RE: [ jEdit-users ] ant farm - large builds on same vm

[Richard W. <rw...@pa...>]

From: John Gellene [mailto:jge...@ny...]
> The current AntFarm display requires an initial parse=20
> operation yielding an
> array of strings - the name of the file and the names of=20
> targets.  Naturally
> you would want to check the XML for validity at load time,=20
> but you don't
> have to keep the document tree in memory.  Why would you not=20

I do not believe the document tree is held in memory, what is
held is the org.apache.ant.Project contained in ant.jar which
has its own methods to load from XML.

> want to?  If
> you're not about to run a build, you would want to minimize the
> application's footprint, while still being able to invoke  a=20

I agree that we want to minimize the footprint, the question is:
how can this be achieved?  I dont think any of us can say for sure
where the problem is.  The evidence I have gathered so far:

1.  the size of the build.xml file does not cause problem
2.  the number of files compiled when doing a javac through antfarm
with the modern compiler on the same VM appears causes a problem
3.  holding on to the Project instance when launching on a separate
VM does not cause a problem
4.  reloading the Project instance from XML does not seem to help any.

What this suggests to me is that tools.jar is the culprit and that
the XML document tree is not.  Before we can make any conclusions,
it would be nice if some kind soul were to try a large build with
JCompiler (I cant seem to get it to run, it always pops up the ant
shell and gives an error and I've noticed no JCompiler shell -
the last time I used JCompiler half a year ago, there was a JCompiler
shell)

If someone could confirm or deny my findings with AntFarm that would
also be nice.  Memory issues are difficult to track and even more
difficult to fix.

----------------------------------------------------------------------
Richard Wan
Servidium Inc.