| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
* if config.on_state_create() is a C function, call it by direct C
closure reconstruction in newly created states
* bumped version to 3.7.6
|
|
|
| |
Actually, you shouldn't need malloc.h at all, it's obsolete; including stdlib.h should take care of what malloc.h used to handle.
|
|
|
|
|
| |
use demote_full_userdata to select between light userdata demotion or
raising an error when attempting to transfer a non-deep full userdata
|
|
|
|
|
|
|
|
|
| |
* bumped version to 3.7.3
* set pthread thread cancel type to PTHREAD_CANCEL_ASYNCHRONOUS
* lane_h:cancel() accepts a 3rd timeout argument used when waiting for
actual thread termination (hitting the timeout raises an error)
* added PROPAGATE_ALLOCF macro to select state creation mode
(lua_newstate or luaL_newstate)
|
|
|
|
|
| |
library
Fix issue #77.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
* fix lanes.threads() not being available in a lane where
lanes.configure() settings didn't contain track_lanes although the
initial configure() call did.
* require "lanes".configure() sequence is only necessary at the first
require "lanes".
* fix a crash at application shutdown where in some situations we could
deinitialize the protected allocator mutex while a lane was still using
it.
* fix timers broken by change 69
|
|
|
|
| |
list is "*"
|
| |
|
|
|
|
|
|
|
|
| |
no longer call core.configure with dummy params when requiring lanes
more than once (fixes potential multithreading issues with LuaJIT
allocator)
activated EnableCrashingOnCrashes() Win32 debug builds
fixed some comments in code
|
|
|
|
|
|
|
|
|
|
|
|
| |
Lanes, and send over native functions a bit faster as well
* Lanes no longer has to internally require modules inside the keeper
states because they no longer need a lookup database. the lookup name is
stored as-is and actually converted in the destination state
*
optimisation: bypass cache when sending native functions over
* removed
all the KEEPER_MODEL_LUA code, as it can no longer work anyway
|
|
|
|
|
|
|
|
| |
* lane:cancel(<negative-timeout>) only causes cancel_test() to return
true but won't interrupt execution of the lane during linda operations
* more explicit errors when trying to transfer unknown source functions
(with new configure option verbose_errors)
* default options wrap allocator around a mutex when run by LuaJIT
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
| |
* new lanes.h header and API call luaopen_lanes_embedded() for embedders
* "lanes.core" is an acceptable library in the generator libs argument
* library "*" wildcard also opens lanes.core
* tweaked code for Xbox 360 build
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* new: API lanes.require(), use it instead of regular require() for modules that export C functions you need to send over.
* new: lanes no longer require 'lanes.core' by default in every created state. Use {required={"lanes.core"}} if you need to transfer lanes functions.
* internal: because of the above, reworked the timer implementation to remove upvalue-dependency on lanes.core
* new: API lanes.timer_lane, to be able to operate on timer lane if need be
* improved: if a module is a full userdata, scan its metatable for function database population
* improved: on_state_create can be a Lua function
* changed: on_state_create is called after the base libraries are loaded
* package[loaders|searchers] is no longer transfered as function naming depends on slot order
* internal: changed separator from '.' to '/' in lookup databases to be able to distinguish search levels and dot coming from module names
* added some mode debug spew
* updated tests to reflect the above changes
|
|
|
|
| |
* A bit of code cosmetics
|
|
|
|
|
|
|
|
| |
* bugfix: take into account the fact that "coroutine" is no longer part of base library in Lua 5.2
* bugfix: if "bit32" was listed in the libraries, it wouldn't open (library list parsing failing on digits)
* bugfix: Use luaL_requiref() to open standard libraries in Lua 5.2 as we should
* bugfix: any Lua state created by Lanes reuses the allocator function of the originating state
* bugfix: don't call on_state_create() while GC is suspended during lua state initialization
|
|
|
|
|
|
|
|
|
|
|
| |
* raise an error if lane generator libs specification contains a lib more than once
* bit32 is a valid lib name in the libs specification (silently ignored by the Lua 5.1 build)
* improved lanes.nameof to search inside table- and userdata- metatables for an object's name
* fixed an unwarranted error when trying to discover a function name upon a failed transfer
* contents of package.[path,cpath,preload,loaders|searchers] are pulled *only once* inside keeper states at initialisation
* Lua function upvalues equal to the global environment aren't copied by value, but bound to the destination's global environment
especially useful for Lua 5.2 _ENV
* fixed loading of base libraries that didn't create the global tables when built for Lua 5.2
|
|
|
| |
Removed a limitation preventing Lua functions with indirect recursive upvalue references from being transferable.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Lua 5.2 introduced a hash randomizer seed which causes table iteration to yield a different key order
on different VMs even when the tables are populated the exact same way.
When Lua is built with compatibility options (such as LUA_COMPAT_ALL),
this causes several base libraries to register functions under multiple names.
This, with the randomizer, can cause the first name of a function to be different on different VMs,
which breaks function transfer.
This means that Lua 5.2 must be built with compatibility off to be able to use Lanes.
Even under Lua 5.1, this may cause trouble (even if this would be much less frequent)
Unfortunately, this fails with string.gfind/string.gmatch when Lua 5.1 is built with LUA_COMPAT_GFIND (which is the case of LuaBinaries),
so for the time being, fail only for Lua 5.2 as the randomizer is the real show breaker here.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
the user find out what's wrong
* new function lanes.nameof()
|
|
|
|
| |
luaL_openlibs()
|
|
|
|
| |
int size)
|
|
|
|
|
|
| |
parameters
* added support for an on_state_create callback called to load custom functions in a state in addition to the base libraries
|
| |
|
|
|
|
| |
* fix a multithreading issue at desinit causing invalid memory accesses (to be verified)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
when atexit() handlers are processed
* Lua 5.2-style module:
* module() is no longer used to implement lanes.lua
* a global "lanes" variable is no longer created when the module is required
* the Lanes module table is returned instead
* Lanes must be initialized before used:
* the first occurence of 'require "lanes"' produces a minimal interface that only contains a configure() function
* the remainder of the interface is made available once this function is called
* subsequent calls to configure() do nothing
* configure() controls the number of keeper states and the startup of timers
* LuaJIT 2 compatibility
* non-Lua functions are no longer copied by creating a C closure from a C pointer, but through 2-way lookup tables
* this means that if a lane function body pulls non-Lua functions, the lane generator description must contain the list of libraries and modules that exports them
* introduces a change in configuration .globals management: contents are copied *after* std libs are loaded
* new .required configuration entry to list modules that must be require()'ed before lane body is transferred
* lane:cancel() wakes up waiting lindas like what is done at lane shutdown
|
| |
|
|
|
|
|
|
| |
LuaJIT2-compatibility (but this is not enough).
* moved keeper-related code in a separate source file
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Refactor lane proxy implementation: it is now a full userdata instead
of a table, and its methods are implemented in C instead of Lua.
* its metatable is no longer accessible.
* writing to the proxy raises an error.
* it is no longer possible to overwrite its join() and cancel() methods
- when a deep userdata idfunc requests a module to be required, manually
check that it is not loaded before requiring it instead of relying on
the require function's loop detection feature.
- when a module must be required, raise an error if the 'require' function
is not found in the target state.
- we know Lanes is loaded in the master state, so we don't force it
to be required in every lane too when a linda deep userdata is copied.
|
|
|
|
| |
errors thrown by inter-state data copy for unsupported types
|
|
|
|
| |
lua-callable and to be able to require the module it was exported from in the target lanes.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
github repository becomes the official Lanes source codebase.
Note that Asko's SVN server holds version 2.0.9, whereas this is version 2.0.10, but I don't see any real need to update SVN if it is to become deprecated.
Next steps:
- upgrade the rockspec to the latest version
- make the html help available online somewhere
Signed-off-by: Benoit Germain <bnt.germain@gmail.com>
|
|
|