Re: 'make check' with Windows/MSVC build?

On 04/11/2012 02:17 PM, Lubos Lunak wrote:
> On Wednesday 11 of April 2012, Stephan Bergmann wrote:
> > However, I guess the mechanisms to create a LO installation against
> > which to run the tests (which was always handled slightly differently on
> > Windows, to avoid problems with long pathnames) has been broken with the
> > various changes how to create that LO installation in general (ooinstall).
>
>   So it can work, it's just "slightly" broken somewhere at the moment? If
> that's so, could somebody who understands this stuff have a look?

There's two points to this:  Originally, the installation against which 
the tests run had been produced in a two-step process.  First, building 
instsetoo_native always produced a tgz (or zip, on Windows) archive 
installation set in addition to what else it was requested to build 
(rpm, deb, etc.).  Second, depending on platform:  On non-Windows, 
building smoketestoo_native unpacked the archive installation set to 
solver/*/installation once, so that all tests could use it from there. 
On Windows, running each individual test unpacked the archive 
installation set into a temp directory (and removed it afterwards).

Today, this has been changed completely, with the installation set being 
built and put into place in one step through ooinstall.  So for Windows 
to work again, first ooinstall needs to be fixed to work on Windows (my 
understanding is that it does not).  Second, as long as long pathnames 
on Windows are still an issue (and my understanding is that they are, at 
least for our interaction with the berkeleydb library), it would be 
advantageous to mimic the original logic, like running ooinstall not 
once but for each test, instructing it to output into a temp directory.

Stephan