array, list, struct and others set() now copies the values. These
values can be created by user, in this case string is just a stack
object and not a real eina_stringshare.
To cope with it, add the string instead of referencing it. Bit slower,
but nicer behavior.
SVN revision: 67886
I did a bad decision to steal memory for Array, List, Hash and Struct
types, it was nice to not have to copy it internally, but breaks when
one needs to set a new value that was set elsewhere. What did not
happen with string, integers and other basic types.
This was exposed by Raphael Kubo using eina_model_property_set() with
complex types (Array, List and Hash) and it was not possible to
correctly set such properties.
Now it's all set, but the behavior changed and the memory is not
stolen and released anymore. Test eina_test_value.c was changed to
reflect it.
SVN revision: 67843
if user get and then set the same value, we should not crash and this
may happen with previous code as the old
string/array/value/list... were released, then you ended with the
released memory still being pointed.
SVN revision: 67841
We should not flush and then setup the memory, instead we leave
vset/pset functions do their own stuff to clean previous data, if any.
SVN revision: 67840
Let's try to help debug by allowing extended reference management that
takes in account an identifier. This identifier is accounted on xref
and xunref and must match.
xrefs_get will return the list of such references, for debugging purposes.
eina_models_list_get() was added to return all live models, just
tracked when EINA_MODEL_DEBUG is enabled.
eina_models_usage_dump() was added and use the same infrastructure as
eina_models_list_get() and eina_model_xrefs_get() to aid debugging :-)
SVN revision: 67821
He worked previously with documentation, now is helping me with
Eina_Model and will keep adding some examples and fixes here and
there.
SVN revision: 67798
Same as the previous commit about "Type" this time about Interface.
This will also hopefully remove the likelihood for hacks.
Previous functions are still available as method_offset_resolve.
SVN revision: 67786
val->value.ptr is just valid on 64bits platforms, as on 32bits both
pointers of the Eina_Value_Struct fits in the 8bytes of val->value.buf.
Always use the function to correctly retrieve it.
SVN revision: 67764
Make it possible to resolve foo (the last item).
struct _My_Object_Type
{
Eina_Model_Type parent_class;
...
void (*foo)(Eina_Model *, int);
};
SVN revision: 67733
1. consider case where the struct was not set for flush and others;
2. flush previous struct on pset;
3. setup the new struct members memory on pset;
4. always use ops->alloc and ops->free if they are provided;
SVN revision: 67662
It's interchangeable with const char *, it's juts there as a visual hint
for functions expecting stringshared values. Docs are not enough...
SVN revision: 67641
Thanks to Tasn review it was noticed that subclasses using "const void *value"
with sub-structure would have to resolve their methods manually, quite bad.
Then the extension will happen with structures inside structures, as
previously used in Evas_Smart_Class. To allow the system to resolve
the methods and keep it, it is now specificed the type_size.
A new function is added to help extensions:
eina_model_type_method_resolve(), together with few helper macros.
To allow Eina_Model_Type to be augmented with more methods without
breaking the ABI, 4 slots are left there. We expect we got it right
from beginning and these will never be used, but we can never be sure
;-)
Last but not least, similar changes were done to Eina_Model_Interface
for consistency matters.
SVN revision: 67635
It is common to already known the member, then no need to query
it. This provides optimized retrieval for people that can manage it.
SVN revision: 67538
this is a killer, should be very efficient in memory and speed to
set/get items: instead of a hash of properties, keep them in a C
struct!
The constraint is that properties have fixed types defined at compile
time and cannot be deleted, but this is expected in many cases (ie:
esql rows).
SVN revision: 67517
Don't print messages from share common, that way we have no way to
know where it happens, moreover the binshare values may lack trailing
'\0'.
Also makes no sense to share the log domain. With separate domains we
can be more selective in what we log/debug.
SVN revision: 67418
This is a simple to use generic hierarchical data access. It includes
properties, children, reference counting and notifications (callbacks).
It can be understood in the same lines as GObject or PyObject, but I
believe the best usage is just to provide data models, not generic
Object Oriented Programming in C.
It misses most of documentation, although the important bits are
there. Need help here :-/
SVN revision: 67405
they are not so expensive and will help detect bugs.
also, safety can be disabled at compile time... if performance is so
critical.
SVN revision: 67393
library files should ALWAYS include config.h, they should not include
Eina.h using "<Eina.h>" to avoid messing with system's version.
Last but definitely not least: include safety checks BEFORE your local
header, otherwise the macros EINA_ARG_NONNULL() will remove every
check for null pointers! eina_safety_checks.h redefines it to empty so
compiler does not optimize these things.
SVN revision: 67392
This removes the annoying:
make[2]: warning: jobserver unavailable: using -j1. Add `+' to parent make rule.
and with the package name I can upload it easily to somewhere else.
SVN revision: 67343