mirror of
https://github.com/TomHarte/CLK.git
synced 2024-11-23 18:31:53 +00:00
228 lines
9.0 KiB
Plaintext
228 lines
9.0 KiB
Plaintext
Programs incompatible with EmuTOS
|
|
=================================
|
|
This is a list of programs that have program bugs or shortcomings that
|
|
prevent them from running properly with EmuTOS, and whose problem has
|
|
been definitively identified. It is mainly intended to prevent these
|
|
programs from being added to 'bugs.txt' in the future.
|
|
|
|
|
|
Program: STOS programs
|
|
----------------------
|
|
Error 1: joystick and/or keyboard input doesn't work.
|
|
|
|
Bug analysis:
|
|
STOS Basic compiler routines check for input using undocumented and
|
|
TOS-specific locations. Programs using these routines work only with
|
|
specific TOS versions, and not with EmuTOS.
|
|
|
|
Workaround:
|
|
Use version of the program that has been fixed to work with modern TOS
|
|
versions.
|
|
|
|
Error 2: STOS error message "Error #046, press any key" during startup.
|
|
|
|
Bug analysis:
|
|
This is caused by a divide-by-zero error in vsm_height() when the
|
|
program is run from the AUTO folder. VDI initialisation does not occur
|
|
until after AUTO-folder programs have been run, so if a VDI function is
|
|
called by an AUTO-folder program, internal variables have not been
|
|
initialised. STOS tries to initialise these variables itself, based on
|
|
a built-in table of TOS-specific locations. These locations are invalid
|
|
for EmuTOS, so the required variables remain zero, causing the error.
|
|
|
|
Workaround:
|
|
Move the program from the AUTO folder to the root of the drive, and run
|
|
it from there. While STOS programs will then start, most of them will
|
|
remain unusable due to error 1 above.
|
|
|
|
|
|
Program: old game using fixed addresses
|
|
---------------------------------------
|
|
Error: panic during game startup.
|
|
|
|
Bug analysis:
|
|
Several old, floppy-only games load their data into fixed memory
|
|
addresses, which won't work when EmuTOS has loaded something else
|
|
there. This can be detected by tracing programs' OS calls with Hatari
|
|
(--trace os_all) and checking the used addresses. For example, under
|
|
older EmuTOS versions, the Gods game demo (from an ST Format cover disk)
|
|
overwrote itself with its game data because of this, and crashed.
|
|
|
|
Workarounds:
|
|
In some games this can be worked around by either using the cartridge
|
|
version of EmuTOS (which uses less memory) or by using a custom high-RAM
|
|
build of EmuTOS, that uses higher RAM addresses for loading programs and
|
|
for memory allocation:
|
|
make 512 UNIQUE=uk DEF="-DSTATIC_ALT_RAM_ADDRESS=0x00080000 -DCONF_WITH_FRB=0"
|
|
|
|
However, this doesn't help with programs which also expect undocumented,
|
|
OS internal functions and variables to be at certain locations. The
|
|
best workaround is to use a version of the game that has been fixed to
|
|
run from HD and with modern TOS versions.
|
|
|
|
|
|
Program: Awele v1.01
|
|
--------------------
|
|
Error: mono desktop colours are inverted after exiting program.
|
|
|
|
Bug analysis:
|
|
This version of Awele was compiled with PureC and linked with a very
|
|
early version of Windform. During WinDOM initialisation, malloc() is
|
|
called to allocate an area to save the palette in. However, although
|
|
malloc() returns the pointer in A0, the WinDOM code assumes it is in D0.
|
|
As a result, an area of low memory is pointed to, which is overwritten
|
|
during Awele execution. At program termination, the palette is restored
|
|
from the overwritten area, resulting in the error seen.
|
|
|
|
Workaround:
|
|
Use v1.02 of the game.
|
|
|
|
|
|
Program: Cameleon
|
|
-----------------
|
|
Error 1: program exits immediately when 'Start game' is selected.
|
|
|
|
Bug analysis:
|
|
The program requires an STe. In order to determine whether it is
|
|
running on an STe, it checks the contents of location 0x995 (hardcoded).
|
|
On Atari TOS, this is where TOS initialisation happens to store the _MCH
|
|
cookie but this is *not* how Atari says you should locate it (and it's
|
|
not where EmuTOS stores it).
|
|
|
|
Error 2: program crashes with a Trace exception on any exit.
|
|
|
|
Bug analysis:
|
|
During startup, the program switches to supervisor state via the Super()
|
|
system call. Subsequently, the supervisor stack overwrites the program's
|
|
user stack. On exit, the user stack pointer is restored and during this
|
|
a corrupted value is loaded into the SR, causing a trace excpetion.
|
|
|
|
|
|
Program: (VDI) Invaders and Anduril
|
|
-----------------------------------
|
|
Error: keys to move an object are ignored (in Invaders, the '-' key; in
|
|
Anduril, the 'h' & 'j' keys)
|
|
|
|
Bug analysis:
|
|
Both programs were written by "M.Dheus" who found that the most recent
|
|
key input from the keyboard was stored at offset 0x6d from the address
|
|
returned by Kbdvbase(), and used that to read the keyboard. This was
|
|
never documented by Atari, but was apparently true for all versions of
|
|
TOS 1. However it is not true for TOS 2, 3, or 4 (or EmuTOS).
|
|
|
|
|
|
Program: Ramses
|
|
---------------
|
|
Error: panic
|
|
|
|
Bug analysis:
|
|
Program calls the Line A initialization $A00A and gets the routine
|
|
vectors in a2. It gets the address of _v_hide_c, then starts doing
|
|
undocumented things with the bytes of the actual routine:
|
|
https://sourceforge.net/p/emutos/mailman/message/30605378/
|
|
|
|
|
|
Program: STVidPlay
|
|
------------------
|
|
Error: "Error in getting file location"
|
|
|
|
Bug analysis:
|
|
Program looks for a specific 2-byte sequence in the hard disk driver
|
|
code pointed to by hdv_rw ($476). If it doesn't find that sequence
|
|
within bytes 6-48 (or thereabouts) of the start of the driver, it
|
|
gives the error message.
|
|
|
|
|
|
Program: Cubase Lite
|
|
--------------------
|
|
Error: panic
|
|
|
|
Bug analysis:
|
|
On TOS 1.62 etv_timer vector is a delegate to an internal private
|
|
function. Cubase Lite tries to guess the address of that private
|
|
function in an undocumented way, which crashes on EmuTOS. (Somebody
|
|
could write a loader or TSR to change the etv_timer function so that
|
|
Cubase will not crash.)
|
|
|
|
|
|
Program: Reservoir Gods games (Bugger, Bunion, SkyFall, Sworm)
|
|
--------------------------------------------------------------
|
|
Error: panic
|
|
|
|
Bug analysis:
|
|
Games use an undocumented TOS4 vector for keyboard input instead of
|
|
accessing kbdvec correctly. This causes EmuTOS to panic.
|
|
|
|
Workaround:
|
|
This can be worked around with the following hack.prg:
|
|
https://sourceforge.net/p/emutos/mailman/message/26841274/
|
|
|
|
|
|
Program: OMIKRON.BASIC V3.01 (interpreter)
|
|
------------------------------------------
|
|
Error: Panic (bus error) during start
|
|
|
|
Bug analysis:
|
|
The program relies on undocumented internal TOS variables at several
|
|
points. First, it expects A0 upon return from Mediach (BIOS function)
|
|
to point to wplatch (floppy write protect latch variable). On EmuTOS
|
|
A0 is 0 and hence a bus error occurs when the program wants to modify
|
|
that variable. Second, it parses the bytes of the 0xA00A (hide cursor)
|
|
line-A routine to get the address of a variable reflecting the internal
|
|
state of the mouse cursor. This is done with the same code used in
|
|
"Ramses" (see above). This also fails on EmuTOS, resulting in another
|
|
bus error. There may be more accesses to undocumented variables.
|
|
|
|
|
|
Program: STSpeech v2.0
|
|
----------------------
|
|
Error: panics due to stack corruption
|
|
|
|
Bug analysis:
|
|
The program installs a custom Timer A interrupt handler, and calls the
|
|
XBIOS from it. If the Timer A interrupt happens to occur just when an
|
|
unrelated BIOS/XBIOS call is manipulating _savptr (saving registers),
|
|
then the nested XBIOS call inside the Timer A handler will trash that
|
|
BIOS/XBIOS save area, possibly modifying the stack pointer. In the
|
|
"Hitchhiker's Guide to the BIOS", Atari documented a workaround for this,
|
|
but STSpeech does not use it.
|
|
|
|
Workaround:
|
|
Because this problem is timing-dependent, it does not show up on Atari
|
|
TOS, and only shows up in EmuTOS if BigDOS is installed (BigDOS issues
|
|
many XBIOS calls). Use program without BigDOS, or anything else doing
|
|
a lot of XBIOS calls.
|
|
|
|
|
|
Program: Protracker v2 STE (Equinox version)
|
|
--------------------------------------------
|
|
Error: crash when "DISK OP." button is selected
|
|
|
|
Bug analysis:
|
|
The program relies on a TOS-specific code sequence, as follows:
|
|
1. it searches the ROM (pointed to by location 4) for the first word
|
|
equal to 0x47e
|
|
2. when found, it uses the longword immediately before as a pointer to
|
|
an address; in TOS2, this is a pointer to the mediachange handler
|
|
3. it stores the long at offset 0x1c from that address in its data
|
|
segment; in TOS2, this is a pointer to (I think) two bytes of
|
|
status for the floppy drives
|
|
Subsequently, when "DISK OP." is selected, the stored long is used as a
|
|
pointer. In TOS2, the value stored is $4216; in EmuTOS, it's zero,
|
|
resulting in a crash.
|
|
|
|
|
|
Program: Spectrum 512
|
|
---------------------
|
|
Error: crash during program initialisation
|
|
|
|
Bug analysis:
|
|
The program relies on a TOS-specific code sequence, as follows:
|
|
1. it searches the VBL handler (pointed to by location $70) for the
|
|
first word containing a value of 0x6100
|
|
2. when found, it uses the word immediately following as an index to
|
|
generate an address, and starts searching at that address for a
|
|
word containing a value of 0x8606
|
|
Under EmuTOS, the address generated is a nonsense address which happens
|
|
to be odd, causing an immediate address error.
|