Shutdown doesn't look clean on Linux, and this might help highlight
current and future problems. Besides, we don't want shutdown to take
ages, do we? ;)
This was SVN commit r2801.
also fix refcnt for tex_wrap temporarily until mem_assign/mem_get_ptr
do.
h_mgr: add disallow-reload support (required for wrapped ogl_tex since
they are not backed by a file)
This was SVN commit r2780.
fix UNREACHABLE def on unix;
string_s: use HAVE_STRING_S;
wdbg_sym: fix option set code; explain why search path isn't needed
WriteScreenshot: make sure there's enough mem; use Handle to free mem
(more efficient+less error prone)
This was SVN commit r2779.
- config: add proper HAVE_ macro for string_s functions rather than
checking MSC ver directly
- string_s: disable parts of the self-test that trigger errors if not
using the string_s.cpp implementation
This was SVN commit r2760.
to that end, provide support for "wrapping" existing textures so that
tex_* and ogl_tex* calls may be used.
also have renderer use ogl_tex for its created-on-the-fly composite
alpha texture
also named atlas's thread (for fun+profit during debugging)
This was SVN commit r2754.
executive summary: memcpy2 aka. ia32_memcpy is a drop-in replacement
that is much faster than memcpy (7% on small buffers, 300% on 192K
buffers)
This was SVN commit r2729.
considerable continued cleanup in ia32
came across article detailing HT detect on multicore systems. that
requires new code, working on it.
pthread: while reading, came across idea of wrapping thread trampoline
in __except -> more reliable exception handling
This was SVN commit r2716.
premake stuff: call assembler objects *.asm.obj instead of *.obj (so
that ia32.asm doesn't collide with ia32.cpp)
ia32.asm: use .text section for proper linkage
This was SVN commit r2709.
lib: add bits() function (actually moved it from zip.cpp)
ia32: start moving code out to external asm file (will use NASM in
pre-build step; avoids needing MSC and GCC-specific inline code). move
memcpy test code out to other file
This was SVN commit r2706.
add tex error codes
move memcpy code to ia32
tex: simplify codec call with separate tex_codec_for* calls
codecs: use error code instead of string; remove redundant params
ia32: heavy WIP - testing memcpy implementations
util: prepare for zero-copy improvement to screenshot code
This was SVN commit r2695.
dyn_array supports "wrapping" other types of memory (stopgap measure)
texture codecs now load via DynArray (intermediate step to all mem stuff
going through res/mem.cpp)
simplified tex codecs by doing as many checks as possible in
tex_load_mem ("template method"-esque)
This was SVN commit r2690.
file: expose block size constant; always round user transfers up
tex: transform() now receives what to change, not what to achieve.
tex*: encoding works with DynArray+VFS backend
GameSetup: free g_BuildingPlacer to avoid mem leak.
fixed some warnings+bugs.
This was SVN commit r2679.
self_test: added another note
wtime: fix evil bug - was casting FILETIME* to u64*, which has different
alignment guarantees.
This was SVN commit r2662.
self_test: add dox (also to wiki)
tex_codecs: further work; renamed+reorg for clarity. separating codecs
from IO implementation
This was SVN commit r2654.
- mem.h: export mem_assign (needed to wrap a mem buffer)
- use stricmp in some file extension comparisons
- wposix: properly sets last-error code when opendir fails
- also: misc comments
This was SVN commit r2650.
tex.cpp now has loader API and codec support routines.
provide for codecs "transforming" the image; this allows e.g.
decompressing DDS, which is currently done in ogl code (but belongs
here)
This was SVN commit r2648.
- ogl_tex_bind now supports multitexturing (avoids need to get id and
then bind manually)
- provide nicer API for querying texture properties (no longer one big
getter). improved GUI determination of texture alpha.
- renderer: another see-if-texture-has-alpha and fixed alpha map code
(wasn't checking if all are same size. only worked because the first map
was larger than the rest
This was SVN commit r2647.
squelch debug asserts during self-tests to avoid warning about intended
errors (passing invalid data into functions to test their error
reporting)
This was SVN commit r2646.
- input.cpp allows feeding in events from the outside (will be used by
main.cpp and Atlas msg loop)
- most of main.cpp split into ps/GameSetup.*
- main.cpp is now a "controller" that drives the game via SDL events
(Atlas is another such controller that can override main.cpp). it calls
GameSetup functions to set up and processes messages.
- add functions to import AtlasUI.dll automatically
also, the usual refactoring ;)
This was SVN commit r2622.
main: add logic for run-time load of Atlas (ATLAS_* section). also moved
GUI code into a GUI_* section, removed console "Testing function
registration", moved input init into a separate function and cleaned up
a bit
This was SVN commit r2619.
- wposix.cpp: initial support for MEM_RESERVE and MEM_COMMIT semantics
in mmap
- cstr: removed no longer necessary serialization header
- xmlutils: wrap new() calls in nommgr/mmgr; allows tracking other
allocs in this file.
- add u64_from_u32
- various minor comments/improvements.
This was SVN commit r2604.
- add note on performance in TextureManager.cpp
debug_stl: more fine-grained #if STL_DINKUMWARE (gets this closer to
working on other STL implementations)
vfs.h: improved dox and moved vfs_path public functions here.
cursor: pass g_mouse_x as a parameter (SwEng)
This was SVN commit r2595.
wsdl: create window at expected size, and retrieve correct mouse coords
CStr: not ISerializable (=> not virtual)
Game: less excessive complaining, since the original problem seems to be
fixed
Entity: fixed evaluation order
Others: minor comments/etc
This was SVN commit r2591.
file: now provides a much more efficient alternative to file_enum that
does not sort all files (this was a problem for check_if_oal_dll which
was scanning the entire system32 dir)
- use the same DirEnt struct for file and VFS due to clever name pointer
trick.
- wposix: avoid malloc for most opendir() and fix mmap() bugs (thanks to
Philip for noticing+reporting them)
- VFS: was getting too large to manage, now split into an additional
file
- WIP: much more efficient mounting; supports other mount sources than
file or zip (e.g. if we want to add network files); clean separation
between VFS submodules; dox; beginning of thread locking.
This was SVN commit r2589.
- debug: add provision for naming threads. allows adding current thread
name to log messages and displays their names in the debugger.
- replaced various if(err < 0) complain() sequences with new variants of
CHECK_ERR (see lib.h)
- fixes to mmgr/VC debug alloc enable code
- improved h_mgr error reporting (now complains when h_free fails)
- US -> UK english (partial)
- fix tex_load double-free bug
- move win32 mouse cursor code into sysdep
- error dialog is now topmost to make sure it's visible (was a problem)
- handle WM_QUIT before displaying error dialog (makes sure it's shown)
also as in previous 3 revisions.
This was SVN commit r2588.
- add convenience macros for config_db(CFG_GET_SYS_VAL)
- VFSUtil::EnumDirEnts now uses flags instead of bool recursive
- UNUSED() for params, UNUSED2 (<- need better name) for variables
- config.h defines must be tested with #if (always defined) -> allows
detecting misspellings thanks to compiler warnings
- replace debug_assert(0) with debug_warn (its sole purpose)
- replace ScriptingHost::ValueToInt et al with ToPrimitive
- use nommgr.h to disable both mmgr and VC debug heap
This was SVN commit r2585.
main: add new variable for clarity
scriptglue: massive refactor. added docs, rearranged into groups,
eliminate forward decls, consistent param checking, some progress on
consistent naming and formatting.
This was SVN commit r2492.
ia32: fix comments and fix EBP warning
wdbg: fix bug if exception locus is unavailable
wdetect: store full DS driver path (-> version reported correctly)
This was SVN commit r2491.
snd: better explain __try hack; avoid any updates if not initialized
ia32: no longer user declspec(naked) (unsafe)
wdbg_sym: also declspec(naked) and refactor display of libstdc++ objects
wdll: remove redundant #undef
wsdl: add comment
This was SVN commit r2487.
h_mgr.h: debug_warn if H_DEREF fails
have vfs / res functions only CHECK_ERR (i.e. complain) once per call
sequence. made sure vfs_load and vfs_open debug_warn and everything else
just passes on the error.
This was SVN commit r2482.
snd: make __try() more obvious and fix old bug that causes a NULL read
when many sources are playing
add alloca implementation (takes care of stack overflow handling)
This was SVN commit r2475.
bring debug_get_nth_caller in line with debug_dump_stack; clarify
parameter dox.
use debug_get_nth_caller instead of debug_dump_stack to get exception
locus (much less evil)
fix out() clamping if buffer is full - no more chicken scratching at end
of buffer
This was SVN commit r2474.
snd.cpp - remove stack size note
vfs_tree - fix uninitialized variable problem
string_s - match official rename of strlen_s to strnlen
ia32 - fix infinite recursion, the apparent cause of bug#64
This was SVN commit r2470.
- add stdext and new STL containers to stl_simplify_name
- fix internal error when container is uninitialized (now check cntr
fields directly instead of calling e.g. size(), which can crash)
- all container wrappers now have a consistent interface, no more
specialization needed in iterator
- fix potential loophole with container adapters
- refactor dump_sequence
This was SVN commit r2464.
debug - documented
debug_stl - WIP; large changes because specialized containers were
breaking in some cases. now have to do it "right" ;p
add SLIST container and MULTISET
This was SVN commit r2461.
Updated DLLs.
Added DirectX headers/libraries, since they're not in the platform SDK.
Fixed invalid STL iterator usage.
Disabled STL in stack trace, since it breaks.
Added comments for some reported warnings.
Corrected some apparent off-by-one errors.
Defined strlen_s to strnlen.
Probably broke pre-beta2 versions of VC2005.
This was SVN commit r2454.
- increase stack trace buffer size (needed for some complicated nested
expressions)
- fix bug in container output - no longer treat iterator (i.e.
x::iterator) as an x string (now uses match_wildcard instead of simple
string compare)
- add unicode match_wildcard
- bugfix for vectored exception handler - was always ignoring exceptions
from main thread (fixed the previous fix where it was always grabbing
all exceptions ;p)
- better array display (display less of huge arrays)
- fix nesting limit (which was too low); make sure infinite
nesting/indirection can't happen
This was SVN commit r2448.
- revised dump_sequence to support arbitrary containers and added
dump_array wrapper for simplicity
- check UDT member offset against total size for safety
- disambiguate type and data ids
- prevent duplicated and cyclic display of pointers
This was SVN commit r2437.
- tentatively activate PCH if gcc > 3.4. remove XP_ defines (which were
undocumented; now in scripting/spidermonkey.h)
- rename some config defines
- move sysdep/debug.h to lib/debug.h (it's portable; was wrong all
along)
- grouped all debug heap related stuff to debug.h:debug_heap_*
- sysdep: fix circular include issue with debug.h
This was SVN commit r2427.
- add new portable display_error (and convenience macro DISPLAY_ERROR,
if advanced params aren't needed; don't want to break C compat). this is
the old assert dialog and more; allows ErrorReaction (see sysdep.h) to
be carried out according to users wishes.
- add win_is_locked for deadlock avoidance
This was SVN commit r2420.
- thread-safe init and stack trace output (more robust)
- better symbol display (no longer suppress all STL; UDT line splitting
- add debug_wprintf and debug_disable_leak_reporting
- exception handler now called from __try and vectored chain (if
supported by OS); will even work while being debugged and isn't stomped
on by some other UnhandledExceptionFilter users
This was SVN commit r2419.
- main init hook no longer calls main() - that was causing stack
weirdness and is forbidden. provide support for the app calling init
directly (in case someone else is already hooking main)
- wdetect: add support for old drivers that set OpenGLDrivers
incorrectly (my laptop, so I can test ;p)
This was SVN commit r2418.
wtime.cpp: fix comment
main: reorder includes, remove deadwood, rename a few functions for
consistency, moved config code together into one section, slight change
to system info text file (for new ogl driver detect code), remove
_control87 (see comment)
This was SVN commit r2392.
lib.h: reorder error codes (more clear) and add a few
detect.cpp, gfx.h: move gfx_mem (amount of video memory) here from ogl
This was SVN commit r2391.
(Just for fun, compressed texture uploads are mapped onto a
decompression function if the user's system doesn't understand S3TC
textures.)
Corrected (unless I'm wrong) GL version identification logic.
This was SVN commit r2384.
revised get-crash-location code (very much trial+error, as StackWalk64
doesn't report errors) - it now looks to be bulletproof.
TODO: symbol addresses in release mode appear to be incorrect => value
display is completely worthless.
This was SVN commit r2376.
wdbg: work around several dbghelp flaws (most importantly concerning
address of symbol, which was leading to completely wrong display).
refactored sym_dump_*. also further cleanup + dox.
This was SVN commit r2373.
wdbg: work around several dbghelp flaws (most importantly concerning
address of symbol, which was leading to completely wrong display).
refactored sym_dump_*. also further cleanup + dox.
This was SVN commit r2372.
Altered CStr/UniDoubler's macros, to avoid polluting/conflicting/etc.
Fixed headers in non-PCH builds.
Used the _d version of vorbisfile*.lib in Debug builds.
This was SVN commit r2366.
asserts now correctly retrieve thread context in release mode (worked
around VC problem with __asm in macros)
added sample code to enable SE_DEBUG_NAME (pending)
This was SVN commit r2361.
wdbg: major improvements to exception handler. now catches and displays
C++ exceptions (no small feat). also fixed a few bugs in stack trace;
minor improvements elsewhere.
This was SVN commit r2346.
debug.*: move crashlog write code here
wdbg: improvements to breakpoint code. major cleanup in exception
handler; de-hackified. mostly done, not yet stress-tested when not
running in debugger
This was SVN commit r2333.
- Ported lockless code to gcc inline assembly
- A few new net messages (gather, attack, add waypoint)
- Support for new messages in network->entity order converter
- Implemented rudimentary JS interface for Interaction
- issueCommand JS API, connected to the network
- Removed Interaction stuff now replaced by JS
- And something in there should probably break VS builds :P
This was SVN commit r2316.
vfs: no longer watch dirs when mounting - need to pass flag. allows
turning this off for screenshots dir (where hotloading doesn't make
sense)
vfs_tree: expand hash table when 3/4 full; 1/2 was quite wasteful.
main: do CPU init early, so timing during init can use TSCmain: do CPU
init early, so timing during init can use TSC
This was SVN commit r2098.
- add check for changing filters while enumerating dirents
- add path_component_valid
- change iterator to only return valid entries (removing that burden
from user code)
This was SVN commit r2067.
- replace std::map with custom filename lookup container (more
efficient)
- split into 3 parts (was too big)
also added struct stat-like fields to vfsDirEnt
still under heavy development.
This was SVN commit r2058.
numerous TODOs remaining - initial update instead of only after
completing first job; fix GUI overdraw issue; decrease granularity
(currently only 3 updates)
This was SVN commit r2033.
replaced that with posix_types.h, which now defines size_t.
this change exposed a lot of code that relied on posix.h being included
implicitly; fixed those as well.
This was SVN commit r2015.
lib/posix_types.h: wrapper for stdint / sysdep/win/wposix_types.h
included by types.h; replaces posix.h there.
(works around wsock conflicts by not pulling in all posix stuff
everywhere)
This was SVN commit r1973.
wdbg: fixed root cause of assert noted during 2005-02-05 meeting (now
skip UDT child symbols that don't have an offset, i.e. uninteresting
stuff like BaseClass symbols)
This was SVN commit r1915.
lib.h: move debug_warn to debug.h; use assert2 instead of assert
wdbg: better explanation of "get current CONTEXT"; remove "tag ==
SymTagData" assert (triggered when dumping UDT)
This was SVN commit r1890.
precompiled: restructured, documented everything, memory system headers
are now included before mmgr.h to prevent conflicts (before, had
#defined include guard of crtdbg.h and malloc.h and manually declared
their contents - a hack)
This was SVN commit r1885.
- hook strdup and getcwd, since they allocate memory internally
- better reporting of owner in case only partial information is
available (don't give up and report "unknown")
This was SVN commit r1879.
- if tag != SymTagDataType, don't dump the symbol (to be safe). this is
unexpected, but Philip reports it triggered.
- assert dialog is resizeable (grr, have to do that manually)
This was SVN commit r1828.
defining operator_new in a global file is a bad idea- if we forget
that, hard-to-track errors result
so, no more operator new; instead, #include "nommgr.h" before local
operator new definitions and #include "mmgr.h" again afterwards
This was SVN commit r1824.
mmgr.h: remove everything if !defined(USE_MMGR); move dox to end of file
debug.h: fix stupid n00b bug in assert2 (expr wasn't in "()" )
This was SVN commit r1822.
mmgr: remove wdbg calls (was prototype); use sysdep/debug interface
wdbg: no longer call it dbghelp63.h; should be installed normally in
compiler dir
win_internal.h: fix inevitable MS screwups in new dbghelp.h (*sigh*)
This was SVN commit r1816.
- sysdep/debug.h includes platform-specific version, so they can define
debug_break
- no longer a debug_break wrapper that calls to specific versions (so
that the debugger breaks in the target function directly)
- unix-specific stuff is only debug ATM; call it udbg instead of unix.*
This was SVN commit r1813.
added flags to vfs_mount that enable looking for archives and mounting
subdirectories (previously always done, but slow)
main: uses above flags when mounting
This was SVN commit r1695.
vfs: fix archive file add routine: wasn't starting at mount point,
didn't handle files in archive root dir correctly
tex: add tex_is_known_fmt
lib: add ERR_TIMED_OUT
This was SVN commit r1693.
tex: improve flip / bgr convert (now generalized "transforms"); add
support for transforming image before encoding
wsdl: add SDL_ShowCursor, minimize when tasking out
main: WriteScreenshot improved, fixes flipped screenshots
This was SVN commit r1689.
- complete+document+fix texture orientation code
- add JPG codec
- file extension check bugfix: now case-insensitive
also some general cleanup.
This was SVN commit r1660.
- problem when using CHECK_ERR with functions returning a Handle
- incorrect sign conversion when passing non-ascii (> 0x80) chars to
FOURCC
precompiled.h: include lib/types.h
This was SVN commit r1654.
2) we attempt to load missing files (see bug #32). with the fix above,
CHECK_ERR would fire every run (which it should until the problem is
fixed :P), which is annoying.
Instead, FileUnpacker now checks if the file exists; if not, it throws
immediately and thereby bypasses the VFS error mechanism.
This was SVN commit r1503.
2) VFS: add path_copy convenience routine
3) various lib code: removed unnecessary const cluttering things (like
const char* _const_ name in param list)
This was SVN commit r1483.
2) wposix: cache file status in struct dirent (Win32 specific)
3) vfs: speed up vfs_mount (via 2) and zip_cb path lookup caching
scheme)
4) temp scaffolding added to file.cpp to measure performance
This was SVN commit r1482.
- reworked tree_lookup: dir names now require trailing '/' to
disambiguate. exception: vfs_open_dir (convenience)
- display VFS tree in stdout.txt unless g_Quickstart
This was SVN commit r1466.
- Connecting client(s) to a server and giving orders to a shared set of
entities now works - Yeah!
- Removed thread sync from CNetServer
This was SVN commit r1180.
- have vfs_load allocate the buffer, so mem_assign is no longer needed.
exposes buffer alignment requirement to VFS, but hey. simplifies things
a good bit.
- moved performance measuring crap to vfs_timed_io, a wrapper function
- various fixes for streaming code
This was SVN commit r1159.
