You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@harmony.apache.org by Tim Ellison <t....@gmail.com> on 2009/11/20 10:15:53 UTC
[build-test] new Hudson job (1.5 all tests)
FYI I have created a new Hudson job [1] that builds the 1.5 branch and
then runs all the tests in the checkout (excluding awt & swing at the
moment).
There are a number of failing tests, but it looks like some of these are
config issues caused by running headless that hopefully will go away
once the environment is set up right.
[1]
http://hudson.zones.apache.org/hudson/view/Harmony/job/Harmony-1.5-head-linux-x86_64-full-tests/
Regards,
Tim
Re: [build-test] new Hudson job (1.5 all tests)
Posted by Mark Hindess <ma...@googlemail.com>.
In message <4B...@gmail.com>, Tim Ellison writes:
>
> On 20/Nov/2009 10:40, Mark Hindess wrote:
> > In message <4B...@gmail.com>, Tim Ellison writes:
> >> FYI I have created a new Hudson job [1] that builds the 1.5 branch
> >> and then runs all the tests in the checkout (excluding awt & swing
> >> at the moment).
> >>
> >> There are a number of failing tests, but it looks like some of
> >> these are config issues caused by running headless that hopefully
> >> will go away once the environment is set up right.
> >
> > I'll fix the xvfb environment set up. "export DISPLAY=:0" only last
> > for the duration of the script not the subsequent ant script so it
> > doesn't work at the moment.
> >
> > There is actually a script to help with this so you can do:
> >
> > xvfb-run ant ... text
>
> Thanks Mark. There are a bunch of warnings when running Xvfb on
> minerva at the moment, which may not be important but I can help fix
> them if they affect our test runs.
The only ones that might will be the font ones but I think the tests
should only use the included dejavu fonts anyway.
> > This is handled by ant now in the run-hdk-tests macro.... hmm...
> > except for a few cases which I'll fix shortly.
>
> Ack - thanks.
Maybe not so shortly but as soon as I can.
> > New build running now ... hopefully a few more tests will pass now.
>
> I left it scheduled only @daily for the moment, thinking that's enough
> nagging to make sure I don't forget it. Of course, we can invoke manual
> builds while fixing the scripts.
Ok.
> > I left the "fetch-depends build" step alone but I was wondering why you
> > didn't use "fetch-depends rebuild" to get a clean build?
>
> Just because 'build' is shorter and that helps at this stage of
> debugging the build system. Once things are working I'll change it to
> do a rebuild.
Okay. I'll leave it for now.
One other thing I noticed is that we are running the gen-report step.
This is quite time-consuming (on some platforms it takes longer than
running the tests) and it is pointless since hudson reports based on
the raw xml files. Out-of-the-box there isn't a no.report flag on our
builds but -Dshort.report=true should be much quicker so I've added that
for now.
Regards,
Mark.
Re: [build-test] new Hudson job (1.5 all tests)
Posted by Tim Ellison <t....@gmail.com>.
On 20/Nov/2009 10:40, Mark Hindess wrote:
> In message <4B...@gmail.com>, Tim Ellison writes:
>> FYI I have created a new Hudson job [1] that builds the 1.5 branch and
>> then runs all the tests in the checkout (excluding awt & swing at the
>> moment).
>>
>> There are a number of failing tests, but it looks like some of these are
>> config issues caused by running headless that hopefully will go away
>> once the environment is set up right.
>
> I'll fix the xvfb environment set up. "export DISPLAY=:0" only last for the
> duration of the script not the subsequent ant script so it doesn't work at
> the moment.
>
> There is actually a script to help with this so you can do:
>
> xvfb-run ant ... text
Thanks Mark. There are a bunch of warnings when running Xvfb on minerva
at the moment, which may not be important but I can help fix them if
they affect our test runs.
> I'm going to prune the properties because some are not needed...
>
> # Link to version 6 of C++ runtime libs
> use.libstdc++6=true
>
> This is the default now.
>
> #Exclude UI tests for now
> exclude.module=awt,swing
>
> This is needed.
>
> # Run simple test on result
> test.jre.home=$WORKSPACE/harmony/target/hdk/jdk/jre
>
> This is the default.
>
> #reduce the crud left behind by redefining some vars
> tmp.dir=$WORKSPACE/harmony/tmp
> user.home=$WORKSPACE
>
> This is handled by ant now in the run-hdk-tests macro.... hmm... except
> for a few cases which I'll fix shortly.
Ack - thanks.
> New build running now ... hopefully a few more tests will pass now.
I left it scheduled only @daily for the moment, thinking that's enough
nagging to make sure I don't forget it. Of course, we can invoke manual
builds while fixing the scripts.
> I left the "fetch-depends build" step alone but I was wondering why you
> didn't use "fetch-depends rebuild" to get a clean build?
Just because 'build' is shorter and that helps at this stage of
debugging the build system. Once things are working I'll change it to
do a rebuild.
Regards,
Tim
Re: [build-test] new Hudson job (1.5 all tests)
Posted by Mark Hindess <ma...@googlemail.com>.
In message <4B...@gmail.com>, Tim Ellison writes:
>
> FYI I have created a new Hudson job [1] that builds the 1.5 branch and
> then runs all the tests in the checkout (excluding awt & swing at the
> moment).
>
> There are a number of failing tests, but it looks like some of these are
> config issues caused by running headless that hopefully will go away
> once the environment is set up right.
I'll fix the xvfb environment set up. "export DISPLAY=:0" only last for the
duration of the script not the subsequent ant script so it doesn't work at
the moment.
There is actually a script to help with this so you can do:
xvfb-run ant ... text
I'm going to prune the properties because some are not needed...
# Link to version 6 of C++ runtime libs
use.libstdc++6=true
This is the default now.
#Exclude UI tests for now
exclude.module=awt,swing
This is needed.
# Run simple test on result
test.jre.home=$WORKSPACE/harmony/target/hdk/jdk/jre
This is the default.
#reduce the crud left behind by redefining some vars
tmp.dir=$WORKSPACE/harmony/tmp
user.home=$WORKSPACE
This is handled by ant now in the run-hdk-tests macro.... hmm... except
for a few cases which I'll fix shortly.
New build running now ... hopefully a few more tests will pass now.
I left the "fetch-depends build" step alone but I was wondering why you
didn't use "fetch-depends rebuild" to get a clean build?
Regards,
Mark.