| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
32bit/64bit values and arrays are misrepresented
- PointerBuffer is used to map 64bit integer values,
which is illegal due to the latest PointerBuffer changes,
where the underlying ByteBuffer is either 32bit wide
or 64bit wide (in respect to the pointer size).
The PointerBuffer semantic itself is correct,
but no more suitable to represent 64bit values and arrays.
A Int64Buffer (LongBuffer does not exist in CDC/CVM patch)
should be used instead.
- Determine use of Int64Buffer and PointerBuffer
Assuming the 1st step being solved, it has to determined
for which cases gluegen shall map a type to a PointerBuffer.
All pointer pointer types, regardless of a Opaque mapping, ie:
+++
typedef struct __MYAPIConfig * MYAPIConfig;
Opaque long MYAPIConfig
void foo(MYAPIConfig * ptrarray);
+++
The argument 'ptrarray' must be represented as a PointerBuffer
otherwise data is misrepresented in case:
- 32bit machines _and_
- array semantics - more than one value is being used
Impl:
java/com/sun/gluegen/JavaEmitter.java:
- Checks ptr-ptr for Opaque values
- Returns PointerBuffer mapping for ptr-ptr types
- Allow PointerBuffer being mapped as String[]
Very elaborated tests .. :)
++++++++++++
Misc Changes:
- Added <Type>.put(<Type>Buffer src) for Int64Buffer/PointerBuffer
|
| |\ |
|
| | | |
|
| |/ |
|
| |\ |
|
| | |\
| | |
| | |
| | |
| | | |
Conflicts:
src/junit/com/jogamp/gluegen/test/junit/Test1.java
|
| | | | |
|
| | |/
|/|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Mixed types in generated code are functioning now,
even though not all permutations are generated - still.
However, this patch merges the indirect object passing,
wheather it is a primitive array or an indirect NIO buffer,
incl. PointerBuffer.
This allows the usage of only one JNI functions for all combinations.
Only in case of NIODirectOnly, the simplified direct only '0'
variation is created - otherwise the parametrized '1' variant.
The junit tests proves the implementation and
almost completes the gluegen junit tests coverage
for JavaEmitter and ProcAddressEmitter.
Impact/Result:
- Working mixed array types
- JOGL GL2 native library shrunk around 30%
- Simplified gluegen code
- Almost complete gluegen junit tests
TODO: Complete permutations of array/NIO arguments,
if desired.
++++
Misc changes:
- NativeLibrary implements DynamicLookupHelper:
lookupFunction() -> dynamicLookupFunction()
|
| |/ |
|
|
|
Adding 'plain' junit tests.
Plain stands for the simple processing of:
ant.junit.compile: gluegen -> java/c files, javac/cc, jar
ant.junit.run: junit batch run
Avoiding 'black magic', ie kicking off gluegen and ant-compilation
from within the junit tests.
Same methodology as the JOGL junit tests,
junit test sources are under 'src/junit'
This way, the migration to other platform tests might be easier,
as well as the we don't need to pass through ant properties (ant - junit - ant),
see 3a32650d4229f9b4ad1f527d9e30c24ddb69bb3f.
|
