axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Steve Loughran" <stev...@iseran.com>
Subject Re: Axis Build System - Possible conversion to Maven
Date Sat, 15 Mar 2003 02:41:56 GMT

----- Original Message -----
From: "Dan Diephouse" <dan@envoisolutions.com>
To: <axis-dev@xml.apache.org>
Sent: Friday, March 14, 2003 13:32
Subject: Re: Axis Build System - Possible conversion to Maven


> This is exactly my point!!!  Maven takes care of all these technologies
for
> you!!!  It will be the only technology you need once the build conversion
is
> through.  It automatically takes care of and downloads the dependencies.
All
> you need to do is tell maven what dependencies there are in your project
file
> like so:
>
>    <dependencies>
>
>      <dependency>
>        <id>log4j</id>
>        <version>1.2.6</version>
>      </dependency>
>
>      <dependency>
>        <id>jaf:activation</id>
>        <version>1.0.2</version>
>      </dependency>
>
>      <dependency>
>        <id>servletapi</id>
>        <version>2.3</version>
>      </dependency>
>
> </dependencies>
>
> It is so amazingly simple.  I spent several hours trying to get axis to
build
> and converted Axis to use Maven in 15 minutes (not with all the tests yet,
mind
> you) - and I am by no means a Maven expert.  It is very, very simple.

1. was getting a copy of the servlet API what caused you your hours of
loss?

2. gump handles this for the main build, but I do see the merits in this
kind of feature -which is why Ant will adopt some variant in due course.

>
> The point is I can't contribute to what you, other developers, and the xml
> community need until I can get a handle on the project.  For example - you
have
> more than 114 ant build files!!!  This is insane!


>
> I know what your bug list is.  The same bugs bother me.  However, I have
> considered doing my own SOAP implementation because Axis is so hard to
work
> with.  I can't fix these things until the build becomes manageable.

The issue is not the build of the axis distro, that is just "ant
compile" for the jars, ant dist for the full package. The many build
files kick in when doing ant test.

And if you went to write your own soap implementation (a lightweight
client would be nice BTW), then you will find the issues are
-interop w/ other systems
-wsdl to java
-functional testing

and your life will soon get complex. Ms have whole test teams dedicated
to interop, we have a hundred+ build files that rerun the test suite
every six hours.

Now, I clearly have a historical bias toward Ant, but I am open to
technology that improves the build process, where that includes your
build, my build of axis, my build of apps that depend on axis, big
builds of products (jboss.net, websphere) that depend on axis. And to
support those, stability of build process is important.

I also think the troublespot is not the core distro, but the tests.

We do need to fix that. The current iteration gives us
-ability to control dependencies in a package
-ability to run a single test
-ability to have non-standard builds in a test

at the expense of
 -the same build file replicated (We could fix that, to an extent)
 -expose memory leaks in javac (pre1.4) and ant (pre 1.5.4 & 1.6alpha).
 -slow build process

 Changing the build process isnt the solution. Or maybe it is, but it
 isnt the real problem, which is we need to reexamine what our test
 process is.

------

[On the subject of testing, I can borrow an Itanium 1 workstation if I
can think of a good excuse. Its too big and noisy for the office, but I
was thinking of bringing up a 64 bit os (unix or nt) and running gump on
it. Thoughts?]



Mime
View raw message