| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
* we were basing the glib allocators on throwing std::bad_alloc just like
the C++ operators. However, this always was unsafe since we were throwing
exceptions across plain-C frames (Glib).
Also, the memory vtable has been deprecated in Glib, resulting in
ugly warnings.
* Instead, we now let the C++ new/delete operators work like Glib
by basing them on g_malloc/g_slice.
This means they will assert and the application will terminate
abnormally in case of OOM. OOMs cannot be handled properly anyway, so it is
more important to have a good memory limiting mechanism.
* Memory limiting has been completely revised.
Instead of approximating undo stack sizes using virtual methods
(which is unprecise and comes with a performance penalty),
we now use a common base class SciTECO::Object to count the memory
required by all objects allocated within SciTECO.
This is less precise than using global replacement new/deletes
which would allow us to control allocations in all C++ code including
Scintilla, but they are only supported as of C++14 (GCC 5) and adding compile-time
checks would be cumbersome.
In any case, we're missing Glib allocations (esp. strings).
* As a platform-specific extension, on Linux/glibc we use mallinfo()
to count the exact memory usage of the process.
On Windows, we use GetProcessMemoryInfo() -- the latter implementation
is currently UNTESTED.
* We use g_malloc() for new/delete operators when there is
malloc_trim() since g_slice does not free heap chunks properly
(probably does its own mmap()ing), rendering malloc_trim() ineffective.
We've also benchmarked g_slice on Linux/glib (malloc_trim() shouldn't
be available elsewhere) and found that it brings no significant
performance benefit.
On all other platforms, we use g_slice since it is assumed
that it at least does not hurt.
The new g_slice based allocators should be tested on MSVCRT
since I assume that they bring a significant performance benefit
on Windows.
* Memory limiting does now work in batch mode as well and is still
enabled by default.
* The old UndoTokenWithSize CRTP hack could be removed.
UndoStack operations should be a bit faster now.
But on the other hand, there will be an overhead due to repeated
memory limit checking on every processed character.
|
|
implementation classes
* whenever the implementation class was not exactly RBEntryType,
it had to have a virtual destructor since RBTree cared about
cleanup and had to delete its members.
* Since it does not allocate them, it is consistent to remove RBTree::clear().
The destructor now only checks that subclasses have cleaned up.
Implementing cleanup in the subclasses is trivial.
* Consequently, RBEntryString no longer has to be virtual.
HelpIndex and GotoTables are completely non-virtual now
which saves memory (and a bit of cleanup speed).
For QRegister, not much changes, though.
|
|
* Using a common implementation in RBTreeString::auto_complete().
This is very efficient even for very huge tables since only
an O(log(n)) lookup is required and then all entries with a matching
prefix are iterated. Worst-case complexity is still O(n), since all
entries may be legitimate completions.
If necessary, the number of matching entries could be restricted, though.
* Auto completes short and long Q-Reg names.
Short names are "case-insensitive" (since they are upper-cased).
Long specs are terminated with a closing bracket.
* Long spec completions may have problems with names containing
funny characters since they may be misinterpreted as string building
characters or contain braces. All the auto-completions suffered from
this problem already (see TODO).
* This greatly simplifies investigating the Q-Register name spaces
interactively and e.g. calling macros with long names, inserting
environment registers etc.
* Goto labels are terminated with commas since they may be part
of a computed goto.
* Help topics are matched case insensitive (just like the topic
lookup itself) and are terminated with the escape character.
This greatly simplifies navigating womanpages and looking up
topics with long names.
|
|
* the old implementation tried to avoid template programming by
making the entry comparison function virtual.
* The new RBTree implementation takes a template argument with the
implementation of RBEntry. It is now partially conventional
that the template argument must be actually derived from RBTree::RBEntry
and must define a "compare" method.
* As an advantage we now get static polymorphism (avoiding virtual
calls and allowing for more compiler optimizations) and the
the RBEntry implementation no longer has to be virtual.
* The only RB-Trees actually used are string-keyed, though.
Therefore there's a common base class RBTreeString now which
defines two synonymous "key" and "name" attributes.
* The entry base class RBEntryString is virtual again because
we do not want to propagate the RBEntryType template parameter
even further and the RBTree base class needs to destroy
entries.
This might be avoided by not defining a RBTree::clear() method,
leaving this task to the implementations.
At least QRegisters have to be virtual, though.
* RBTreeString only depends on the strcmp() and strncmp() functions
used now and only case-sensitive and case-insensitive versions
are actually required, so we instantiate these templates statically
in rbtree.cpp.
This means there are still only two instantiations of the RBTree
in the binary.
* RBTreeString defines convenient wrappers for find() and nfind()
to look up by string.
This uses the RBEntryString base class, so no allocations whatsover
are required for lookups and less space is wasted on the call stack.
* A RBEntryOwnString base class is also provided which frees the
implementations from memory managing the tree keys.
* RBTreeString can now be used to add other common functionality
like auto-completions for Q-Registers, goto labels and help topics.
* some minor optimizations
* updated TODO
|
|
batch mode
* by using variadic templates, UndoStack::push() is now responsible
for allocating undo tokens. This is avoided in batch mode.
* The old UndoStack::push(UndoToken *) method has been made private
to avoid confusion around UndoStack's API.
The old UndoStack::push() no longer needs to handle !undo.enabled,
but at least asserts on it.
* C++11 support is now required, so variadic templates can be used.
This could have also been done using manual undo.enabled checks;
or using multiple versions of the template with different numbers
of template arguments.
The latter could be done if we one day have to support a non-C++11
compiler.
However since we're depending on GCC 4.4, variadic template use should
be OK.
Clang supports it since v2.9.
* Sometimes, undo token pushing passed ownership of some memory
to the undo token. The old behaviour was relied on to reclaim the
memory even in batch mode -- the undo token was always deleted.
To avoid leaks or repeated manual undo.enabled checking,
another method UndoStack::push_own() had to be
introduced that makes sure that an undo token is always created.
In batch mode (!undo.enabled), this will however create the object
on the stack which is much cheaper than using `new`.
* Having to know which kind of undo token is to be pushed (taking ownership
or not) is inconvenient. It may be better to add static methods to
the UndoToken classes that can take care of reclaiming memory.
* Benchmarking certain SciTECO scripts have shown 50% (!!!) speed increases
at the highest possible optimization level (-O3 -mtune=native -march=native).
|
|
|
|
* acts as a safe-guard against uninterrupted infinite loops
or other operations that are costly to undo in interactive mode.
If we're out of memory, it is usually too late to react properly.
This implementation tries to avoid OOMs due to SciTECO behaviour.
We cannot fully exclude the chance of an OOM error.
* The undo stack size is only approximated using the
UndoToken::get_size() method.
Other ways to measure the exact amount of allocated heap
(including size fields in every heap object or using sbrk(0) and
similar) are either costly in terms of memory or platform-specific.
This implementation does not need any additional memory per heap
object or undo token but exploits the fact that undo tokens
are virtual already. The size of an undo token is determined
at compile time.
* Default memory limit of 500mb should be OK for most people.
* The current limit can be queried with "2EJ" and set with <x>,2EJ.
This also works interactively (a bit tricky!)
* Limiting can be disabled. In this case, undo token processing
is a bit faster.
* closes #3
|
|
* also did some whitespace cleanup in SciTECO now that tabs are
displayed properly
|
|
|
|
minor change that fixes Doxygen-generated documentation
|
|
normally, since SciTECO is not a library, this is not strictly
necessary since every library should use proper name prefixes
or namespaces for all global declarations to avoid name clashes.
However
* you cannot always rely on that
* Scintilla does violate the practice of using prefixes or namespaces.
The public APIs are OK, but it does define global functions/methods,
e.g. for "Document" that clashed with SciTECO's "TECODocument" class at
link-time.
Scintilla can put its definitions in a namespace, but this feature
cannot be easily enabled without patching Scintilla.
* a "SciTECO" namespace will be necessary if "SciTECO" is ever to be
turned into a library. Even if this library will have only a C-linkage
API, it must ensure it doesn't clutter the global namespace.
So the old "TECODocument" class was renamed back to "Document"
(SciTECO::Document).
|
|
|
|
* specifications resulted in runtime errors (unexpected exception) when bad_alloc ocurred
* specs should be used scarcely: only when the errors that may be thrown are all known
and for documentary purposes
|
|
|
|
|
|
|
