#531 mem handling reform

open
8
2008-08-06
2008-07-31
No

The newly available allocCache field in Interp can be used to speed up most alloc/free as well as and Tcl_Obj allocation and freeing.

Specifically, a simple extra interface to the functions in generic/tclThreadAlloc that receives a Cache pointer would bypass accessing the TSD. For Tcl_Obj allocation and freeing, the work can be done mostly in macros (see the macro TclAllocObjStorageEx() in HEAD).

What this means is that we should pass the interp where available to the alloc and obj creation/deletion functions or macros (NULL if unknown). Most of the core activity happens with an interp at hand.

This is a fairly trivial project, but rather bothersome (2000+ locs to change). Note that the "upgrade" can be done gradually, starting with the most critical parts. Note also that this would make normal allocations compete with TclStackAlloc in terms of speed, so that we could remove it and be free of its bothersome limitations.

Discussion

  • miguel sofer

    miguel sofer - 2008-08-06

    Logged In: YES
    user_id=148712
    Originator: YES

    This reform would also allow to keep memory usage statistics per interp, cheaply and independently of the allocator being used. It opens the path to implement [interp memlimit], which is one of the biggest remaining holes in the protection against DOS attacks on safe interps.

    My first idea would be to keep mem usage statistics on non-Tcl_Obj memory only.

     
  • miguel sofer

    miguel sofer - 2008-08-06
    • summary: Use the ekeko to speed up threaded allocs --> mem handling reform
     

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:

JavaScript is required for this form.





No, thanks