“Interesting. I wonder if this is related to the “BOTH” button that got cut in SQ4.” — @ATMcashpoint

I don’t know about the both button that ScummVM adds to some SCI games, but there’s quite literally no way it could work by just adding a third button state. There’s a fair bit of script logic that’d need to be overhauled. Here’s why that is, and here’s how I did it in The Dating Pool.

I could have used the SCI Companion template game to compare against and document, but to be honest it’s a bit of a mess, as you could expect from a decompilation. The leaked system scripts are much neater to work with, even though the actual script code is basically identical.

Original Messager.sc:

(method (sayNext theMod theNoun theVerb theCase theSeq &tmp aTalker [theBuf 200] msgkey)
  ; If we were called with arguments, grab the text for that entry.
  ; If not, grab the next entry in the sequence.
  (if argc
    (= aTalker (Message msgGET theMod theNoun theVerb theCase theSeq @theBuf))
  else
    (= aTalker (Message msgNEXT @theBuf))
  )
 
  ; If we have voice enabled, allocate space and grab the entry's tuple.
  ; This block is missing in SQ4CD.
  (if (& gMessageType CD_MSG)
    (= msgkey (Memory memALLOC_CRIT 12))
    (Message msgLAST_MESSAGE msgkey)
  )
 
  (if (and  aTalker
            (or  (not lastSequence)
                 (and  lastSequence
                       (<= curSequence lastSequence)
          )
        )
      )
    ; Look up the Talker (or Narrator) by number.
    ; aTalker was a number, but now it'll be an object pointer.
    (= aTalker (self findTalker: aTalker))
 
    (if (!= aTalker -1)
      (talkerSet add: aTalker)
 
      ; Now let our Talker handle the rest.
      (if (& gMessageType CD_MSG)
        (aTalker
          modNum: theMod,
          say:    msgkey self ;<-- pass ONLY the tuple
        )
      else
        (aTalker
          modNum: theMod,
          say:    @theBuf self ;<-- pass ONLY the string
        )
      )
      ; In SQ4, we just always pass only @theBuf. There's some major
      ; voodoo involved in getting it to work.
 
      (++ curSequence)
    )
    ; Cutting a bit of irrelevant fastcast voodoo
  )
  ; Dispose of the space we allocated for the voice tuple, if needed.
  (if (& gMessageType CD_MSG)
    (Memory memFREE msgkey)
  )
)

Catdate’s Messager.sc:

; Exactly the same as in the template BUT...
(if (!= aTalker -1)
  (talkerSet add: aTalker)
  ; Pass both the buffer AND the tuple, no matter our settings.
  ; That does mean that msgkey may be null, but say won't use it in
  ; that case anyway.
  (aTalker
    modNum: theMod
    say: @theBuf msgkey
  )
  (++ curSequence)
)

In SQ4CD, the Narrator has extra noun, verb, and sequence properties that get set to allow the text to work. It’s really quite a bit of a mess, and my hat’s off to whoever on the ScummVM team got that Both mode to work. I was going to document it but got lost trying, it’s that wild.

On to the Narrator and by extension Talker!

Original Talker.sc:

(method (say theBuf whoCares)
  (if theIconBar (theIconBar disable:))
  (if (not initialized) (self init:))
 
  (= caller
    (if (and (> argc 1) whoCares)
      whoCares
    else
      null
    )
  )
 
  ; Figure out what to do with the message.
  ; Note that in one case, theBuf is a string...
  (if (& gMessageType TEXT_MSG)
    ; (method (startText theBuf &tmp strLength)
    (self startText: theBuf)
  )
  ; ...but in the other it's a tuple!
  (if (& gMessageType CD_MSG)
    ; (method (startAudio theKeys &tmp m n v c s)
    (self startAudio: theBuf)
  )
 
  ; cut a bit...
 
  ; start___ will have set ticks to the length
  ; of the string or recording. We add one more
  ; second regardless.
  (= ticks (+ ticks 60 gameTime))
  (return true)
)

Catdate’s Talker.sc:

; much the same, but
  (method (say theText theAudio whoCares)
    ; ...
    (if (& gMessageType TEXT_MSG)
      (self startText: theText)
    )
    (if (& gMessageType CD_MSG)
      (self startAudio: theAudio)
    )
    ; ...
  )

Now, this works fine. If I record a quick bit of gibberish, then load up the game, switch to Both, and click, I get a perfectly readable message and hear my gibber. But if I were to revert my little change and use the original code…

That’s what we in the business call mistaking a bunch of numbers for a valid string. I specifically get this result because the first value in the tuple is the module number, which is 110 (0x6E ‘n‘) in this case, and all numbers in SCI are 16-bit so there’s a terminating null right after.

What’s funny is that after all this, I can’t see how SQ4 is supposed to support Both mode, and ScummVM only needs to add that third button state. There is no patch to adjust the script, and I can’t for the life of me figure out how this would work:

(method (say theVoodoo whoCares &tmp newEvent)
  ; ...
  (if (& gMessageType TEXT_MSG)
    ; Note: noun, tVerb, and tSequence are properties. theVoodoo is now "case".
    (self startText: modNum noun tVerb theVoodoo tSequence &rest)
  )
  (if (& gMessageType CD_MSG)
    (self startAudio: theVoodoo)
  )
  ; ..
)

The weird part is that I can’t find anywhere those properties are set.

…At least with the KQ6/LB2 patches they actually do overhaul quite a bit of the scripts’ logic, which are otherwise just the same system scripts as above. Not the way I did it for my game, but clearly in a way that works out.