Robert Lemmen wrote:
> ok, i got the thing about the variables, although i think i want to avoid
> effectively having two variables with the same name on one object ...
>
> now to these frobs. i understand how to use them and how to restrict a method
> to be frob-only, but what i don't get is why to make the effort. what is the
> advantage of a frobbed call?
I'll let Brandon or someone else answer this.
> btw, how is the status of this project, because it is very cool, but also seems
> to be a bit dead/sleeping
There's been a lot of work done on Genesis of late, none of which has
been released yet because I haven't had the time to clean everything up
nicely and finish up some of the changes. It is pretty stable though
since we've been running TEC (http://www.eternal-city.com/) on it for a
couple of months.
Partial changelog:
* cleanup config() and add a new 'cachelog parameter (Brad)
* add logging of object name and size at all db_put()'s (Brad)
* add pre-allocation to cBuf's (Brad)
* add sync() builtin to allow cache_sync without backup (Brad)
* fix pad to disallow 0 length pad string (Brad)
* method cache fixes: don't flush cache in as many
cases, and also cache negative method lookups (Bruce)
* added a test for the above pad() bug, as well as
some other missing tests. (Bruce)
* added bufsub() and $buffer.bufsub() (Bruce)
* Change task_stack() to let it obtain the stack for any
given frame and to generate either line numbers or
opcode index values. (Bruce)
* Add log_task_stack() and log_all_task_stacks() callable
from C. (Bruce)
* Re-enable AC_INLINE check in configure (Bruce)
* cleanup cache.c to have a dirty list and reduce overhead of
cache holder array (Brad)
* Add linefeed to panic() message. (Bruce)
* panic() now tries to call log_all_task_stacks(). (Bruce)
* Integrate patch to support using doubles for the ColdC
float type. (Brian, trivial modifications by Bruce and Brad)
* Adjust size_* functions to all return Int. (Bruce)
* html_escape needs to encode if there are doublequotes,
otherwise Bruce's added case won't be used. (Allen Noe)
* Corrected random(x)'s throw message for x < 1 (Allen Noe)
* Make method_bytecode() act like list_method(), specifically
not returning bytecode for methods on parent objects (Allen Noe)
* Also fix a bug in method_bytecode where it checks the wrong
variable. (Allen Noe)
* Win32 build fix involving VRELEASE. (Allen Noe)
* Add func_init_0_to_2(). (Bruce)
* Make stack() accept parameters to specify task id and
whether or not line numbers are desired. (Bruce)
* Add 'log_malloc_size parameter for config(). (Bruce)
* Add explode_quoted() patch. (Jeremy Weatherford)
* Add calling_method() patch. (Allen Noe)
* Make method_bytecode() output info on jumps. (Allen Noe)
* Makefile fix, putting textdb.c into EVERYTHING_O (Jeff Dubrule)
* Change binarydb/dbpack routines to read/write entire object at
once going through a cBuf. (Brad)
* Fix db_alloc to call grow_bitmap once when expanding to fit an
object that's > the size of the standard increment (Brad)
* Add ancestor caching to speed up has_ancestor() and coldcc.
(Bruce, Brad)
* Added a cache_check cache in handle_varcmd (Brad)
* Minor code cleanups in binarydb.c (Brad)
* Add percent done to coldcc compile and decompile (Brad)
* valid() works on frobs (normal and handled-frobs). Tests
added to the test suite. (Bruce)
* Error handling optimization to shift expense from throw()
to traceback(). (Bruce)
* Add background cleaner thread to write dirty objects to disk (Brad)
* Further method cache cleanups, including a config()
option to control logging of information about
invalidations. (Bruce)
* If new and old parents lists are equal, change_parents()
should do no work. (Including cache invalidations) (Bruce)
* Add cache_stats() and related data gathering code. (Bruce)
* Add support for Obj*'s as a subtype of cData's. (Brad)
# Add config('cleanerwait, ##) to control the number of seconds
between cleaner thread runs (Brad)
* Add config('cleanerignore, #[]) to give the cleaner thread objects
to ignore, the key is the only relevant part.
* Add config('cachewatch, $obj) and config('cachewatchcount, ##)
to log the call stack of a specific $obj every cachewatchcount
dirtys. (Brad)
* Add ability to get both size and char* data from an ident to avoid
useless strlen() calls. (Bruce)
* Avoid writing back size/offset values if they haven't changed in
binarydb.c. (Bruce)
* Lookup and bind native methods while compiling the DB so that they
can be used within eval blocks in coldcc to let you test natives
in the test suite. (Bruce)
* data_cmp() compares objnums by their numerical value now. (Bruce)
* New lookup implementation that uses Berkeley DB. (Bruce)
* Add has_method() to ColdC to make up for catch ~methodnf and
find_method() being so slow. has_method() does and -must-
search ancestry. That's why it exists. (Bruce)
* Switch from DBM to BDB for the index lookup for now.
The cache cleaning thread will be going away soon in favor of some code
that can do asynchronous cache flushes, and then we're well along the
way to having the DB on disk be consistent at all times, so if there is
a crash, you likely won't have to revert back to the last backup.
That's the plan anyway. :)
There is also more debug/monitoring stuff coming as and when I get the
time. That stuff is intended to help tune the server and find
performance problems in people's ColdC code.
But, for most people, what's available now as Genesis 1.1.9 is pretty
decent and usable. Most of the above changes have been motivated by
work in making TEC run faster.
- Bruce
|