Problem was that the file-to-id cache was cleared
upon calling internal Ruby functions (begin_exec
was triggered). In one test case, the execution
performance improved from 6s/320s (without/with debugger)
to 6s/7s (without/with debugger)
Problem was that destruction of database objects was postponed during
callbacks, specifically the interpreter callback.
This caused db::Region objects to pile up.
The original feature was introduced to prevent issues when Ruby/C++
code interacts inside Qt UIs. The new fix does not disable GC, but
protects Ruby objects during method calls. Maybe that solves the problem too.
Goal is to support "*!" which is a variant of multiplication
without the commutative nature.
The refactoring yields a more consistent handling of
Python specializations. Now this happens when the methods
are collected, rather than later. This way, specific behavior
for identically named synonyms can be implemented for example.
This is the case for the "*" operator in Trans which is partially
commutative and partially not.
* Fixed issue #1029
The problem happened because in the described case the
tl::Variant used as a intermediate container holds the
Python QBrush object and when it gets deleted, the
QBrush object is deleted too.
* Added tests
Needed to represent QIODeviceBase as a second base class for
QIOdevice, QFile etc. Additional base classes are supported but
in a rather limited way. They only contribute constants like enums.
No methods can be provided this way (a limitation of the method
enumeration scheme which only supports one base classe)
* Fixed the issue - Byte array variant was not properly converted to Python/Ruby objects.
* Added tests + properly converting byte arrays to byte array variants in Python.
On MSVC2017 builds, a crash is observed in the RBA basic
tests. There is no obvious reason, but empirically, the
instantation of a std::vector<VALUE> caused this problem.
Compiler bug or strange interaction with Ruby's GC?
