Use alternative mechanisms to allow us to use the default new/delete implementations #101947
Conversation
…nt to allow us to use the standard allocators.
… for OOM logging and we can handle std::bad_alloc
| @@ -27,9 +27,9 @@ | |||
| // case we need to wrap memory allocations, in the latter there is no | |||
| // infrastructure to support this. Detect which way we're building and provide a | |||
| // very simple abstraction layer (handles allocating bytes only). | |||
| #ifdef _BLD_CLR | |||
There was a problem hiding this comment.
The comment above is stale after removing this ifdef.
src/coreclr/vm/ceemain.cpp
Outdated
| } | ||
| else | ||
| { | ||
| throw std::bad_alloc(); |
There was a problem hiding this comment.
Hmm, this makes me wonder - couldn't we just make this handler throw our OutOfMemoryException here instead of making all the changes in the exception handling macros?
There was a problem hiding this comment.
We can't as the new handler is technically process-global. So with this change, we would technically report more OOMs (as we'd report an OOM with the stack from ICU for example) but we wouldn't break a C++ library that is expecting to catch std::bad_alloc.
There was a problem hiding this comment.
Ah, makes sense, thank you for the explanation.
There was a problem hiding this comment.
We can't as the new handler is technically process-global.
Does that mean that some other library loaded in the process can break us, or vice versa?
There was a problem hiding this comment.
I am wondering whether it would be better to drop this stresslogging. I do not think it is particularly important.
There was a problem hiding this comment.
I agree that is it not particularly important. Both from dump and live debugging, it will be clear that there was an OOM, so having it in the stresslog doesn't seem to provide any benefit.
There was a problem hiding this comment.
I'll remove it from this path then.
| @@ -32,9 +32,10 @@ set(SOURCES | |||
|
|
|||
| include_directories(../../../coreclr/pal/prebuilt/inc) | |||
|
|
|||
| add_compile_definitions(SOS_INCLUDE) | |||
There was a problem hiding this comment.
This define can use a better name.
There was a problem hiding this comment.
I'll come back to this and clean this up in a future PR.
set_new_handler.std::bad_allocand translate it to our pre-allocatedOutOfMemoryException.With these changes, we can switch to the global
newanddeleteimplementations provided by the C++ Standard Library.Also remove
_BLD_CLRdefine as we always define it in the product. Adjust the one test native binary that was using the headers without defining it.