Koromix
commented
1 year ago To implement C callbacks in dynamic FFI code, you need to use trampolines, which are small functions that receive the call and redirect it to Koffi.
The only thing Koffi (and other FFI libraries) can use to differentiate two callbacks is the address of the call itself. You don't control the C code and so you can't use any kind of argument; only the raw address of the callback function.
Given these limitations, there are two ways to provide callbacks in dynamic FFI libraries:
- You can generate a trampoline at runtime, by writing some machine code to memory, which will then redirect the call to Koffi with the correct callback information.
- You can use a predefined static set of trampolines, and keep the info about the callback in a global array. So when trampoline 12 gets called, it looks at callbacks[12] to get all the information needed to call into JS.
The former option is more complex and can fail on some platforms that prevent remapping writeable memory to executable memory (such as Linux PaX builds, OpenBSD, etc.). So Koffi uses the latter approach and has a set of hard-coded 1024 trampolines for FFI callbacks.
So you can't configure it, but I can increase this limit. The main downside is bigger binaries.
wanglin86769
EPICS PV (Process Variable) values can be subscribed to monitor instead of poll, and for each PV, two callbacks are needed for monitoring, they are connection state change callback and value change callback, which means only 512 PVs can be monitored at the same time with koffi. However, there are usually more than 100000 PVs in the EPICS control system of a particle accelerator facility.
Why does koffi restrict callback count? Is it possible to increase the count or make it configurable?
EPICS home page is as follows, https://epics.anl.gov/