There's no point in continuing to write other parts if the final result
will be a failure anyway. Moreover this avoids an assert
in ZipOutputStream::writeLOC() if writing of the previous part resulted
in an error (e.g. ZipException because of broken zip CRC for the stream)
that skipped calling ZipOutputStream::rawCloseEntry().
Change-Id: I5095b97a31cac9befcab5e82bd8cda2dfa53c7f7
Reviewed-on: https://gerrit.libreoffice.org/73646
Tested-by: Jenkins
Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
ZipPackageStream::saveChild() already had one threaded compression,
but that still uses only one thread for one stream. Many documents
contain many streams (where this is useful), but large documents
often contain one huge content.xml, which then would be compressed
using just one thread.
But it is in fact possible to do deflate in parallel on the same data,
at the cost of somewhat increased CPU usage (spread over threads).
This is handled separately from the background thread path, as
integrating these two approaches would probably be needlessly complex
(since they both internally use ThreadPool, the tasks should often
intermix and parallelize anyway).
On my 4-core (8 HT threads) machine this reduces the compression time
of tdf#113042 from 3s to 1s.
Change-Id: Ifbc889a27966f97eb1ce2ce01c5fb0b151a1bdf8
Reviewed-on: https://gerrit.libreoffice.org/73032
Tested-by: Jenkins
Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
It can be easily separated out, it looked like hacked in. And
I will need to do more refactoring of the class, so this shouldn't
be more complex than necessary.
Change-Id: I302da55409e9195274907ca4939c37fbb2427b18
Reviewed-on: https://gerrit.libreoffice.org/73031
Tested-by: Jenkins
Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
E.g. with the bugdoc from tdf#93553, most of the streams to save
are actually small, they just do not provide XSeekable. Since it
seems all the streams are already fully available by the time
the zip packaging is done, fallback to XInputStream::available()
to find out if the stream is small.
With this, the zipping part of saving tdf#93553 is now down
from ~1.5s to ~0.5s.
This presumably also makes the hack for tdf#93553 unnecessary.
Change-Id: Id9bb7d835400d6d8f147f5c11ade684a549aba53
Reviewed-on: https://gerrit.libreoffice.org/73030
Tested-by: Jenkins
Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
A threadpool controls a number of threads that execute a number
of thread *tasks* from a queue. The API even says they are tasks.
So it's damn confusing when ZipPackageStream::saveChild()
claims to limit the number of threads to 4x the maximum number
of threads. It limits the number of queued thread tasks.
Change-Id: I317497f27a82d92a7c8566b14aaeae73a4ffef1f
Reviewed-on: https://gerrit.libreoffice.org/72677
Tested-by: Jenkins
Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
regression from
commit 306758ab3e
Date: Fri Apr 5 15:40:27 2019 +0200
tdf#117066 Saving ODT document with ~1500 bookmarks is slow, part 5
Before the above commit, we could have multiple child items with the
same name. Restore that behaviour, while keeping the fast lookup, by
using a std::unordered_map<OUString, std::vector<...
Also remove name from SotElement_Impl so there is no chance of the name
in the key of the map, and the name in the element getting out of sync.
Change-Id: I65c294ddc409d9b8a7006e4f4338c9a2a4446a92
Reviewed-on: https://gerrit.libreoffice.org/72544
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Similar to clang-tidy readability-container-size-empty
Change-Id: Idd67f332b04857a39df26bad1733aae21236f105
Reviewed-on: https://gerrit.libreoffice.org/71764
Tested-by: Jenkins
Reviewed-by: Michael Stahl <Michael.Stahl@cib.de>
if we're doing a find/insert on a set or a map, it is better to just do
a conditional insert/emplace operation than triggering two lookups.
Change-Id: I80da5097f5a89fe30fa348ce5b6e747c34287a8d
Reviewed-on: https://gerrit.libreoffice.org/70937
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Individually, these don't make much difference, but they add up
to a halving the time to save on my machine.
OStorageImpl is spending time iterating over its m_aChildrenVector to
find stuff by name, so just use a std::unordered_map.
Also return iterator from FindElement, so we can avoid searching the map
twice.
This was probably the biggest win.
Change-Id: I30776bad5377d14144fc7a47e86527e2cdb62a83
Reviewed-on: https://gerrit.libreoffice.org/70313
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Individually, these don't make much difference, but they add up
to a halving the time to save on my machine.
ManifestImport::characters was spending time adding data to an OUString,
so convert that to an OUStringBuffer.
Change-Id: I267e701f4e7998044763f44199b1fe8a37325b68
Reviewed-on: https://gerrit.libreoffice.org/70311
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Also use scope where possible. This allows to limit guard scope at
language level; visualises the scope clearly; and helps avoiding
errors like fixed in commit 61e4437c85.
Change-Id: Ifeca96e2df8e8a0897770d9546b2536806275f41
Reviewed-on: https://gerrit.libreoffice.org/70376
Tested-by: Jenkins
Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
V572 It is odd that the object which was created using 'new' operator
is immediately cast to another type.
Change-Id: I54976062dc3f62eaaa79f89eff54454f0b24ac2c
Reviewed-on: https://gerrit.libreoffice.org/69989
Tested-by: Jenkins
Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
V572 It is odd that the object which was created using 'new' operator
is immediately cast to another type.
Change-Id: I6d1523e71b3e06be1cf41abaabb44e49fe11cd8e
Reviewed-on: https://gerrit.libreoffice.org/69369
Tested-by: Jenkins
Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
...which has no effect on the result, but silences benign Clang
-fsanitize=implicit-signed-integer-truncation warnings
Change-Id: I0953914a35f2a8c4caa6f75d4918e3b3366d07e8
Reviewed-on: https://gerrit.libreoffice.org/68628
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
Add exceptionToString() and getCaughtExceptionAsString() methods in
tools.
Use the new methods in DbgUnhandledException()
Add special-case case code for most of the exceptions that contain extra
fields, so all of the relevant data ends up in the log
Change-Id: I376f6549b4d7bd480202f8bff17a454657c75ece
Reviewed-on: https://gerrit.libreoffice.org/67857
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Since it is now possible to use C++14, it's time to replace
the temporary solution with the standard one
Change-Id: I8bee1344f7df82536f31bc5e4ec4fd379cac1d04
Reviewed-on: https://gerrit.libreoffice.org/66704
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Found with bin/find-unneeded-includes
Only removal proposals are dealt with here.
Change-Id: I22ba2c8aec235e34cd7835b8a0a716bf3057db7a
Reviewed-on: https://gerrit.libreoffice.org/60837
Tested-by: Jenkins
Reviewed-by: Miklos Vajna <vmiklos@collabora.co.uk>
...warning about (for now only) functions and variables with external linkage
that likely don't need it.
The problems with moving entities into unnamed namespacs and breaking ADL
(as alluded to in comments in compilerplugins/clang/external.cxx) are
illustrated by the fact that while
struct S1 { int f() { return 0; } };
int f(S1 s) { return s.f(); }
namespace N {
struct S2: S1 { int f() { return 1; } };
int f(S2 s) { return s.f(); }
}
int main() { return f(N::S2()); }
returns 1, both moving just the struct S2 into an nunnamed namespace,
struct S1 { int f() { return 0; } };
int f(S1 s) { return s.f(); }
namespace N {
namespace { struct S2: S1 { int f() { return 1; } }; }
int f(S2 s) { return s.f(); }
}
int main() { return f(N::S2()); }
as well as moving just the function f overload into an unnamed namespace,
struct S1 { int f() { return 0; } };
int f(S1 s) { return s.f(); }
namespace N {
struct S2: S1 { int f() { return 1; } };
namespace { int f(S2 s) { return s.f(); } }
}
int main() { return f(N::S2()); }
would each change the program to return 0 instead.
Change-Id: I4d09f7ac5e8f9bcd6e6bde4712608444b642265c
Reviewed-on: https://gerrit.libreoffice.org/60539
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
Found with bin/find-unneeded-includes
Only removal proposals are dealt with here.
Change-Id: I7bfeef47abaf94cfb355db95c0fdb928ce36c0a6
Reviewed-on: https://gerrit.libreoffice.org/60232
Tested-by: Jenkins
Reviewed-by: Miklos Vajna <vmiklos@collabora.co.uk>
comment from sberg:
aren't these changes broken in general, when the called function
may throw an exception before it takes ownership of the passed-in pointer?
So revert, except for
(a) PlainTextFilterDetect::detect, which was definitely a leak
(b) SwCursor::FindAll, where unique_ptr was being unnecessarily used
This reverts commit 7764ae70b0.
Change-Id: I555e651b44e245b031729013d2ce88d26e8a357e
Reviewed-on: https://gerrit.libreoffice.org/60301
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Tested-by: Noel Grandin <noel.grandin@collabora.co.uk>
Caolán McNamara