[GSOC] Tiled Rendering, ahunt, Weekly Report 10 / WK30

Andrzej Hunt andrzej at ahunt.org
Mon Jul 28 05:24:06 PDT 2014

And finally, managed to miss the largest part of last weeks work:

* Viewport / overlapping tile matching issues:
** Reverted the assert disabling.
** Revert the DocShell::SetVisArea hack -- using this seems to confuse
    the rest of sw, causing the viewport asserts (i.e. previous point).
** Instead used SwViewSh:VisPortChgd-- which works as expected (i.e.
    correct and matching output, no spurious asserts).
** Allows comparison of all tiles in the lok tiledrendering test.
** But the test is currently still disabled due to loading issues on
    some TBs (and ~useless debug output from framework/sfx2).

For completeness sake, current branches:
* feature/soffice_main:
** Enables the rendering of images for calc/impress.
** Could probably be merged -- but a review would be good, especially
    since I'm unsure of the naming.
** Also currently sometimes segfaulting on shutdown since rebasing on
    master at the weekend, still debugging that...
    (Seems to be about 50/50 reliability.)
* feature/calctiledrendering5
** All the calc tiled rendering work.
** (Still needs further work as to not break the usual UI.)



On 28/07/14 10:23, Andrzej Hunt wrote:
> Hi All,
> * soffice_main LOK integration:
> ** Integrated configuration initialisation for unit tests into gbuild:
>     we now run a simple LOK binary to ensure that the unit test doesn't
>     have to deal with restarting itself instead.
>     ( gb_CppunitTest_use_confpreinit )
> ** Disabled Impress Remote for headless -- it's more or less useless,
>     and could block a real LO instance from getting hold of the correct
>     ports.
> ** Made CommandLineArgs ignore '--protector': soffice_main/Desktop init.
>     fails if there are "unknown" arguments (which --protector previously
>     was) -- this hasn't been an issue before as unit tests typically
>     don't run soffice_main.
> ** Added "--headless" arg to lok_initialize: we already manually set
>     headless on vcl/Application, but Desktop/CommandLineArgs has a
>     separate concept of headless (which is used e.g. to decide whether
>     or not to show document recovery dialogs) -- the latter is actually
>     used to set the former (the vcl/Application headless is then used
>     to cancel any later dialogs that might be shown).
> *** (Might be better to allow manual setting of the headless flag
>      directly rather than by passing string arguments which are then
>      parsed?)
> * Moved all this into feature/lok_sofficemain since it's probably not
>    far off being mergeable.
> * Rebased calc work atop of that (feature/calctiledrendering5).
> * Moved lok_init from shim.c into LibreOfficeKit.h
>    (i.e. killing the static library)
> ** https://gerrit.libreoffice.org/#/c/10584/
> ** Made it a static function -- can't use inline since we need to
>     support c89 -- since lok_init only exists on Linux we could
>     potentially use c99 there, but I imagine it would be a can
>     of worms to selectively enable c99 when virtually all the other
>     LO C code has to stay c89 compatible anyways.
> ** And we have to use #pragma to disbale -Wunused-function...
> Some screens of calc/impress with images visible are attached.
> Cheers,
>      Andrzej

More information about the LibreOffice mailing list