Re: Explicitly triggering the Garbage Collector

[Rici Lake <lua@...>]

From the reference manual (http://www.lua.org/ftp/refman-5.0.pdf), page 
27 (33 if your pdf reader doesn't understand page numbers):

Lua uses two numbers to control its garbage collection: the count and 
the threshold (see §2.9). The first counts the amount of memory in use 
by Lua; when the count reaches the threshold, Lua runs its garbage 
collector. After the collection, the count is updated and the threshold 
is set to twice the count value.

You can access the current values of these two numbers through the 
following functions:
   int lua_getgccount (lua_State *L);
   int lua_getgcthreshold (lua_State *L);

Both return their respective values in Kbytes.

You can change the threshold value with void lua_setgcthreshold 
(lua_State *L, int newthreshold); Again, the newthreshold value is 
given in Kbytes. When you call this function, Lua sets the new 
threshold and checks it against the byte counter. If the new threshold 
is less than the byte counter, then Lua immediately runs the garbage 
collector. In particular lua_setgcthreshold(L,0) forces a garbage 
collection. After the collection, a new threshold is set according to 
the previous rule.

Note: I believe this interface will change with Lua 5.1.

On 7-Aug-05, at 11:20 AM, Uli Kusterer wrote:

> Hi, it's me again.
>
>  I've found luaC_collectgarbage() in lgc.h for explicitly triggering 
> the garbage collector. However, if I understand this correctly, we're 
> only supposed to use the "public" Lua API, i.e. the stuff in lua.h and 
> lualib.h (and lauxlib.h). Is it okay to use this GC call, or am I 
> asking for trouble and making it harder to update to future versions 
> of Lua?
>
>  Since my app has several distinct phases, and thus certain clear 
> breaks at which I *know* there will be lots of objects in need of 
> being reclaimed, I'd like to be able to explicitly trigger the GC. 
> People will be waiting between these phases anyway, and thus won't 
> notice the time the GC could take. This function seems to work, but 
> since it's not part of the public API, I fear I may have overlooked 
> some more obvious means of achieving the same. Any clues will be 
> appreciated.
>
> Cheers,
> -- M. Uli Kusterer
> http://www.zathras.de