Write binary data to TRD from ASM

[Hikaru]

I've made a little progress. The error returned in $5D0F was 1 No file. i wasn't padding my filename with spaces in the file descriptor.

Now I get error 12 The channel is not open in $5D0F. But I have already been using TR-DOS - the code this is running from is loaded from basic with a TR-Dos call, so Dos must be initialised, surely?

Meh. It fails since it's actually doing a 'verify' operation - because guess what, there are actually two variables responsible for this, the second one being #5D10. Neither Rodionov's book nor the disassembly mention that one (I don't even see it listed as a TR-DOS variable), although it's possible to trace where the check occurs in the disassembly.
It looks like @keith56 has figured this out on his own previously, since both #5CF9 and #5D10 are being set up in his code.

With this in mind, the loader example finally becomes:

Code: Select all

LoadExample
	ld hl,FileDescriptor
	ld c,#13		;#3D13 C=#13: copy the file descriptor to TR-DOS system variables
	call #3D13
	xor a
	ld (#5CF9),a		;#00: load, #FF: verify
	ld (#5D10),a
	dec a			;load the file to an address in HL
	ld hl,MyAddress
	ld c,#0E
	call #3D13		;#3D13 C=#0E: load/save file
	ret

FileDescriptor
	DEFB "FILENAME"		;8 characters
	DEFB "C"		;file type: code

Fingers crossed! xD

On a bit unrelated note,
1. OUTputting a random value from #5B5C to (#7FFD) at the start of your main code and hoping for the best isn't exactly a good idea. There's no telling which ROM/screen/page that would activate, or indeed whether you won't be screwed the next time an IM1 interrupt comes. It's better to set both of them to a known value of your own if you have to.
2. The correct spelling for the last option is 'Обнулить рекорды' (or 'Обнуление рекордов'). Regardless that's pretty cool I think, you don't often see people translating their games to Russian on their own.

[Ast A. Moore]

Neither Rodionov's book nor the disassembly mention that one (I don't even see it listed as a TR-DOS variable), although it's possible to trace where the check occurs in the disassembly.

Well, there’s this, which describes the variable at #5d10 as “MSB of TR-DOS error code, cleared on 15616 calling, should be cleared when you call 15635.”

Then there’s the manual for BETA 128 interface itself, which isn’t particularly useful, but has a little assembly snippet nevertheless.

Code: Select all

CHADD  EQU   23645         Location of SOS variable CHADD
       ORG   XXXXX         XXXX=address of this code
       LD    HL,(CHADD)    Start to save true CHADD
       LD    (TEMP), HL    Temporary store of true CHADD
       LD    HL,49000      Address of SAVE routine
       LD    (CHADD),HL    CHADD now points to our routine
       CALL  15363         ENTER TRDOS SAVE via chadd
       JP    BACK          Jump to program point from which
                           the whole routine was called
       LD    HL,(CHADD)    
       LD    (TEMP),HL     The routine for LOAD now repeats
       LD    HL,49500      the above with just the address
       LD    (CHADD),HL    changed.
       CALL  15363         
BACK   LD    HL,(TEMP)     Start to restore CHADD
       LD    (CHADD),HL    Reload original CHADD
       RET                 Return from where you came
TEMP                       Label allocating memory for
                           temporary storage

[Seven.FFF]

Yay, ld (#5D10),a makes it work! Cheers guys

Will write a longer reply later.

[Seven.FFF]

Meh. It fails since it's actually doing a 'verify' operation - because guess what, there are actually two variables responsible for this, the second one being #5D10. Neither Rodionov's book nor the disassembly mention that one (I don't even see it listed as a TR-DOS variable), although it's possible to trace where the check occurs in the disassembly.

Awesome, thanks. I will get there eventually. Somehow it's harder integrating new info in a foreign language until you reach a tipping point. I don't know how you multilingual guys do it, but you rock.

1. OUTputting a random value from #5B5C to (#7FFD) at the start of your main code and hoping for the best isn't exactly a good idea. There's no telling which ROM/screen/page that would activate, or indeed whether you won't be screwed the next time an IM1 interrupt comes. It's better to set both of them to a known value of your own if you have to.

You're right. This is code that's I copied and pasted between projects, and somehow I have strayed away from the original intention - which was to set up the environment in a deterministic state after external loaders (mostly DivMMC/exsDOS) have dumped you into the starting point. Clearly I'm not doing what I intended here!

2. The correct spelling for the last option is 'Обнулить рекорды' (or 'Обнуление рекордов'). Regardless that's pretty cool I think, you don't often see people translating their games to Russian on their own.

Thanks! I have had help I'm afraid. Mikhail Sudakov (Castlevania Spectral Interlude) kindly offered to help with translations. He already warned me about context and word-ending changes. I woke up to the same message from him as I did from you!! Обнулить таблицу рекордов was too long - and I just now saw that he gave me Обнулить рекорды as an alternative right from the beginning

Well, there’s this, which describes the variable at #5d10 as “MSB of TR-DOS error code, cleared on 15616 calling, should be cleared when you call 15635.”

Another piece of the puzzle, cheers

Then there’s the manual for BETA 128 interface itself, which isn’t particularly useful, but has a little assembly snippet nevertheless.

I had been trying to get this one working too. This is the "official" API that passes tokenized basic commands to TR-Dos. I ran into issues with my tokenization - the stuff in my reference REM statements were being tokenised as "C", "O", "D", "E" etc and it was causing confusion with some other problems I was having at the same time. It clearly works in my basic loader though, so I could revisit this later. But I don't really need to.

I now have an asm loader that loads everything except $6000-$7FFF (and updates my progress bar). Loading this block is breaking TR-Dos, so I need to debug to see what's going on. I don't really want to rearchitect the game because the codebase is shared with the non-Pentagon version (and space in unpageable RAM is TIGHT!!), so I will have to figure out a way to have code from $6000-> and stack from <-$5FFF, and still keep Dos working well enough to save my high scores every game end. Also I'm using the ZX Printer buffer as a general purpose buffer in a ton of places, so will need to figure out something different there.

Are there any special Pentagon-only horizontal MMU schemes that would help, by any chance? Mostly a rehetorical question - I have lots of reading to do...

[Ast A. Moore]

Something interesting for you to investigate. This version of Battle Command has an unusual loader. It displays a scrolling text and plays music while loading the game. Might be worth reverse engineering.

[Hikaru]

I now have an asm loader that loads everything except $6000-$7FFF (and updates my progress bar). Loading this block is breaking TR-Dos, so I need to debug to see what's going on. I don't really want to rearchitect the game because the codebase is shared with the non-Pentagon version (and space in unpageable RAM is TIGHT!!), so I will have to figure out a way to have code from $6000-> and stack from <-$5FFF, and still keep Dos working well enough to save my high scores every game end. Also I'm using the ZX Printer buffer as a general purpose buffer in a ton of places, so will need to figure out something different there.

File level operations use a temporary extra buffer of about 256 bytes that is located after the regular TR-DOS variables (110-something bytes). Any BASIC data in the area gets temporarily saved higher in the memory. So I'd suggest you to clear the BASIC part of your loader first. 24576 doesn't seem that low in the memory, so that might just be enough to make it work.

I think this should do the trick, but don't take my word for it.

Code: Select all

ClearBasic
	ld hl,(#5C59)		;E-LINE: end-marker of the current variables area - the '80-byte'
	ld de,(#5C53)		;start of BASIC program area
	dec hl
	call #19E5		;RECLAIM-1

[Seven.FFF]

Still messing around with this. I had several errors in how I was saving my binary chunks and also how I was loading them. No wonder it wasn't working. Hopefully will have time to finish debugging this weekend. Thanks @Hikaru.

In the meantime I knocked up a repeating attract thingy on the menu. Amazingly I managed to get it locked to the alien timings so it doesn't drift out of sync. It's a bit jerky cos I don't have a good gif grabber that works with Pentagon timings.

[Ast A. Moore]

It's a bit jerky cos I don't have a good gif grabber that works with Pentagon timings.

Fuse > fmfconv (Fuse utilities), maybe?

The menu looks splendid, by the way. Love the proportional font.

[Seven.FFF]

Another thing I'm stuck with - at the conceptual level.

I'm trying to save two files to disk. One is only 1 byte (the selected language, 1 sector), the other is 960 bytes (the high score table), 4 sectors.

I can write files with call #0B (Save <C> file). This only works if I save a new file, though. If I save an existing file, I get error 0 returned, suggesting it worked, but the contents don't change.

I can also delete the file with call #12. Then saving it works, but creates a new file at the end of the catalog, and leaves the sectors of the original file marked as deleted.

I would prefer to save the same file repeatedly, with the same filename, in the same sector. I'd rather not MOVE the disk programmatically to reclaim the deleted sectors.

There exist the sequential and random access files that can be used with OPEN #/PRINT #/INPUT#/CLOSE #. Will they do the job? Can they be only called the "official" API way, by placing tokenized Basic commands in memory? It looks like these have the disadvantage of not being able to be used on <C> files, as they have their own special <#> type.

There also exist PEEK/POKE "filename"Buffer_address, Sector_numbers commands. It looks like these can be used to read and write <C> files. Do they also have to be used with the "official" API?

There are calls to read (#05) and write(#06) individual sectors, but I can't see how to read the track and sector numbers from the catalogue. I know what my initial sectors are, as I built the disk, but I would corrupt the disk if I assumed that, as someone might do their own disk operations that moved the file.

What do you think the best and easiest way to do this is?

I also discovered that writing to a read-only disk displays the nice Retry, Ignore, Abort message, which clears my screen. I worked round this by saving and restoring the screen before and after all my disk commands. I read in Russian about a service screen - is this what's happening here? Has my screen already been saved, and I could restore it by calling TR-DOS?

[Hikaru]

The whole 'leaving gaps' issue is normal in a sense, and is expected of programs that operate on the file level. Notice that, as long as the file you're working with is the last file on the disk, no gap would be left if it is deleted and subsequently saved back.
On the other hand, doing a MOVE within a game program is certainly an overkill and not what most people would expect. It's a bit like if a PC game tried to defragment the HDD on its own!

Random access and other unusual formats are best avoided. I doubt most people have any idea what those are (myself included), and I don't remember a single program that used those on a regular basis for its functionality.

(Why not just keep a single 961 byte file? Loading 4 sectors and 1 sector takes virtually the same time, and this is more efficient in terms of using the disk space.)

There are calls to read (#05) and write(#06) individual sectors, but I can't see how to read the track and sector numbers from the catalogue. I know what my initial sectors are, as I built the disk, but I would corrupt the disk if I assumed that, as someone might do their own disk operations that moved the file.

You can use a combination of #0A and #08 to read the track/sector information, and then read/save the data with #05 and #06. Sort of like this:

Code: Select all

Overwrite
	ld hl,FileDescriptor
	ld c,#13		;copy the file descriptor to TR-DOS system variables
	call #3D13
	ld a,9			;find a file with matching 9 first bytes in the descriptor (filename and type)
	ld (#5D06),a		;uses the extra buffer, as do file level operations
	ld c,#0A
	call #3D13
	ld a,c
	inc c			;C will contain #FF if the file with this name wasn't found
	jr z,NoFile
	ld c,#08		;#3D13 C=#08: loads a file descriptor specified in A at #5CDD
	call #3D13
	ld hl,SaveAddress	;or LD HL,(#5CE6) to assume the same saving address as in the file descriptor
	ld de,(#5CEB)		;file descriptor: track and sector
	ld a,(#5CEA)		;file descriptor: length, sectors
	ld b,a
	ld c,#06		;#3D13 C=#06: save sectors
	call #3D13
	ret

Another option is to do the #0A/#08 thing once in the beginning, store the file locations/lengths in memory, and then just use #05/#06 as needed. This would help to reduce unnecessary drive head movements to track 0 and back (is faster). Also, #05/#06 have the advantage of not requiring the extra buffer in order to work.

I also discovered that writing to a read-only disk displays the nice Retry, Ignore, Abort message, which clears my screen. I worked round this by saving and restoring the screen before and after all my disk commands. I read in Russian about a service screen - is this what's happening here? Has my screen already been saved, and I could restore it by calling TR-DOS?

It's definitely not doing anything that convenient.

The correct way to go about this is to intercept the Retry/Abort/Ignore message. The following code comes from my unfinished #3D13 driver so again, pretty much untested, but I believe it should be something along these lines.

Code: Select all

OneTimeInit
	ld a,#C3		;#5CC2: is used by TR-DOS to call the BASIC 48 ROM routines, typically contains a RET
	ld (#5CC2),a		;is also executed upon error during a #3D13 call (see #0201)
	ld hl,ErrorHandler	;here, it is set to 'JP ErrorHandler'
	ld (#5CC3),hl
	ret

DOS
	ld,(iy+0),#FF		;set #5C3A to 'no error', this is checked before exiting in #0201 but isn't reset by the DOS
	push hl
	ld (#5C3D),sp		;#5C3D: ERRSP
	ld hl,0			;zero out the two byte error code at #5D0F
	ld (#5D0F),hl
	pop hl
	call #3D13
	ret z			;Z: OK, NZ: ERROR
	ld a,c			;C = error code
	cp 20
	jr z,Break		;20: break
	sub 6
	jr z,NoDisc		;6: no disc
	dec a
	jr z,DiscError		;7: disc error (abort)
	jr UnhandledError


;intercept the Retry/Abort/Ignore message
ErrorHandler
	ex (sp),hl		;will eventually return if entered upon NO DISC or ABORT, with the error code in C
	push af
	ld a,h
	cp #0D			;the RIA situation is indicated by #0D6B on the stack (both 5.03 and 5.01)
	jr z,.ria
	pop af			;otherwise, we continue
	ex (sp),hl		;LIST, CAT, FORMAT and the like display additional messages, not handled here
	ret
.ria
	ld hl,12		;10+2
	add hl,sp
	ld sp,hl
	call RIAMessage		;our custom RIA procedure here, A should return an "R"/"I"/"A" ASCII code
	ld hl,#3F7E		;5.01 uses #3F74
	push hl			;return the user's response to TR-DOS
	jp #3D2F