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.

@node Startup Paths, Packages, Command Switches, Top
@comment  node-name,  next,  previous,  up
@section How XEmacs finds Directories and Files

@cindex startup paths
@cindex directories

XEmacs deals with a multitude of files during operation.  These files
are spread over many directories, and XEmacs determines the location of
most of these directories at startup and organizes them into various
paths.  (A @dfn{path},
@cindex path
for the purposes of this section, is simply a list of directories which
XEmacs searches successively in order to locate a file.)

@subsection XEmacs Directory Hierarchies
@cindex hierarchies
@cindex directory hierarchies

Many of the files XEmacs looks for are located within the XEmacs
installation itself.  However, there are several views of what actually
constitutes the "XEmacs installation": XEmacs may be run from the
compilation directory, it may be installed into arbitrary directories,
spread over several directories unrelated to each other.  Moreover, it
may subsequently be moved to a different place.  (This last case is not
as uncommon as it sounds.  Binary kits work this way.)  Consequently,
XEmacs has quite complex procedures in place to find directories, no
matter where they may be hidden.

XEmacs will always respect directory options passed to @code{configure}.
However, if it cannot locate a directory at the configured place, it
will initiate a search for the directory in any of a number of
@dfn{hierarchies} rooted under a directory which XEmacs assumes contain
parts of the XEmacs installation; it may locate several such hierarchies
and search across them.  (Typically, there are just one or two
hierarchies: the hierarchy where XEmacs was or will be installed, and
the one where it is being built.)  Such a directory containing a
hierarchy is called a @dfn{root}.
@cindex root of a hierarchy
Whenever this section refers to a directory using the shorthand
@code{<root>}, it means that XEmacs searches for it under all
hierarchies XEmacs was able to scrounge up.  In a
running XEmacs, the hierarchy roots are stored in the variable
@code{emacs-roots}.
@vindex emacs-roots

@subsection Package Hierarchies
@cindex package hierarchies

Many relevant directories and files XEmacs uses are actually not part of
the core installation.  They are part of any of the many packages
usually installed on top of an XEmacs installation.  (@xref{Packages}.)
Hence, they play a prominent role in the various paths XEmacs sets up.

XEmacs locates packages in any of a number of package hierarchies.
Package hierarchies fall into three groups: @dfn{early}, @dfn{late},
and @dfn{last},
@cindex early package hierarchies
@cindex late package hierarchies
@cindex last package hierarchies
according to the relative location at which they show
up in the various XEmacs paths.  Early package hierarchies are at the
very front, late ones somewhere in the middle, and last hierarchies are
(you guessed it) last.

By default, XEmacs expects an early package hierarchy in the
subdirectory @file{.xemacs/xemacs-packages} of the user's home
directory.

Moreover, XEmacs expects late hierarchies in the subdirectories
@file{site-packages}, @file{mule-packages}, and @file{xemacs-packages}
(in that order) of the @file{<root>/lib/xemacs} subdirectory of one of
the installation hierarchies.  (If you run in-place, these are direct
subdirectories of the build directory.)  Furthermore, XEmacs will also
search these subdirectories in the @file{<root>/lib/xemacs-<VERSION>}
subdirectory and prefer directories found there.

By default, XEmacs does not have a pre-configured last package
hierarchy.  Last hierarchies are primarily for using package hierarchies
of outdated versions of XEmacs as a fallback option.  For example, it is
possible to run XEmacs 21 with the 20.4 package hierarchy as a last
hierarchy.

It is possible to specify at configure-time the location of the various
package directories with the @code{--with-user-packages} (an alias for
@samp{--with-early-packages}), @code{--with-system-packages} (an alias
for @samp{--with-late-packages}), and @code{--with-legacy-packages} (an
alias for @samp{--with-last-packages}) options to configure.
@cindex package path
At run time, the package directories may also be specified via the
@code{EMACSEARLYPACKAGES}, @code{EMACSLATEPACKAGES}, and
@code{EMACSLASTPACKAGES} environment variables.

An XEmacs package hierarchy is laid out just like a normal installed
XEmacs directory.  It may have @file{lisp}, @file{etc}, @file{info}, and
@file{lib-src} subdirectories.  (The @file{lib-src} subdirectory
contains architecture-independent general-purpose scripts interpreted by
the shell or Perl.  Java is also being widely used, but Java programs
are generally found under @file{etc}, because they are specific to
particular packages such as @file{JDE} and @file{xslt}.)  XEmacs adds
these at appropriate places within the various system-wide paths.

There may be any number of package hierarchy directories.

@subsection Directories and Paths
@cindex paths

Here is a list of the various directories and paths XEmacs tries to
locate during startup.  XEmacs distinguishes between directories and
paths specific to @dfn{version}, @dfn{site}, and @dfn{architecture}
when looking for them.

@table @code
@item version-specific
@cindex version-specific directories
directories are specific to the version of XEmacs they belong to and
typically reside under @file{<root>/lib/xemacs-<VERSION>}.
@item site-specific
@cindex site-specific directories
directories are independent of the version of XEmacs they belong to and
typically reside under @file{<root>/lib/xemacs}
@item architecture-specific
@cindex architecture-specific directories
directories are specific both to the version of XEmacs and the
architecture it runs on and typically reside under
@file{<root>/lib/xemacs-<VERSION>/<ARCHITECTURE>}.
@end table

During installation, all of these directories may also reside directly
under @file{<root>}, because that is where they are in the XEmacs tarball.

If XEmacs runs with the @code{-debug-paths} option (@pxref{Command
Switches}), it will print the values of these variables, hopefully
aiding in debugging any problems which come up.

@table @code

@item lisp-directory
@vindex lisp-directory
Contains the version-specific location of the Lisp files that come with
the core distribution of XEmacs.  XEmacs will search it recursively to a
depth of 1 when setting up @code{load-path}.

@item load-path
@vindex load-path
Is where XEmacs searches for XEmacs Lisp files with commands like
@code{load-library}.
@findex load-library
It contains the package lisp directories (see further down) and the
version-specific core Lisp directories.  If the environment variable
@code{EMACSLOADPATH} is set at startup, its directories are prepended to
@code{load-path}.
@vindex EMACSLOADPATH

@item Info-directory-list
@vindex Info-directory-list
Contains the location of info files.  (See @ref{(info)}.)  It contains
the package info directories and the version-specific core
documentation.  Moreover, XEmacs will add @file{/usr/info},
@file{/usr/local/info} as well as the directories of the environment
variable @code{INFOPATH}
@vindex INFOPATH
to @code{Info-directory-list}.

@item exec-directory
@vindex exec-directory
Is the directory of architecture-dependent files that come with XEmacs,
especially executable programs intended for XEmacs to invoke.

@item exec-path
@vindex exec-path
Is the path for executables which XEmacs may want to start.  It contains
the package executable paths as well as @code{exec-directory}, and the
directories of the environment variables @code{PATH}
@vindex PATH
and @code{EMACSPATH}.
@vindex EMACSPATH

@item doc-directory
@vindex doc-directory
Is the directory containing the architecture-specific @file{DOC} file
that contains documentation for XEmacs' commands.

@item data-directory
@vindex data-directory
Is the version-specific directory that contains core data files XEmacs uses.
It may be initialized from the @code{EMACSDATA}
@vindex EMACSDATA
environment variable.

@item data-directory-list
@vindex data-directory-list
Is the path where XEmacs looks for data files.  It contains package data
directories as well as @code{data-directory}.

@end table