William Ahern
On Wed, Apr 08, 2015 at 07:15:02PM -0700, daurnimator wrote: > Interesting tangentially related article: https://blog.sandstorm.io/news/2015-04-08-osx-security-bug.html I've long thought about making SO_OOBINLINE the default, precisely out of concern about this...
Now I remember why I hadn't implemented cqueue:close, and why I had a nagging feeling that simply reusing cqueue__gc as the close method was problematic. Commit 96bc2f2f contains an interim...
See comment for issue #9. The :close method actually destroys the underlying object. The userdata object becomes a NULL pointer. Lua does the same thing: file:close() calls fclose and then...
Yikes. That's alot of code for something simple. I'd rather hold off on this. I was thinking it was a small diff that reduced the overall SLoC. Regarding branching, I...
"Optionally bind core C functions with FFI" How do we do this? Can we import a C function through a lightuserdata, or do we need to define the function as...
Pasting this link so I can find it again. http://www.freelists.org/post/luajit/ffi-function-pointers-and-light-userdata,1
Clarification regarding distinction between direct and indirect calls. http://www.freelists.org/post/luajit/Cost-of-wrapping-ffi-functions,1 Probably will just use indirect calls, as otherwise we need to reload the module through ffi.load, but we could already be...
I've started an experimental branch for this, 60-luajit-ffi. I've got monotime working both as a boxed pointer and through a C namespace.
Do you think you can reduce the issue to a small example which reliably reproduces poor performance?
Will LuaJIT discard a trace if there's any branch that might call a Lua C API routine, or only if that branch is taken? :send/:recv could easily be made fully...