LOADB/SAVEB hang on invalid device

[jkotlinski]

12 device parse-name foo $7000 loadb <-- hangs

12 device $7000 $7100 parse-name foo saveb <-- hangs

[Whammo]

IEC commands do not freeze on invalid device.
#482

 code ds
      w stx,
    $00 lda,#
    $90 sta, \ clear st
    $ba lda,
  $ffb1 jsr, \ listen
    $6f lda,#
  $ff93 jsr, \ second #15
  $ffae jsr, \ unlisten
    $90 lda, \ readst
+branch bne, \ device not present?
\ checking for the device is optional
\ but a good idea.
    $ba lda,
  $ffb4 jsr, \ talk #8
    $6f lda,#
  $ff96 jsr, \ tksa #15
here
    $90 lda, \ readst
+branch bne,
  $ffa5 jsr, \ acptr get byte
  $ffd2 jsr, \ chrout
swap    jmp,  \ loop
:+
  $ffab jsr, \ untalk
:+
      w ldx, ;code

[Whammo]

Which means DOS will also freeze when device is invalid.

[Whammo]

It's not load or save that freezes, it's open.

[Whammo]

It might be simpler to vector kernal open through something like this:

        LDA #$00
        STA $90       ; clear STATUS flags

        LDA $BA       ; device number
        JSR $FFB1     ; call LISTEN
        LDA #$6F      ; secondary address 15 (command channel)
        JSR $FF93     ; call SECLSN (SECOND)
        JSR $FFAE     ; call UNLSN
        LDA $90       ; get STATUS flags
        BNE .devnp    ; device not present
        JMP open
.devnp
        ... device not present handling ...
        RTS

[jkotlinski]

The original code for reading error channel is here: https://codebase64.org/doku.php?id=base:reading_the_error_channel_of_a_disk_drive

" A small warning: Both the BASIC and the Assembler versions will deadlock if the device is not present. "

Edit: It also presents an alternate version which avoids the deadlock.

Edit 2: OK, it looks like you found the same code :-)

[jkotlinski]

OK, I think the above commit might work.

The codebase64 page says:

"In this version the status-byte is accessed directly which reduces the portability of the code." <-- That's not great, but OK...

"This will limit you to IEC-bus devices" I am not sure what practical consequences that has. Would one ever want to LOADB/SAVEB to something else?

[Whammo]

You can use kernal readst, but unlike basic, it doesn't clear after reading.
EOF (st=64) isn't really an error is it?
Save to cassette??

[Whammo]

We have to reset status before IEC commands because they only change ST on failure.

[Whammo]

Yes, DOS, INCLUDED and OPEN, and anything with kernal open is subject to hang on device not present.

[Whammo]

All the info I used to master kernal IEC

https://commodore.software/downloads/download/215-articles-and-guides/15601-iec-disected

[Whammo]

Another advantage of IEC commands is they don't consume file numbers.

[jkotlinski]

When testing with #501, these cases produce error messages and no hang:

12 device parse-name foo included

12 device dos scratch0:test

Maybe everything is fine with #501? Am I missing something?

[Whammo]

12 device dos
no string freezes
12 device dos i
okay. workaround?
INCLUDED working correctly is good news.

[Whammo]

The problem is a bug with the Kernal opening a file with no file name.
It creates the file entries in the tables so chkin doesn't error , then exits.
The command channel doesn't require an (IEC) open or a close,
so it doesn't check if the device is there before it goes to chrin, (acptr) then it gets stuck waiting.
One solution is to never close 15.

[jkotlinski]

One could add a 0 check to send-cmd, but maybe it is equally fine to let user exit the bad state by pressing RESTORE? After all, DOS with empty string is not a real use-case?

…

On Wed, 28 Dec 2022 at 10:56, Whammo ***@***.***> wrote: The problem is a bug with the Kernal opening a file with no file name. It creates the file entries in the tables so chkin doesn't error , then exits. The command channel doesn't require an open or a close, so it doesn't check if the device is there before it goes to chrin it gets stuck waiting. — Reply to this email directly, view it on GitHub <#499 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAY34O3CHTE3DBGEVCRIWBTWPQFFXANCNFSM6AAAAAATKIP26I> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[Whammo]

DOS with an empty string prints the current device status if it's there, and hangs otherwise.
This is why BASIC insists upon device numbers for anything other than tape.

[jkotlinski]

I see, so it is a use case.
Maybe send-cmd should be rewritten, too, to use IEC instead of OPEN?

[Whammo]

: send-cmd ( addr len -- )
?dup if
$f $f open ioabort 
clrchn $f chkin ioabort
begin chrin emit readst until
clrchn $f close cr
else drop _errorchread then ;

perhaps?

But yes it's doable.

[jkotlinski]

That seems like a fine solution. Added!

[Whammo]

I think there were some additions to V that can be rolled back now.

[jkotlinski]

Oh really? What V additions?

[Whammo]

added v check for bad device# ?

[jkotlinski]

ah right! great catch. thank you!

[Whammo]

At some point in the future maybe, we can get durexForth off of the filesystem and onto IEC commands so file numbers used by the system would go away.

[jkotlinski]

That stuff is a bit above my head probably :-)

[Whammo]

Serial IEC is certainly arcane. The documentation is sparse. The IEEE parallel PET stuff is well documented, of which the serial stuff is a tiny subset, the overview makes it understandable.

[Whammo]

Wow! Thanks. I'll try not to f- anything up. 🥇