-DEV-L] hackyInstaller and sensor.properties > Sigh. I guess this proves that I can't put off dealing with the > 'redundant > sensor code in sensor.*.jar and sensorshell.jar' issue any longer. > Thanks, > Aaron, for articulating the problem so cogently. > > I see the following possible approaches: > > (1) Change HackyInstaller so it defaults sensors to disabled, not > enabled, > when it doesn't find a setting for them in sensor.properties. This > is a > band-aid, of course, but at least stops data from being sent when > the > developer didn't even install the sensor explicitly. > (HackyInstaller crew: > please make this change ASAP.) > > (2) Eliminate the redundancy. There are two ways: > > (a) Stop creating sensor.*.jar files entirely; use sensorshell.jar > as > container for all Java-based sensor code. Advantages: simplicity > in build > process; reliability; simplified client-side installation (download > one > sensorshell.jar, get many sensors installed). Disadvantages: no > way to > uninstall a subset of Java-based sensors on client side. > Implications: > local.build.xml files will no longer create the sensor.*.jar file; > hackyInstaller must be changed so that it can understand that > installing > one sensor implicitly installs many others, and that uninstalling > one > sensor could potentially uninstall several others, documentation > updates. > (b) Stop including sensor code in sensorshell.jar file. > Advantages: > supports individual uninstallation of sensors. Disadvantages: > build > process more complicated; possibility of missing classes on client > side. > Implications: build process must be changed to strip sensor classes > from > sensorshell before jarring. > > What do people think about (a) vs. (b)? They seem to be about > equally > complicated to me, so it's more about which approach seems to > result in a > more usable system. > > Cheers, > Philip > > --On Friday, October 7, 2005 10:25 AM -1000 Aaron Akihisa Kagawa > <[EMAIL PROTECTED]> wrote: > > > Not to mention that the HackyInstaller Uninstall functions does not > > remove sensorshell.jar nor disable the sensors. > > > > thanks, aaron > > > > ----- Original Message ----- > > From: Aaron Akihisa Kagawa <[EMAIL PROTECTED]> > > Date: Friday, October 7, 2005 10:12 am > > Subject: [HACKYSTAT-DEV-L] hackyInstaller and sensor.properties > > > >> Hey guys, > >> > >> I found a bug. > >> > >> I am hacking on some Ant code installing the Build Sensor in some > >> projects. I wrote the Ant code, ensured that sensor.build.jar > wasn't>> installed (i have sensorshell and sensor.locc.jar > installed in > >> Ant/lib),and executed Ant to be sure that I wouldn't send Build > >> Data. Well, the > >> bug is that I did send build data. > >> > >> Here's why: > >> 1) sensorshell.jar contains all the classes in Hackystat, so I > don't>> really need to have sensor.build.jar in my Ant/lib > directory. thats > >> another bug. > >> > >> 2) hackyInstaller seems to enable all sensors in > sensor.proerties even > >> though they are not installed. So, when I opened > >> hackyInstaller.jar I > >> found that my build sensor was not installed. When I checked my > >> sensor.properties I saw that my build sensor was enabled. > >> > >> So, you should be able to see the problem when 1) and 2) are > combined.>> hackyInstaller thinks a sensor isn't installed and so > does the user. > >> But, still I send data. > >> > >> thanks, aaron > >> >
