- 25 8月, 2010 16 次提交
-
-
由 Marek Safar 提交于
-
由 Marek Safar 提交于
-
由 Mark Probst 提交于
Our previous inline card table write barrier did not perform any checks on the pointers, which means that it marked cards for the major heap even if the references stored there were not to nursery objects. Doing that is slightly faster in the mutator but the nursery collections might take a lot longer because objects might have to scanned unnecessarily. This patch implements an inline write barrier that also performs a nursery check. To that end we introduce a new IL opcode, OP_CARD_TABLE_WBARRIER. It is currently only implemented on x86. For architectures that don't support it (yet), we automatically regress to the old, non-checking write barrier.
-
由 Atsushi Eno 提交于
-
由 Mark Probst 提交于
We didn't update the heap boundaries correctly in the fixed-heap mark&sweep collector, which could lead to pins being lost.
-
由 Mark Probst 提交于
There was still an issue with thread-safety of binary protocols: If the world was stopped while a thread was in the process of using a binary protocol buffer, the GC would write all buffers to the file and release the memory, making the thread crash once it restarted. We fix this here by keeping a count of how many threads currently use the buffers. If they are in use when we try to save and free them, we don't. They'll be saved and freed eventually.
-
由 Atsushi Eno 提交于
-
由 Rodrigo Kumpera 提交于
-
由 Rodrigo Kumpera 提交于
* mini.c (create_jit_info): Handle the case where the basic block right after a try block is dead. For example: .try { .... leave ... } nop //HERE .catch { ... } The above nop is dead code and the branch opts will nullify it which caused the code in create_jit_info to assert when calculating the try block lenght. This is a very rare case since only obfuscators or broken compilers would put instruction between the try and handler blocks. The fix is to simply iterate over all CIL addresses of the try block in reverse order looking for the first block which was not removed. Fixes 633291
-
由 Zoltan Varga 提交于
Rename a few functions to better reflect their purpose, Use free_up_hreg in more places.
-
由 Zoltan Varga 提交于
-
由 Robert Nagy 提交于
-
由 Gonzalo Paniagua Javier 提交于
-
由 Jérémie Laval 提交于
-
由 Jérémie Laval 提交于
-
由 Jérémie Laval 提交于
-
- 24 8月, 2010 24 次提交
-
-
由 Zoltan Varga 提交于
This was only needed because of the buggy assemblyref in mscorlib to itself. This reverts commit 77d0ee57.
-
由 Zoltan Varga 提交于
the last change to dwarfwriter.c.
-
由 Zoltan Varga 提交于
-
由 Zoltan Varga 提交于
-
由 Zoltan Varga 提交于
-
由 Jérémie Laval 提交于
-
由 Jérémie Laval 提交于
-
由 Jérémie Laval 提交于
-
由 Jérémie Laval 提交于
Since the call is also implemented as a simple long load operation in the runtime when on x86-64 there is no need to do the differentiation in managed code.
-
由 Jonathan Pryor 提交于
No mdoc changes, but since 'mdoc update --exceptions=all' scans across assemblies, and since System.Delegate.Combine() now throws additional exceptions (according to our IL scanning algorithm), the expected output differs, leading to false negatives about regressions. Furthermore, 'gmcs /doc' has been fixed (it used to cause a compilation error) and improved, so more XML documentation is now imported because gmcs is generating the correct XML documentation crefs. (Go gmcs, go!)
-
由 Jonathan Pryor 提交于
Fixes #624242 This also allows master 'gmcs /doc' to build OpenTK, gitorious.org/cadenza, and build the mcs/tools/mdoc unit tests w/o compilation error. The bug was caused by a NotSupportedException when calling ArrayContainer.Name within DocUtil.GetSignatureForDoc(TypeSpec). This is directly fixed by checking explicitly for ArrayContainer. Another bug fixed was wrt type parameter indexing in nested types, hence "counting" the number of type parameters declared in all declaring types.
-
由 Marek Safar 提交于
-
由 Marek Safar 提交于
-
由 Atsushi Eno 提交于
-
由 Marek Safar 提交于
-
由 Atsushi Eno 提交于
-
由 Atsushi Eno 提交于
Disable separate loop for HttpListener context acquisition. It prevents HttpStandaloneChannelListener from closing itself.
-
由 Marek Safar 提交于
-
由 Marek Safar 提交于
-
由 Marek Safar 提交于
-
由 Marek Safar 提交于
-
由 Marek Safar 提交于
-
由 Miguel de Icaza 提交于
-
由 Marek Habersack 提交于
When a server-side control processes its children (in the markup) as properties, it needs to instantiate an object of the property type and assign it to the appropriate property of the control. Failing to do so will cause markup which references properties to fail in certain cases (for instance when the control's property type is a class and it is not instantiated by the control's constructor).
-