The static_buffer optimization seems to come from the time where the
on-heap buffer was allocated using global memory operations. At this
point the temporary buffer and temporary string storage all come from
the evaluation stack (that can be partially allocated on heap...), so
the extra logic isn't relevant for performance.
In compact mode, we currently can not support zero-allocation moves
since some pointer assignments required during the move need to allocate
hash table slots.
This is mostly applicable to xml_document_struct::first_child, since the
pointer to this element is used as a hash table key, but there are some
contrived cases where parents of root's children need a hash slot and
didn't have it before.
These cases can be fixed by changing the compact encoding to be a bit
more move friendly, but for now it's easier to handle the error and
throw/return during move.
When this happens, the source document doesn't change.
After move some nodes in the hash table can have keys that point to
other; this makes the table somewhat larger but this does not impact
correctness.
The reason is that for us to access a key in the hash table, there
should be a compact_pointer/string object with the state indicating that
it is stored in a hash table, and with the address matching the key. For
this to happen, we had to have put this object into this state which
would mean that we'd overwrite the hash entry with the new, correct
value.
When nodes/pages are being removed, we do not clean up keys from the
hash table - it's safe for the same reason, and thus move doesn't
introduce additional contracts here.
This change implements the initial version of move construction and
assignment support for documents.
When moving a document to another document, we always make sure move
target is in "clean" state (empty document), and proceed by relocating
all structures in the most efficient way possible.
Complications arise from the fact that the root (document) node is
embedded into xml_document object, so all pointers to it have to change;
this includes parent pointers of all first-level children as well as
allocator pointers in all memory pages and previous pointer in the first
on-heap memory page.
Additionally, compact mode makes everything even more complicated
because some of the pointers we need to update are stored in the hash
table (in fact, document first_child pointer is very likely to be there;
some parent pointers in first-level children will be using
compact_shared_parent but some won't be) which requires allocating a new
hash table which can fail.
Some details of this process are not fully fleshed out, especially for
compact mode; and this definitely requires many tests.
It's not clear whether we still need PUGI__MSVC_CRT_VERSION, but it's
more consistent for now to use it for _snprintf_s since this is relying
on a CRT extension, not on a compiler feature.
These functions were deprecated via comments in 1.5 but never got the
deprecated attribute; now is the time!
Using deprecated functions produces a warning; to silence it, this
change moves the relevant tests to a separate translation unit that has
deprecation disabled.
The macro only works correctly when the input argument is an array with
a statically known size - pointers or arrays decayed to pointers won't
work silently.
While this is unlikely to surface issues that aren't caught in
tests/code review, use _countof for MSVC to prevent such code from
compiling.
Instead of branching code at each invocation site, use variadic macros
to create a wrapping macro that use snprintf for the buffer of a
statically known size.
Variadic macros are supported by all C++11 compilers, as is snprintf;
on MSVC 2005+ we don't necessarily have snprintf, but we can use
_snprintf_s with _TRUNCATE to get the same behavior. In all other cases
we fall back to sprintf, that (theoretically) can lead to a stack buffer
overflow.
In practice all snprintfs used in pugixml use buffers that should be
large enough to never be overflown but snprintf is safe even if this is
not the case.
We use references to arrays elsewhere in the codebase and there's just
one caller for this function so it's easier to fix the size.
This will simplify snprintf refactoring.
Integer sanitizer is flagging unsigned integer overflow in several
functions in pugixml; unsigned integer overflow is well defined but it
may not necessarily be intended.
Apart from hash functions, both string_to_integer and integer_to_string
use unsigned overflow - string_to_integer uses it to perform
two-complement negation so that the bulk of the operation can run using
unsigned integers. This makes it possible to simplify overflow checking.
Similarly integer_to_string negates the number before generating a
decimal representation, but negating is impossible without unsigned
overflow or special-casing certain integer limits.
For now just silence the integer overflow using a special attribute;
also move unsigned overflow into string_to_integer from get_value_* so
that we have fewer functions marked with the attribute.
Fixes#133.
This reverts commit 79109a8546.
This warning does not happen on gcc-4.8.4; the workaround introduces an
unsigned integer overflow which results in a runtime error when compiled
with integer sanitizer.
This is accomplished by putting a // fallthrough
comment at the right place.
This seems to be more portable than an attribute-based
solution like [[fallthrough]] or __attribute__((fallthrough)).
Instead of a separate implementation for find/insert, use just one that
can do both. This reduces the code size and simplifies code coverage;
the resulting code is close to what we had in terms of performance and
since hash table is a fall back should not affect any real workloads.
Instead of a complicated partitioning scheme that tries to maintain the
equal area in the middle, use a scheme where we keep the equal area in
the left part of the array and then move it to the middle.
Since generally sorted arrays don't contain many duplicates this extra
copy is not too expensive, and it significantly simplifies the logic and
maintains good complexity for sorting arrays with many equal elements
nonetheless (unlike Hoare partitioning).
Instead of a median of 9 just use a median of 3 - it performs pretty
much identically on some internal performance tests, despite having a
bit more comparisons in some cases.
Finally, change the insertion sort threshold to 16 elements since that
appears to have slightly better performance.
The previous implementation opted for doing two comparisons per element
in the sorted case in order to remove one iterator bounds check per
moved element when we actually need to copy. In our case however the
comparator is pretty expensive (except for remove_duplicates which is
fast as it is) so an extra object comparison hurts much more than an
iterator comparison saves.
This makes sorting by document order up to 3% faster for random
sequences.
Instead of relying on a specific string in the parse result, use
allocator error state to report the error and then convert it to a
string if necessary.
We currently have to manually trigger the OOM error in two places
because we use global allocator in rare cases; we don't really need to
do this so this will be cleaned up later.
The code works fine regardless of the *j->name check, and omitting this
makes the code more symmetric between the "count" and "write" stage;
additionally this improves coverage - due to how strcpy_insitu works
it's not really possible to get an empty non-NULL name in the node.
Instead of rolling back the allocation and trying to allocate again,
explicitly handle inplace reallocate if possible, and allocate a new
block otherwise.
This is going to be important once we use reallocate_nothrow from a
non-throwing context.
This allows us to gradually convert exception handling of out-of-memory
during evaluation to a non-throwing approach without changing the
observable behavior.
W3C specification does not allow predicates after abbreviated steps.
Currently this results in parsing terminating at the step, which leads
to confusing error messages like "Invalid query" or "Unmatched braces".
Any time an allocation fails xpath_allocator can set an externally
provided bool. The plan is to keep this bool up until evaluation ends,
so that we can use it to discard the potentially malformed result.
For both allocate and reallocate, provide both _nothrow and _throw
functions; this change renames allocate() to allocate_throw() (same for
reallocate) to make it easier to change the code to remove throwing
variants.
Propagate the failure to the caller manually. This is a first step to
parser structure that does not depend on exceptions or longjmp for error
handling (and thus matches the XML parser). To preserve semantics we'll
have to convert error code to exception later.
