We now have the --enable-canonical-installation-tree-structure option
which does what was suggested.
Change-Id: I41215275d6d8c8f11b4a45390d44998c967fcc53
PythonTests are run on a dev-install tree where (when using code
signing on OS X) the Python framework has already been signed, so
scribbling bytecode cache files into it will invalidate the signature.
Change-Id: Ic53043b59c9a59373e2383d8dcca9120457d4ba9
SQLFetchScroll uses the bound columns, and writes to them.
This has two problems:
1) For non-bookmark columns, this happens in our buffers,
and at first sight could be innocuous.
However, it allows the driver to disallow us to use
SQLGetData on the same columns, which we intend to do...
2) For bookmark column, the "read bookmark from" and
"write bookmark to" buffers end up being the same.
This unnecessarily exposes us to driver bugs in this corner case.
Change-Id: I594d34cf5a64564766f06b49c86f8737961573c0
the buffer (aBookmark) is written to by SQLBulkOperations or
SQLSetPos, not by SQLBindCol, which merely sets up the instruction
that the next call should write there. So its finak size is known
only *after* the call that makes the write, obviously.
Change-Id: I4faaaeffc060433645188f0eda3d9958b087cf6c
This code was completely inverted. The row update operation
*reads* a bookmark (that is, what row to update), not *writes*
a bookmark.
So we were passing an empty bookmark, and thus the update was failing
because we were refering to a non-existent row.
Change-Id: I676b1a7727a88e13a3e465bd96cbbaf18dad2fa6
We have to register them one at a time, like windows, this means we detect our
bundled opensymbol which makes the "format->bullets and numbering" preview work
again.
Doesn't however fix fdo#68192 yet, more magic smoke required.
Change-Id: I430c91b48d96fb8d62e3b073898b72cf2b8ddea6
Moves the cpu intensive code to ScMatrix, where we can use our
knowledge of the internal structure: use mdds::multi_type_matrix::walk
instead of browsing cells one by one.
Change-Id: Ie1df20e2be6414b8e21b4d58b7697a0801222c1e
Reviewed-on: https://gerrit.libreoffice.org/5455
Reviewed-by: Kohei Yoshida <kohei.yoshida@suse.de>
Tested-by: Kohei Yoshida <kohei.yoshida@suse.de>
A value of 0 for Opcode eOp produces a debug assertion that is
not in place as opcode with value does exist (ocPush, SC_OPCODE_PUSH).
Change-Id: If529cc2ecd8d4d3a5e31f084dfa72e13c9c0fe37
Reviewed-on: https://gerrit.libreoffice.org/5586
Reviewed-by: Thorsten Behrens <tbehrens@suse.com>
Tested-by: Thorsten Behrens <tbehrens@suse.com>
We try and copy the shared one to the user location the first time we create a
customized user file. If the dir doesn't exist then that migration doesn't
happen. The new autocorr file is then written, creating the dir, and any
further operations will reattempt the skipped share copy, and this time succeed
because the dir now exists. Which overwrites the previously written customized
content.
This is similar to fdo#65501 "ensure configured backup dir exists before using
it", so reuse that code and move it into unotools.
Change-Id: I38fa621b8d7295d70b09172a028569ea95404120
...they had originally been
aURL.equalsIgnoreAsciiCaseAsciiL( "vnd.sun.star.pkg", 16 )
etc., so where likely confused with the OUString compareTo functions that take a
maxLength argument.
Change-Id: Ie12df4f589dda310b7e49eb93535ad797f88a8a7
...(like is done for most of LO's non-URE libs already) to reduce likelihood of
name clashes, esp. on Windows where URE libs are found via PATH.
This introduces PRIVATELIBS_URE, and removes now-unused UNOLIBS_URE.
Change-Id: Ib95dd45f18de140a54e62d632dbf2239f83c232e