ant-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jesse Tilly <>
Subject RE: Some Thoughts on Ant 1.3 and 2.0
Date Mon, 30 Oct 2000 19:17:49 GMT
if anyone's done parallel builds using 'make -j' on SMP machines can relate
as to how frustratingly useful (when the makefile is written right) or
useless (98% of the time) parallel compilation really is.


-----Original Message-----
From: Thomas Christen []
Sent: Monday, October 30, 2000 1:32 PM
Subject: RE: Some Thoughts on Ant 1.3 and 2.0

You're probably right, having parallel targets could be much more complex to
be implemented than "only" parallel (or as you wrote asynch) tasks. If I'm
understanding correct, and Stefan has already mentioned, there need to be an
implicit join at the end of a target.

My last description of parallel targets was not quite complete. There's
another thing witch could make parallel targets worse. Parallelism leads
from a pure sequential processing to complex network of dependencies between
targets. Means, each target could get its own thread and being started
immediately. Then the target would need to wait until all dependent targets
have finished ... but in a context of a lot of targets we probably would run
out of threads. A workaround could be a thread pool and a mechanism for
evaluating the next target ready to be executed and so on.

Thomas Christen

> -----Urspr√ľngliche Nachricht-----
> Von: Conor MacNeill []
> Gesendet am: Montag, 30. Oktober 2000 01:02
> An:
> Betreff: RE: Some Thoughts on Ant 1.3 and 2.0
> Forgot one of my favourites. I have in the past proposed supporting the
> multithreading of tasks within a target by the addition of a boolean async
> attribute. I think this is a valuable feature for running test
> harnesses on
> servers, without the need to start multiple ants. I have used this with
> JUnit tests on a Weblogic EJB server.
> Thomas Christen has asked about having parallel targets within a
> project. I
> don't really support that as I think it is significantly more complex than
> multithreading tasks within a target. Nevertheless I think we
> should examine
> both ideas as part of this process to see if either of them has merit in a
> future ant.
> --
> Conor MacNeill
> Cortex eBusiness

View raw message