If the clBuildProgram() failed, the lastSecondProgram variable kept as its
value the already released program handle, and later we called
clReleaseProgram() on it again. This is certainly wrong, and caused a crash at
least with my OpenCL implementation.

As such I am not sure if that lastOneProgram, lastSecondProgram etc dance is
sane...

Change-Id: Ia30426fbba9118ad7e20f13ef35daa5f78a1f0a1

Change-Id: Icb28114d3939063dedaedbd0ce370210b3721fc5

Change-Id: I96ef1c6758eaffb88e2167acee7c9d810317ca44
Reviewed-on: https://gerrit.libreoffice.org/14384Tested-by: Michael Stahl <mstahl@redhat.com>
Reviewed-by: Michael Stahl <mstahl@redhat.com>

Added our own map of MIME types because we cannot rely on Android's.

Change-Id: I11d4b639f42d30aa8adb0ea092797fb30b4fc8b9

Change-Id: I965405ab5821e86b2fa93b42db50e36dc9dbd0ad

Change-Id: Id367dee6c4fe55fe039ebf28603c883014194832

Change-Id: Ic118450e11ca6a4f705670dd71015ff74072fa05

regression from the SwapIn work.

The graphic size during SetGrfFlySize appears to assume its the pre-swapped in
size, i.e. the old 0 by 0 "size is not set" for an html image used to be the
value used here. So get that value (and update it if necessary) before updating
the image data and pass that to SetGrfFlySize

Without the swap-in in the html filter the data gets swapped in during the
initial draw of the image which is too late to update the image size, so
restore the equivalent of that swap-in via GetTwipSize

commit 61a5abd5
Author: Zolnai Tamás <tamas.zolnai@collabora.com>
Date:   Fri Nov 7 10:43:34 2014 +0100

    Make SwGrfNode swapping methods private

    Change the filter test accordingly.

    Change-Id: Ide3043f2f245c097a7b4c07ba2e0713510296b3e

Change-Id: I0afe244f79d8628236b1e552036587a2b4540a76

Change-Id: Idf857f5681766923675bdc210c02a796e0f58024

Change-Id: Ieb13095abb399662e449fad5a056999343165025

... in SwPageFrm::_UpdateAttr()

It looks like SwFmtChg always is sent in pairs and shouldn't contain
null pointers.

Change-Id: Ib14650d5ac7ed579af915806e738aec8a92add32

IsTableMode() should imply cursor is in table.

Change-Id: Ibb94911430f947038abba2ebe013fc82fae7dd54

Change-Id: I7a3a9593cabec68c0d60c17a991207b5d3dab69c

Change-Id: Idb4e50a98a2fcdd1746ef14fc0dbe53bf1d7db8a

Remove icons that simply duplicate the default images, but are incorrectly stored in Locale specific directories (e.g. /cmd/pt-BR/). If they are removed, the default will be used anyway. De-duplication makes the task of icon theming easier TDF#30425.

Galaxy - 124x removed:
- lc_bold x10, 2.4KB. Default: 'B'.
- sc_bold x10, 1.6KB.
- lc_italic  x13, 2.8KB. Default: 'I'.
- sc_italic x13, 2.1KB.
- lc_underline x12, 2.8KB. Default: 'U'.
- sc_underline x12, 2.0KB.
- lc_underlinedouble x9, 2.2KB. Default: 'U'.
- sc_underlinedouble x9, 1.7KB.
- sc_label x26, 5.4KB. Default: 'ABC'.
- lc_numberformatdecimal x5, 1.7KB Default: '0.0'.
- sc_numberformatdecimal x5, 1.2KB.

Check and remove corresponding icons from other themes: Crystal - 82x.
Hicontrast - 97x.
Human - 76x.
Industrial - 76x.
Oxygen - 79x.
Tango - 81x deletions from links.txt: "# Removing language specificity…".
Tango_testing - 25x.

Change-Id: I4e42b494cbdfac2b377fae077001306778addfd0
Signed-off-by: Andrew <dent.ace@gmail.com>
Reviewed-on: https://gerrit.libreoffice.org/14378Reviewed-by: Jan Holesovsky <kendy@collabora.com>
Tested-by: Jan Holesovsky <kendy@collabora.com>

Change-Id: Idcd0cc761121dd356e307003441f56c239efd4f1

Change-Id: I98060f1adae0ba8ec03b2f0d6b0db6d5a1c0385c

DlgEdObj::_propertyChange() should have same exception spec. as
XPropertyChangeListener::propertyChange().

Change-Id: I8d5b628b7cc382dae55a9e17bf813fba647453eb

Change-Id: I38ff790419562cf0e4f2b584696867eaf4614166

Change-Id: I5e41337705bb2952a98e7e5778a217cfb2caccdb

Change-Id: Iab669a18ed990a23c77b5bb3d6b0a935a8af2c5d

Change-Id: I85b0d880b0579f33a6e874ca250d343f7761dbcc

...so drop "static" from all Clang for now (and revert back to OUStringLiteral)

Change-Id: I33a3fca7f529005529f083d3e89ddfbd80bbc9bb

Change-Id: Idad25f82d884c40c55b886fd7634a6c6e51693c4

The fix for Bug 76611 caused ~20% performance loss in XLS import.
With optional cloning of ScTokenArray of the shared XLS formulas
depending on the need of address wrapping, it is possible to gain
back near the original performance.

Note: The original patch for Bug 76611 has already got an unit test,
too (see wrapped-refs.xls), and that unit test works with this
improved patch, too.

Change-Id: Ibfb59d1543ef9c4b8a075d5c4e37f77ab451aaa0

...so that LibreOffice.app dmgs built with --enable-macosx-code-signing with an
appstore-enabled identity will hopefully no longer be rejected on Mac OS X >=
10.9.5 as "'soffice' can't be opened because the identity of the developer
cannot be confirmed."  (Which I cannot verify for lack of an appstore-enabled
certificate, though.)

First of all, do not ignore errors from calls to codesign utitlity.  Really.

That reveals that soffice cannot be signed as soon as it is linked, as it
requires all the other stuff in the app to be already signed.  So just don't
sign it after linking, it will be signed last step in macosx-codesign-app-bundle
anyway.

Second, --resource-rules exemptions are no longer allowed per
<https://developer.apple.com/library/mac/technotes/tn2206/_index.html> "OS X
Code Signing In Depth."

Third, the handful of remaining shell scripts in MacOS/ need to be signed too.
(Signing them adds extended attributes to the files.)

Unfortunately, as discussed at
<http://porkrind.org/missives/mac-os-x-codesigning-woes/> "Mac OS X codesigning
woes," "hdiutil makehybrid" drops extended attributes from the generated dmg (so
the dmg's LibreOffice.app would no longer be considered properly signed, as the
shell scripts would no longer be signed).  So switch from "hdiutil makehybrid"
to "hdiutil create."

Change-Id: I4b587f87d504666f7a1d0e3a24a8be76f22014c5

Change-Id: I3a7023dba7621f9bd066edb9c0894df48313b117

The Document Provider framework will implement access to network
services in the future, and network operations must be run in separate
threads.

Change-Id: Ifcba8f28cc4dbffaf2946c12749aa38d0eb11923

The Document Provider framework will implement access to network
services in the future, and network operations must be run in separate
threads.

The Document Provider framework will implement access to network
services in the future, and network operations must be run in separate
threads.

64-bit floating point became an optional part of OpenCL core in 1.2 and is no
longer an extension. The pragma is not needed in 1.2 and later. It causes a
warning that is indeed harmless, but still clutters the compiler diagnostic
output that is displayed when there is some actual error in generated OpenCL
code.

Change-Id: Id8ea4b39f7e8e788d7683f8b389f53eadc57ea18

Change-Id: Ic94fceef0d4291a02bc3da7bcbeeb2c19670200c

Change-Id: I38b2ecf86d4334d0179362079a216df4301bf184

mnUsedSize was never decreased causing the cache to pretending to be full,
which caused lots of needless work (image loading, id creation (which are
basically a crc of all bytes of an image, ...))

Change-Id: I1e0a6a07631435346c3509822fae43a067d33303
Reviewed-on: https://gerrit.libreoffice.org/14389Reviewed-by: Michael Stahl <mstahl@redhat.com>
Reviewed-by: Zolnai Tamás <zolnaitamas2000@gmail.com>
Tested-by: Zolnai Tamás <zolnaitamas2000@gmail.com>

As drawinglayer::primitive2d::GraphicPrimitive2D constructor expects a
GraphicObject ref, we can simply pass in rGrfObj instead of building a
(implicit) temp GraphicObjet using the GetGraphic() result.

Change-Id: I8c54f20804390c773fdf9406f534787ce7143e84
Reviewed-on: https://gerrit.libreoffice.org/14390Tested-by: Michael Stahl <mstahl@redhat.com>
Reviewed-by: Michael Stahl <mstahl@redhat.com>

The UI FDIST is LEGACY.FDIST in ODFF and is the same in Excel. Only the
"real" FDIST has 4 parameters, which is F.DIST in Excel.

Change-Id: I0b926b51a3cc7282a54cce448e46fed6be3c61fa

Specifically, the record EMR_EXTSELECTCLIPRGN specifies the default
clip region if the RegionMode field is set to RGN_COPY.
See EMF specification section 2.3.2.2 available from
http://msdn.microsoft.com/en-us/library/cc230624.aspx

A unit test had to be changed for this - instead of checking for
a specific clip region, it now checks that no clip region is
specified. This is under the assumption that the default clip
region for our device context is "show everything" - i.e. no clip.

Note also that RGN_COPY seems to be a common mode value for this
record type.

Change-Id: I7bd4fe305dda184d121465005fe09d3c113e3063

Not sure if it makes sense to keep having OpDiv a subclass of Reduction. There
is no DIV() function that would take a range of cells, so it isn't really
comparable to the other Reducion subclasses. But let's keep that as it is for
now.

We need to handle three cases specially in the OpenCL: Dividing by an empty
cell which should produce an #DIV/0! error, dividing an empty cell by zero
which also should produce #DIV/0!, and dividing an empty cell with anything
else number which should produce 0.

Change-Id: I86d86f652047d6f9e3c095c3ef135a8f5396b000