* Arguments for max number of parts and tiles to render (optional).
+ Automatic estimation of maximum tiles to render based on max parts
for Writer docs, since there is only 1 part, this caps the number
of pages to render, similar to other doc types.
* Fixed rendering of Writer documents over and over (as many times as pages).
+ Writer has a single part, unlike other doc types.
+ No point in rendering the whole document in writer to a single tile,
since that's completely unrealistic and impractical (it takes
forever for large docs and artificially spikes the memory).
* Rendering starts at the current part and not the first.
+ This gives the spreadsheet of interest priority (if saved as visible).
* The tile size is now more realistic as we use the same dimensions
as the Online client does.
* When rendering tiles at scale, we use the same dimensions as the
Online client rather than splitting the width by 4, for realism.
* Rendering of tiles is done rows-first, rather than columns-first,
which is similar to what the Online client does, which is more
cache friendly, therefore more realistic.
* Enabled compiling of tilebench when GTK3 is disabled, which
was erroneous, since tilebench doesn't have any dependency on GTK.
+ Now it's possible to compile with local Cairo/Pixman libs.
Reviewed-on: https://gerrit.libreoffice.org/44936
Reviewed-by: Jan Holesovsky <kendy@collabora.com>
Tested-by: Jan Holesovsky <kendy@collabora.com>
(cherry picked from commit 7db98521548de9eab955ee25a5aacaaef42df323)
Change-Id: I6ad2e97f39572778dd7d0c12d14550841c1d6963
Reviewed-on: https://gerrit.libreoffice.org/46984
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Now since, LOKDocView has started to get clients, we need to ship
all of its dependencies with the widget.
Reviewed on:
https://gerrit.libreoffice.org/21605
Change-Id: I874eb416587a80b7760c5f00183d6004dfca4cce
Remove the unnecessary ones, but make sure that e.g. the sofficeapp library is
built (which was not covered by depending on services.rdb).
Change-Id: I1adfa4aaaf60787c944d71cc289cdafba8351c1b
The main problem was that UserInstallation did not point to the usual
workdir/unittest/, so initialization failed early, not finding the
soffice.cfg directory under
workdir/CppunitTest/libreofficekit_tiledrendering.test.user/.
Disable one testcase that still fails, though.
Change-Id: I3fdae5ead47dd2a805a0ed8d3cb407938f868e3b
This is a simple executable which can be used to ensure there is a
pe-pared configuration for unit-tests to avoid the usually required
re-start on the first run.
Change-Id: I49b706e8daa93bc34db508a24615b63b9babddba
It looks like the cleanest method of getting lok_init into
a LibreOfficeKitInit.h header (in a c89 compatible way) is to
have it as a static function.
(inline is only available in C99 or later -- this is actually
available on Linux which is the only place that we can actually
use lok_init anyways currently, however given we have to keep
c89 for the C code (for MSVC) compatibility, selectively enabling
c99 would likely be more messy.)
Change-Id: I0493e7a68ed5397479220bb6ba8c3db870b6dd32
I.e. we render the same area as one larger tile, and then
as 4 sub-tiles (which, when put together, should be identical
to the larger tile). However currently only the top-left sub-tile
actually matches the larger tile, so we have to disable the test
for the remaining sub-tiles.
Change-Id: If1130022b43898e20fefff3e9f592102da3e413a
Ashod Nakashian