Is there a recommended way of generating jtreg java test source?

Jonathan Gibbons jonathan.gibbons at oracle.com
Mon Oct 20 20:12:27 UTC 2014


In general, I think that designing something that works well in all 
OpenJDK repos is very difficult, which is why I generally advocate more 
customized libraries.  I have also found it to be logistically difficult 
to design and provide something to satisfy everyone, with the added 
complication that when you update a sharedf test library, you risk 
affecting other tests with unforeseen consequences.

For example, in the langtools world, we are interested in being able to 
work with the error messages from javac, and to exercise different 
invocation modes.   In other repos, you're more likely to just be 
interested in just getting your source code compiled and getting an 
exception if there is a compilation error.

ToolProvider and JavaCompiler are definitely the right building blocks 
to use.   It also helps to have available a method to create a 
JavaFileObject from a string.  ToolBox (that I referred to earlier) also 
has some smarts to derive a filename for a class by using regexes to 
find the package and class declarations.

-- Jon


On 10/20/2014 01:03 PM, Martin Buchholz wrote:
> So ... I looked at various testing tools in openjdk, and I think it 
> would be good if there was easy-to-use support this kind of thing 
> available from every jtreg test, but the design of such a general 
> utility would be difficult.
>
> For my own use, I ended up using ToolProvider and JavaCompiler, and 
> the result was not bad, but it would be a little better if just the 
> right infrastructure was available.
>
> On Wed, Oct 15, 2014 at 2:17 PM, Jonathan Gibbons 
> <jonathan.gibbons at oracle.com <mailto:jonathan.gibbons at oracle.com>> wrote:
>
>
>     On 10/15/2014 01:04 PM, Martin Buchholz wrote:
>
>         Occasionally, one would like to generate the java sources to
>         be tested instead of checking them in (e.g. the generated
>         source files may be huge).  Is there a way to do that?
>
>         Ideally, you would want to have some java code be @run to
>         create the source files, and those could then in turn be
>         @compile'd and @run'd, but that does not seem to be possible.
>
>
>     Martin,
>
>     We routinely do this in the OpenJDK langtools/test test suite,
>     where we have a style of testing that generates hundreds,
>     sometimes thousands of classes on the fly, to provide the test
>     cases for a feature.
>
>     You can generate the Java source code in memory, and then use the
>     Java Compiler API to compile it (javax.tools.JacaCompiler) either
>     to class files in memory or to class files on disk, depending on
>     your ultimate usage.  Note if you compile to memory, you can
>     execute the classes using an appropriate ClassLoader.
>
>     But, this all has to be done within a test class -- there is no
>     inherent support in jtreg to @compile and @run generated stuff.
>
>     You can look at the file langtools/test/tools/lib/ToolBox.java for
>     a utility class to facilitate creating in memory source objects,
>     compiling them, and so on.  We wrote and used this class to
>     replace many of the remaining shell tests in the langtools repo.
>
>     -- Jon
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/jtreg-use/attachments/20141020/d0233733/attachment.html>


More information about the jtreg-use mailing list