Work around a compiler bug - mark variables that are sent to functions that aren't yet
defined as FETCH_W (because they might end up being sent by reference)

* Centralized shutdown
* Change shutdown order again

Call the request_shutdown on modules before destroying symbol tables, so that
the session module can be implemented

- Fixed Karl's bug report. It's not really a thorough fix, we really need to rethink the INIT_FCALL/DO_FCALL issue.
- Fixed numerous AiCount problems

New $GLOBALS init

Fixes

* Support getThis() for internal functions.
* Fix 'new object or die' and AiCount issue thoroughly (earlier fix didn't
work with the optimizer).
* Add new macros for standardized defini

* Support getThis() for internal functions.
* Fix 'new object or die' and AiCount issue thoroughly (earlier fix didn't
work with the optimizer).
* Add new macros for standardized definition of classes.
* Only report AiCount problems if shutdown was not silent.

show more ...

* Add struct name to all typedef's so that they can be debugged with MSVC
* Fix an AiCount bug - list(...) = $var was using $var multiple times, and thus
causing AiCount to be decreased multi

* Add struct name to all typedef's so that they can be debugged with MSVC
* Fix an AiCount bug - list(...) = $var was using $var multiple times, and thus
causing AiCount to be decreased multiple times even though it was increased only
once for $var. Mark all FETCH_DIM's so that they won't decrease AiCount, and only
decrease AiCount on the last FETCH_DIM.
* Fix a stupid bug - forgot to pass CLS_C to some compiler function. For some reason
MSVC doesn't report these :I

show more ...

Avoid leaking fd's in case of failures

- Updates we did today

- Fix a small problem with class decelerations.

* Fix all hash checks that checked Bucket.arKey for NULL, when it was changed
to char[1], these checks should have been changed to Bucket.nKeyLength==0
* Support runtime declaration of func

* Fix all hash checks that checked Bucket.arKey for NULL, when it was changed
to char[1], these checks should have been changed to Bucket.nKeyLength==0
* Support runtime declaration of functions. I ended up changing the grammar
to catch top level functions vs. nested functions. The reason is simple -
if we don't have functions properly declared at compile-time, function calls
cannot be resolved at compile time, and have to be resolved at runtime, which
ends up being much much slower (without the optimizer, that is).
It's no biggy though, the grammar change isn't that bad.

show more ...

* Fix a bug that occured in case of parse errors. We need to restore the lexical state
even if the compilation failed.

* Fix a problem with constant quoted strings, that was causing Thies's problem
* Remove a development-time printf

Another small fix

Cleanups, remove old ts code

Make token names uniform, they all begin with T_ now.

Thread safety patch. It works now with 'just in time' resource initialization!

Thread safety patch. We're still not quite there but it compiles again, and
more logic has been implemented.

Support =unset as arguments

Whatnot:
* updated alloc_persist to use critical sections
* changed extension shutdown to two-phase
* updated dependencies
* PR support (don't remember if there was any really)

- Should fix the pass by reference problem. This happened because we moved
start from arg 1 now and not arg 0. There might be more places which need fixing
like in the executor but the bug

- Should fix the pass by reference problem. This happened because we moved
start from arg 1 now and not arg 0. There might be more places which need fixing
like in the executor but the bug seems OK now.

show more ...

-Tiny patches

- Fix various memory leaks.

Destroy the resource list after destroying the symbol table, otherwise the
auto-destructor for resources are run when the resource list is no longer valid