Re: [eigen] Improved dashboard submissions...

[Eamon Nerbonne <eamon@xxxxxxxxxxxx>]

As a bit of anecdotal evidence; when I built the tests (just once), I avoided running and submitting tests because it wasn't clear exactly how to do so nor whether that was welcome.  A batch file somewhere may work well; but it's not immediately obvious whether it's maintained, whether you need an account on some dashboard, whether people are supposed to run the tests and store the results and submit that file or the script does that for you, etc.

Integrating it into the cmake/make targets makes it clearer that the target should Just Work.

--eamon@xxxxxxxxxxxx - Tel#:+31-6-15142163

On Tue, Feb 15, 2011 at 07:59, Hauke Heibel <hauke.heibel@xxxxxxxxxxxxxx> wrote:

I want to clarify again while trying to be short.

Dashboard submissions should be possible with

 "make experimental"

on all systems (or "make nightly", etc.). The reasons are numerous. I
bet, we will get many more users who submit do the dashboard. This
means faster detection of regressions. It could lead to a better
coverage of more systems. It would prevent "empty" dashboard
submissions... and I could continue to fill the list.

The only thing standing in the way of this feature is that we exclude
the testing targets from "all" to prevent accidentally building all
tests. I still do not quite get, why we are doing this. You will never
have to wait for ever but at most until you hit "ctrl break" or
whatever shell escape sequence works on your system to abort such a
build.

In case you prefer to discuss this on the bug tracker as a feature I
can add a request.

Finally, I want to say, that I would love to see this feature now
since it will for sure improve our 3.0 release while being easy to
integrate - we can hardly break anything.

- Hauke

On Sun, Feb 13, 2011 at 12:53 PM, Hauke Heibel
<hauke.heibel@xxxxxxxxxxxxxx> wrote:
> On Sun, Feb 13, 2011 at 12:44 PM, Jitse Niesen <jitse@xxxxxxxxxxxxxxxxx> wrote:
>> On Sun, 13 Feb 2011, Hauke Heibel wrote:
>> I don't know ctest/cdash that well. From your comments, I get the impression
>> that in the Windows IDE you get automatically an Experimental (or something
>> like that) target which submits the tests to the dashboard. Any idea how it
>> is supposed to work with Unix make?
>
> If I am not totally wrong you would simply "make Experimental" and
> cmake would build and submit the tests. No need to configure or touch
> anything extra besides running CMake once.
>
>> That has been there from before my time, but isn't it because otherwise all
>> the test programs are built when you do 'make' or 'make install' and that
>> takes a very long time?
>
> Right, there was something like that. I just don't get why you would
> ever type "make" if you don't intend to build everything? That was
> probably just a precaution.
>
> Maybe we could agree upon allowing this type of convenient dashboard
> submissions on Windows by default.
> (EIGEN_LEAVE_TEST_IN_ALL_TARGET==true for win32 systems).
>
> - Hauke
>
> p.s. I attached a new patch which also works with nmake. Benoit, by
> the way, our scripts do properly propagate the SSE flags for nmake.
>