On 12/8/06, Adam Lally <[EMAIL PROTECTED]> wrote:
On 12/8/06, Lev Kozakov <[EMAIL PROTECTED]> wrote: > Unfortunately, the problem is more complicated that I thought at the > beginning. ><snip/> Let me try to simplify. I'll propose a solution and you can tell me why it wouldn't work. 1) When PearInstaller launches the JVM for CVD, it will set its classpath to the contents of the java.class.path system property of the PearInstaller's own JVM. Thus CVD will run with the same classpath as PearInstaller. Since CVD is in the same jar as PearInstaller and both depend on uima-core.jar, this should work. (I don't think we need to support cases where a user has pulled apart our jar files.)
This will work, of course, but requires significant modifications in the code :) 2) Stop launching an external browser to view the help file. Instead
use a JEditorPane. I know that control doesn't have the greatest HTML support, but this is only a help file, so it should suffice. You can now store the piHelp.html file in uima-tools.jar and locate its content using the ClassLoader. uima.home will no longer be needed for this purpose.
We can simplify the help file and use a message box to display it. 3) CVD doesn't need the uima.home JVM arg.
4) I don't know what uima.datapathchild is. Do you mean uima.datapath? Shouldn't this be determined by what's in the metadata for the PEAR file that's being tested? Or, just set this to whatever uima.home was set to when PEAR Installer itself was run.
uima.datapath is used by Gladis to search for sample text files. This is just a 'convenience' feature for users who want to test their installed PEARs with sample UIMA documents, and we can easily drop it. Summary: PEAR Installer shouldn't need to know what the uima install
location is. If we remove the dependencies on that, it will make life easier. -Adam
We would prefer to avoid modifying the old PEAR Installer code and rather wait for the future RCP application, unless it's absolutely necessary for this SDK release. -- Lev
