redo memory-usage mechanism, add way of dynamically initializing Lisp objects -------------------- ChangeLog entries follow: -------------------- lisp/ChangeLog addition: 2010-03-18 Ben Wing <ben@xemacs.org> * diagnose.el (show-memory-usage): Rewrite to take into account API changes in memory-usage functions. src/ChangeLog addition: 2010-03-18 Ben Wing <ben@xemacs.org> * alloc.c: * alloc.c (disksave_object_finalization_1): * alloc.c (lisp_object_storage_size): * alloc.c (listu): * alloc.c (listn): * alloc.c (Fobject_memory_usage_stats): * alloc.c (compute_memusage_stats_length): * alloc.c (Fobject_memory_usage): * alloc.c (Ftotal_object_memory_usage): * alloc.c (malloced_storage_size): * alloc.c (common_init_alloc_early): * alloc.c (reinit_alloc_objects_early): * alloc.c (reinit_alloc_early): * alloc.c (init_alloc_once_early): * alloc.c (syms_of_alloc): * alloc.c (reinit_vars_of_alloc): * buffer.c: * buffer.c (struct buffer_stats): * buffer.c (compute_buffer_text_usage): * buffer.c (compute_buffer_usage): * buffer.c (buffer_memory_usage): * buffer.c (buffer_objects_create): * buffer.c (syms_of_buffer): * buffer.c (vars_of_buffer): * console-impl.h (struct console_methods): * dynarr.c (Dynarr_memory_usage): * emacs.c (main_1): * events.c (clear_event_resource): * extents.c: * extents.c (compute_buffer_extent_usage): * extents.c (extent_objects_create): * extents.h: * faces.c: * faces.c (compute_face_cachel_usage): * faces.c (face_objects_create): * faces.h: * general-slots.h: * glyphs.c: * glyphs.c (compute_glyph_cachel_usage): * glyphs.c (glyph_objects_create): * glyphs.h: * lisp.h: * lisp.h (struct usage_stats): * lrecord.h: * lrecord.h (enum lrecord_type): * lrecord.h (struct lrecord_implementation): * lrecord.h (MC_ALLOC_CALL_FINALIZER_FOR_DISKSAVE): * lrecord.h (DEFINE_DUMPABLE_LISP_OBJECT): * lrecord.h (DEFINE_DUMPABLE_SIZABLE_LISP_OBJECT): * lrecord.h (DEFINE_DUMPABLE_FROB_BLOCK_LISP_OBJECT): * lrecord.h (DEFINE_DUMPABLE_FROB_BLOCK_SIZABLE_LISP_OBJECT): * lrecord.h (DEFINE_DUMPABLE_INTERNAL_LISP_OBJECT): * lrecord.h (DEFINE_DUMPABLE_SIZABLE_INTERNAL_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_SIZABLE_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_FROB_BLOCK_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_FROB_BLOCK_SIZABLE_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_INTERNAL_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_SIZABLE_INTERNAL_LISP_OBJECT): * lrecord.h (MAKE_LISP_OBJECT): * lrecord.h (DEFINE_DUMPABLE_MODULE_LISP_OBJECT): * lrecord.h (DEFINE_DUMPABLE_MODULE_SIZABLE_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_MODULE_LISP_OBJECT): * lrecord.h (DEFINE_NODUMP_MODULE_SIZABLE_LISP_OBJECT): * lrecord.h (MAKE_MODULE_LISP_OBJECT): * lrecord.h (INIT_LISP_OBJECT): * lrecord.h (INIT_MODULE_LISP_OBJECT): * lrecord.h (UNDEF_LISP_OBJECT): * lrecord.h (UNDEF_MODULE_LISP_OBJECT): * lrecord.h (DECLARE_LISP_OBJECT): * lrecord.h (DECLARE_MODULE_API_LISP_OBJECT): * lrecord.h (DECLARE_MODULE_LISP_OBJECT): * lstream.c: * lstream.c (syms_of_lstream): * lstream.c (vars_of_lstream): * marker.c: * marker.c (compute_buffer_marker_usage): * mc-alloc.c (mc_alloced_storage_size): * mc-alloc.h: * mule-charset.c: * mule-charset.c (struct charset_stats): * mule-charset.c (compute_charset_usage): * mule-charset.c (charset_memory_usage): * mule-charset.c (mule_charset_objects_create): * mule-charset.c (syms_of_mule_charset): * mule-charset.c (vars_of_mule_charset): * redisplay.c: * redisplay.c (compute_rune_dynarr_usage): * redisplay.c (compute_display_block_dynarr_usage): * redisplay.c (compute_glyph_block_dynarr_usage): * redisplay.c (compute_display_line_dynarr_usage): * redisplay.c (compute_line_start_cache_dynarr_usage): * redisplay.h: * scrollbar-gtk.c (gtk_compute_scrollbar_instance_usage): * scrollbar-msw.c (mswindows_compute_scrollbar_instance_usage): * scrollbar-x.c (x_compute_scrollbar_instance_usage): * scrollbar.c (compute_scrollbar_instance_usage): * scrollbar.h: * symbols.c: * symbols.c (reinit_symbol_objects_early): * symbols.c (init_symbols_once_early): * symbols.c (reinit_symbols_early): * symbols.c (defsymbol_massage_name_1): * symsinit.h: * ui-gtk.c: * ui-gtk.c (emacs_gtk_object_getprop): * ui-gtk.c (emacs_gtk_object_putprop): * ui-gtk.c (ui_gtk_objects_create): * unicode.c (compute_from_unicode_table_size_1): * unicode.c (compute_to_unicode_table_size_1): * unicode.c (compute_from_unicode_table_size): * unicode.c (compute_to_unicode_table_size): * window.c: * window.c (struct window_stats): * window.c (compute_window_mirror_usage): * window.c (compute_window_usage): * window.c (window_memory_usage): * window.c (window_objects_create): * window.c (syms_of_window): * window.c (vars_of_window): * window.h: Redo memory-usage mechanism, make it general; add way of dynamically initializing Lisp object types -- OBJECT_HAS_METHOD(), similar to CONSOLE_HAS_METHOD(). (1) Create OBJECT_HAS_METHOD(), OBJECT_HAS_PROPERTY() etc. for specifying that a Lisp object type has a particular method or property. Call such methods with OBJECT_METH, MAYBE_OBJECT_METH, OBJECT_METH_OR_GIVEN; retrieve properties with OBJECT_PROPERTY. Methods that formerly required a DEFINE_*GENERAL_LISP_OBJECT() to specify them (getprop, putprop, remprop, plist, disksave) now instead use the dynamic-method mechanism. The main benefit of this is that new methods or properties can be added without requiring that the declaration statements of all existing methods be modified. We have to make the `struct lrecord_implementation' non-const, but I don't think this should have any effect on speed -- the only possible method that's really speed-critical is the mark method, and we already extract those out into a separate (non-const) array for increased cache locality. Object methods need to be reinitialized after pdump, so we put them in separate functions such as face_objects_create(), extent_objects_create() and call them appropriately from emacs.c The only current object property (`memusage_stats_list') that objects can specify is a Lisp object and gets staticpro()ed so it only needs to be set during dump time, but because it references symbols that might not exist in a syms_of_() function, we initialize it in vars_of_(). There is also an object property (`num_extra_memusage_stats') that is automatically initialized based on `memusage_stats_list'; we do that in reinit_vars_of_alloc(), which is called after all vars_of_() functions are called. `disksaver' method was renamed `disksave' to correspond with the name normally given to the function (e.g. disksave_lstream()). (2) Generalize the memory-usage mechanism in `buffer-memory-usage', `window-memory-usage', `charset-memory-usage' into an object-type- specific mechanism called by a single function `object-memory-usage'. (Former function `object-memory-usage' renamed to `total-object-memory-usage'). Generalize the mechanism of different "slices" so that we can have different "classes" of memory described and different "slices" onto each class; `t' separates classes, `nil' separates slices. Currently we have three classes defined: the memory of an object itself, non-Lisp-object memory associated with the object (e.g. arrays or dynarrs stored as fields in the object), and Lisp-object memory associated with the object (other internal Lisp objects stored in the object). This isn't completely finished yet and we might need to further separate the "other internal Lisp objects" class into two classes. The memory-usage mechanism uses a `struct usage_stats' (renamed from `struct overhead_stats') to describe a malloc-view onto a set of allocated memory (listing how much was requested and various types of overhead) and a more general `struct generic_usage_stats' (with a `struct usage_stats' in it) to hold all statistics about object memory. `struct generic_usage_stats' contains an array of 32 Bytecounts, which are statistics of unspecified semantics. The intention is that individual types declare a corresponding struct (e.g. `struct window_stats') with the same structure but with specific fields in place of the array, corresponding to specific statistics. The number of such statistics is an object property computed from the list of tags (Lisp symbols describing the statistics) stored in `memusage_stats_list'. The idea here is to allow particular object types to customize the number and semantics of the statistics where completely avoiding consing. This doesn't matter so much yet, but the intention is to have the memory usage of all objects computed at the end of GC, at the same time as other statistics are currently computed. The values for all statistics for a single type would be added up to compute aggregate values for all objects of a specific type. To make this efficient, we can't allow any memory allocation at all. (3) Create some additional functions for creating lists that specify the elements directly as args rather than indirectly through an array: listn() (number of args given), listu() (list terminated by Qunbound). (4) Delete a bit of remaining unused C window_config stuff, also unused lrecord_type_popup_data.