Sorry for going back and forth with constness, but in some cases it was impossible to ensure that input params aren't overwritten, thus I took const away to avoid making promises that can't be met (see the case of fetch that uses a supposedly input-only param to return data). Overwriting memory that was declared const initially is a serious error in C++ with undefined outcome so we better do not lie.
1. Avoid ALLR memory allocation routine (use new / delete).
2. Use common (TypedHandle) class to control consistency of handles.
3. Make ctors and dtors work - in most cases more work is needed to make them meaningful.
2. Introduce new private API calls, fb__shutdown() and fb__shutdown_callback().
3. Introduce fb_smp_server - binary to run superclassic on posix.
4. Various cleanups.
and CORE-1079: Every attach of fbclient/fbembed library leaks 64KB of memory
To do it:
- created new template GlobalPtr, controlling access to destructors of global objects.
- applied it to global objects.
Also some related cleanup.
const correctness
BOOLEAN -> bool
Vars in scope
Formatting
JRD_REQ -> jrd_req*
JRD_TRA -> jrd_tra*
JRD_REL -> jrd_rel*
JRD_FLD -> jrd_fld*
event -> event_t, EVENT_T dropped, EVENT should go in the future
Frank's corrections
alice's STR -> alice_str*
remote's STR -> rem_str*
etc.
It creates entities and backups and restores for me. :-)
-Replace codes.h with iberror.h (or remove)
-Move gds__vax_integer and gds__event_block* to alt.cpp and isc_ to gds.cpp
-Use ISC_QUAD internally
-Use isc_vax_integer internally
placed some variables in context
fixed some function signatures
closed a few possible buffer overruns
sorry to the platform maintainers, I can't verify what I did for non-Win32 builds
With previous logic all spare threads were notified and performed
expensive loop on each request arrival. New logic wakes up only
one thread per request. (the same fix was performed in 2.0 SS lock manager
to avoid waking up all blocked threads on each lock request resolution).
As an (unexpected) result of this changes SS performance significantly
improved. Some artificial tests show timing improvements by
factor between 2 and 3.
