mirror of
https://github.com/libretro/scummvm.git
synced 2025-02-23 20:51:14 +00:00

The Mac GUI code is getting unwieldy, and this is an attempt at breaking it up into more manageable chunks.
2329 lines
67 KiB
C++
2329 lines
67 KiB
C++
/* ScummVM - Graphic Adventure Engine
|
|
*
|
|
* ScummVM is the legal property of its developers, whose names
|
|
* are too numerous to list here. Please refer to the COPYRIGHT
|
|
* file distributed with this source distribution.
|
|
*
|
|
* This program is free software: you can redistribute it and/or modify
|
|
* it under the terms of the GNU General Public License as published by
|
|
* the Free Software Foundation, either version 3 of the License, or
|
|
* (at your option) any later version.
|
|
*
|
|
* This program is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
*
|
|
*/
|
|
|
|
|
|
|
|
#include "common/config-manager.h"
|
|
#include "common/unicode-bidi.h"
|
|
#include "audio/mixer.h"
|
|
|
|
#include "scumm/actor.h"
|
|
#include "scumm/charset.h"
|
|
#include "scumm/dialogs.h"
|
|
#include "scumm/file.h"
|
|
#include "scumm/imuse_digi/dimuse_engine.h"
|
|
#ifdef ENABLE_HE
|
|
#include "scumm/he/intern_he.h"
|
|
#include "scumm/he/localizer.h"
|
|
#endif
|
|
#include "scumm/macgui/macgui.h"
|
|
#include "scumm/resource.h"
|
|
#include "scumm/scumm.h"
|
|
#include "scumm/scumm_v2.h"
|
|
#include "scumm/scumm_v6.h"
|
|
#include "scumm/scumm_v7.h"
|
|
#include "scumm/verbs.h"
|
|
#include "scumm/he/sound_he.h"
|
|
|
|
#include "scumm/ks_check.h"
|
|
|
|
namespace Scumm {
|
|
|
|
|
|
|
|
#pragma mark -
|
|
#pragma mark --- "High level" message code ---
|
|
#pragma mark -
|
|
|
|
|
|
void ScummEngine::printString(int m, const byte *msg) {
|
|
switch (m) {
|
|
case 0:
|
|
// WORKAROUND bug #12734: The script tries to clear the currently
|
|
// displayed message after Rapp gives you the map, but that means
|
|
// you'll never see Guybrush's reaction to finding a map piece.
|
|
//
|
|
// It's a bit hard to pin down the exact case, since it happens
|
|
// at a few different points during the script. We limit it to
|
|
// when the player has the map piece.
|
|
//
|
|
// We have to do it here, because we don't want to delay the
|
|
// animation of Rapp turning back to Ashes.
|
|
if (_game.id == GID_MONKEY2 && _roomResource == 19 &&
|
|
vm.slot[_currentScript].number == 203 &&
|
|
_actorToPrintStrFor == 255 && strcmp((const char *)msg, " ") == 0 &&
|
|
getOwner(200) == VAR(VAR_EGO) && VAR(VAR_HAVE_MSG) &&
|
|
enhancementEnabled(kEnhMinorBugFixes)) {
|
|
return;
|
|
}
|
|
|
|
// WORKAROUND bug #13378: In the German CD version, Sam's
|
|
// reactions to Max beating up the scientist run much too quick
|
|
// for the animation to match. We get around this by slowing
|
|
// down that animation.
|
|
//
|
|
// In the italian CD version, the whole scene is sped up to
|
|
// keep up with Sam's speech. We compensate for this by slowing
|
|
// down the other animations.
|
|
if (_game.id == GID_SAMNMAX && vm.slot[_currentScript].number == 65 && enhancementEnabled(kEnhTimingChanges)) {
|
|
Actor *a;
|
|
|
|
if (_language == Common::DE_DEU && strcmp(_game.variant, "Floppy") != 0) {
|
|
if (memcmp(msg + 16, "Ohh!", 4) == 0) {
|
|
a = derefActorSafe(2, "printString");
|
|
if (a)
|
|
a->setAnimSpeed(3);
|
|
}
|
|
} else if (_language == Common::IT_ITA && strcmp(_game.variant, "Floppy") != 0) {
|
|
if (memcmp(msg + 16, "Ooh.", 4) == 0) {
|
|
a = derefActorSafe(3, "printString");
|
|
if (a)
|
|
a->setAnimSpeed(2);
|
|
a = derefActorSafe(10, "printString");
|
|
if (a)
|
|
a->setAnimSpeed(2);
|
|
}
|
|
}
|
|
}
|
|
|
|
actorTalk(msg);
|
|
break;
|
|
case 1:
|
|
drawString(1, msg);
|
|
break;
|
|
case 2:
|
|
debugMessage(msg);
|
|
break;
|
|
case 3:
|
|
showMessageDialog(msg);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
|
|
void ScummEngine::debugMessage(const byte *msg) {
|
|
byte buffer[500];
|
|
convertMessageToString(msg, buffer, sizeof(buffer));
|
|
|
|
if ((buffer[0] != 0xFF) && _debugMode) {
|
|
debug(0, "DEBUG: %s", buffer);
|
|
return;
|
|
}
|
|
|
|
if (buffer[0] == 0xFF && buffer[1] == 10) {
|
|
uint32 offset, length;
|
|
int channel = 0;
|
|
|
|
offset = buffer[2] | (buffer[3] << 8) | (buffer[6] << 16) | (buffer[7] << 24);
|
|
length = buffer[10] | (buffer[11] << 8) | (buffer[14] << 16) | (buffer[15] << 24);
|
|
|
|
// Sam and Max uses a caching system, printing empty messages
|
|
// and setting VAR_V6_SOUNDMODE beforehand. See patch #8051.
|
|
if (_game.id == GID_SAMNMAX)
|
|
channel = VAR(VAR_V6_SOUNDMODE);
|
|
|
|
if (channel != 2)
|
|
_sound->talkSound(offset, length, DIGI_SND_MODE_SFX, channel);
|
|
}
|
|
}
|
|
|
|
void ScummEngine::showMessageDialog(const byte *msg) {
|
|
// Original COMI used different code at this point.
|
|
// Seemed to use blastText for the messages
|
|
byte buf[500];
|
|
|
|
convertMessageToString(msg, buf, sizeof(buf));
|
|
|
|
if (_string[3].color == 0)
|
|
_string[3].color = 4;
|
|
|
|
if (isUsingOriginalGUI()) {
|
|
if (_game.version > 4)
|
|
VAR(VAR_KEYPRESS) = showBannerAndPause(0, -1, (const char *)msg).ascii;
|
|
else
|
|
VAR(VAR_KEYPRESS) = showOldStyleBannerAndPause((const char *)msg, _string[3].color, -1).ascii;
|
|
} else {
|
|
InfoDialog dialog(this, Common::U32String((char *)buf));
|
|
VAR(VAR_KEYPRESS) = runDialog(dialog);
|
|
}
|
|
|
|
}
|
|
|
|
#pragma mark -
|
|
#pragma mark --- Core message/subtitle code ---
|
|
#pragma mark -
|
|
|
|
|
|
bool ScummEngine::handleNextCharsetCode(Actor *a, int *code) {
|
|
uint32 digiTalkieOffset = 0;
|
|
uint32 digiTalkieLength = 0;
|
|
int color, frme, c = 0, oldy;
|
|
bool endLoop = false;
|
|
byte *buffer = _charsetBuffer + _charsetBufPos;
|
|
while (!endLoop) {
|
|
c = *buffer++;
|
|
if (!(c == 0xFF || (_game.version <= 6 && c == 0xFE))) {
|
|
break;
|
|
}
|
|
c = *buffer++;
|
|
|
|
if (_newLineCharacter != 0 && c == _newLineCharacter) {
|
|
c = 13;
|
|
break;
|
|
}
|
|
|
|
switch (c) {
|
|
case 1:
|
|
c = 13; // new line
|
|
_msgCount = _screenWidth;
|
|
endLoop = true;
|
|
break;
|
|
case 2:
|
|
_haveMsg = 0;
|
|
_keepText = true;
|
|
endLoop = true;
|
|
break;
|
|
case 3:
|
|
_haveMsg = _game.version == 7 && !(_game.id == GID_FT && _game.features & GF_DEMO) ? 1 : 0xFF;
|
|
_keepText = false;
|
|
_msgCount = 0;
|
|
endLoop = true;
|
|
|
|
// WORKAROUND bug #13378: Some of the speech is badly
|
|
// synced to the subtitles, particularly in the
|
|
// localized versions. This happens because a single
|
|
// speech line is used for a text that's broken up by
|
|
// one or more embedded "wait" codes. Rather than
|
|
// relying on the calculated talk delay, hard-code
|
|
// better ones.
|
|
if (_game.id == GID_SAMNMAX && enhancementEnabled(kEnhTimingChanges) && isScriptRunning(65)) {
|
|
typedef struct {
|
|
const char *str;
|
|
const int16 talkDelay;
|
|
const byte action;
|
|
} TimingAdjustment;
|
|
|
|
TimingAdjustment *adjustments;
|
|
int numAdjustments;
|
|
|
|
// We identify the broken up strings that need
|
|
// adjustment by the upcoming text.
|
|
|
|
TimingAdjustment timingAdjustmentsEN[] = {
|
|
{ "It's just that", 100, 0 },
|
|
{ "you're TOO nice", 90, 0 },
|
|
{ "^unpredictable.", 170, 0 },
|
|
{ "Yikes!", 120, 0 },
|
|
{ "Huh?", 90, 0 },
|
|
{ "Why do you", 110, 0 },
|
|
{ "Maybe we can", 75, 0 },
|
|
{ "Mind if I drive?", 160, 0 }
|
|
};
|
|
|
|
TimingAdjustment timingAdjustmentsDEFloppy[] = {
|
|
{ "Und daf\x81r^", 110, 0 },
|
|
{ "Es ist blo\xe1^", 120, 0 },
|
|
{ "Hey.", 50, 0 },
|
|
{ "Klasse Schlag!", 30, 0 },
|
|
{ "Uiii!", 80, 0 },
|
|
{ "H\x84h?", 60, 0 },
|
|
{ "Kann ich seine", 110, 0 },
|
|
{ "Warum, glaubst", 110, 0 },
|
|
{ "La\xe1 uns von", 220, 0 },
|
|
{ "Vielleicht", 90, 0 },
|
|
{ "Kann ich fahren?", 220, 0 }
|
|
};
|
|
|
|
TimingAdjustment timingAdjustmentsDECD[] = {
|
|
{ "Und daf\x81r^", 110, 0 },
|
|
{ "Es ist blo\xe1^", 120, 0 },
|
|
{ "Hey.", 130, 0 },
|
|
{ "Klasse Schlag!", 150, 0 },
|
|
{ "Uiii!", 185, 1 },
|
|
{ "H\x84h?", 150, 0 },
|
|
{ "Kann ich seine", 110, 0 },
|
|
{ "Warum, glaubst", 110, 0 },
|
|
{ "Vielleicht", 90, 0 },
|
|
{ "Kann ich fahren?", 240, 0 }
|
|
};
|
|
|
|
TimingAdjustment timingAdjustmentsITFloppy[] = {
|
|
{ "E per questo^", 140, 0 },
|
|
{ "E' che^ecco^", 100, 0 },
|
|
{ "^imprevedibile.", 170, 0 },
|
|
{ "Huh?", 110, 0 },
|
|
{ "Perch\x82 pensi", 90, 0 },
|
|
{ "Andiamocene da", 230, 0 },
|
|
{ "Forse possiamo", 75, 0 },
|
|
{ "Ti dispiace", 160, 0 }
|
|
};
|
|
|
|
TimingAdjustment timingAdjustmentsITCD[] = {
|
|
{ "E per questo^", 120, 0 },
|
|
{ "Forse sei", 75, 0 },
|
|
{ "^imprevedibile.", 170, 0 },
|
|
{ "Oh.", 20, 0 },
|
|
{ "Ehi, bel colpo.", 30, 0 },
|
|
{ "Yikes!", 90, 0 },
|
|
{ "Huh?", 50, 0 },
|
|
{ "Posso tenere", 100, 0 },
|
|
{ "Perch\x82 pensi", 120, 0 },
|
|
{ "Andiamocene", 250, 0 },
|
|
{ "Forse possiamo", 90, 0 },
|
|
{ "Ti dispiace", 200, 0 }
|
|
};
|
|
|
|
TimingAdjustment timingAdjustmentsFRFloppy[] = {
|
|
{ "Et pour me", 120, 0 },
|
|
{ "C'est que^euh^", 100, 0 },
|
|
{ "vous \x88tes", 65, 0 },
|
|
{ "^impr\x82visible", 170, 0 },
|
|
{ "Pourquoi est-ce", 100, 0 },
|
|
{ "Filons de cet", 190, 0 },
|
|
{ "Nous pourrons", 65, 0 },
|
|
{ "Je peux conduire", 170, 0 },
|
|
{ "Je n'oublierai", 90, 0 }
|
|
};
|
|
|
|
TimingAdjustment timingAdjustmentsFRCD[] = {
|
|
{ "Oh.", 85, 0 },
|
|
{ "H\x82, pas mal.", 80, 0 },
|
|
{ "Yiik!", 110, 0 },
|
|
{ "Je peux garder", 130, 0 },
|
|
{ "Pourquoi est-ce", 120, 0 },
|
|
{ "Nous pourrons", 80, 0 },
|
|
{ "Je peux conduire", 220, 0 }
|
|
};
|
|
|
|
TimingAdjustment timingAdjustmentsES[] = {
|
|
{ "Y por eso^", 130, 0 },
|
|
{ "es simplemente", 100, 0 },
|
|
{ "eres DEMASIADO", 90, 0 },
|
|
{ "\xa8Hug?", 110, 0 },
|
|
{ "\xa8Por qu\x82", 110, 0 },
|
|
{ "Tal vez podamos", 75, 0 },
|
|
{ "\xa8Te importa", 160, 0 }
|
|
};
|
|
|
|
switch (_language) {
|
|
case Common::EN_ANY:
|
|
adjustments = timingAdjustmentsEN;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsEN);
|
|
break;
|
|
case Common::DE_DEU:
|
|
if (strcmp(_game.variant, "Floppy") == 0) {
|
|
adjustments = timingAdjustmentsDEFloppy;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsDEFloppy);
|
|
} else {
|
|
adjustments = timingAdjustmentsDECD;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsDECD);
|
|
}
|
|
break;
|
|
case Common::IT_ITA:
|
|
if (strcmp(_game.variant, "Floppy") == 0) {
|
|
adjustments = timingAdjustmentsITFloppy;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsITFloppy);
|
|
} else {
|
|
adjustments = timingAdjustmentsITCD;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsITCD);
|
|
}
|
|
break;
|
|
case Common::FR_FRA:
|
|
if (strcmp(_game.variant, "Floppy") == 0) {
|
|
adjustments = timingAdjustmentsFRFloppy;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsFRFloppy);
|
|
} else {
|
|
adjustments = timingAdjustmentsFRCD;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsFRCD);
|
|
}
|
|
break;
|
|
case Common::ES_ESP:
|
|
adjustments = timingAdjustmentsES;
|
|
numAdjustments = ARRAYSIZE(timingAdjustmentsES);
|
|
break;
|
|
default:
|
|
adjustments = nullptr;
|
|
numAdjustments = 0;
|
|
break;
|
|
}
|
|
|
|
byte action = 0;
|
|
|
|
for (int i = 0; i < numAdjustments; i++) {
|
|
int len = strlen(adjustments[i].str);
|
|
if (memcmp(buffer, adjustments[i].str, len) == 0) {
|
|
_talkDelay = adjustments[i].talkDelay;
|
|
action = adjustments[i].action;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (_language == Common::DE_DEU) {
|
|
Actor *act;
|
|
|
|
switch (action) {
|
|
case 1:
|
|
act = derefActorSafe(2, "handleNextCharsetCode");
|
|
if (act)
|
|
act->setAnimSpeed(2);
|
|
|
|
// The actor speaks so slowly that the background
|
|
// animations have run their course. Try to restart
|
|
// them, even though it won't be quite seamless.
|
|
|
|
int actors[] = { 3, 10 };
|
|
|
|
for (int i = 0; i < ARRAYSIZE(actors); i++) {
|
|
act = derefActorSafe(actors[i], "handleNextCharsetCode");
|
|
if (act) {
|
|
act->startAnimActor(act->_initFrame);
|
|
act->animateActor(249);
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
break;
|
|
case 8:
|
|
// Ignore this code here. Occurs e.g. in MI2 when you
|
|
// talk to the carpenter on scabb island. It works like
|
|
// code 1 (=newline) in verb texts, but is ignored in
|
|
// spoken text (i.e. here). Used for very long verb
|
|
// sentences.
|
|
break;
|
|
case 9:
|
|
frme = buffer[0] | (buffer[1] << 8);
|
|
buffer += 2;
|
|
if (a)
|
|
a->startAnimActor(frme);
|
|
break;
|
|
case 10:
|
|
// Note the similarity to the code in debugMessage()
|
|
digiTalkieOffset = buffer[0] | (buffer[1] << 8) | (buffer[4] << 16) | (buffer[5] << 24);
|
|
digiTalkieLength = buffer[8] | (buffer[9] << 8) | (buffer[12] << 16) | (buffer[13] << 24);
|
|
buffer += 14;
|
|
if (_game.heversion >= 60) {
|
|
#ifdef ENABLE_HE
|
|
((SoundHE *)_sound)->playVoice(_localizer ? _localizer->mapTalk(digiTalkieOffset) : digiTalkieOffset, digiTalkieLength);
|
|
#else
|
|
((SoundHE *)_sound)->playVoice(digiTalkieOffset, digiTalkieLength);
|
|
#endif
|
|
} else {
|
|
_sound->talkSound(digiTalkieOffset, digiTalkieLength, DIGI_SND_MODE_TALKIE);
|
|
}
|
|
_haveActorSpeechMsg = false;
|
|
break;
|
|
case 12:
|
|
color = buffer[0] | (buffer[1] << 8);
|
|
buffer += 2;
|
|
if (color == 0xFF)
|
|
_charset->setColor(_charsetColor);
|
|
else
|
|
_charset->setColor(color);
|
|
break;
|
|
case 13:
|
|
debug(0, "handleNextCharsetCode: Unknown opcode 13 %d", READ_LE_UINT16(buffer));
|
|
buffer += 2;
|
|
break;
|
|
case 14:
|
|
oldy = _charset->getFontHeight();
|
|
_charset->setCurID(*buffer++);
|
|
buffer += 2;
|
|
memcpy(_charsetColorMap, _charsetData[_charset->getCurID()], 4);
|
|
_nextTop -= _charset->getFontHeight() - oldy;
|
|
break;
|
|
default:
|
|
error("handleNextCharsetCode: invalid code %d", c);
|
|
}
|
|
}
|
|
_charsetBufPos = buffer - _charsetBuffer;
|
|
*code = c;
|
|
return (c != 2 && c != 3);
|
|
}
|
|
|
|
#ifdef ENABLE_HE
|
|
bool ScummEngine_v72he::handleNextCharsetCode(Actor *a, int *code) {
|
|
const int charsetCode = (_game.heversion >= 80) ? 127 : 64;
|
|
uint32 digiTalkieOffset = 0;
|
|
uint32 digiTalkieLength = 0;
|
|
int i, c = 0;
|
|
char value[4096];
|
|
bool endLoop = false;
|
|
bool endText = false;
|
|
byte *buffer = _charsetBuffer + _charsetBufPos;
|
|
while (!endLoop) {
|
|
c = *buffer++;
|
|
if (c != charsetCode) {
|
|
break;
|
|
}
|
|
c = *buffer++;
|
|
switch (c) {
|
|
case 84:
|
|
i = 0;
|
|
c = *buffer++;
|
|
while (c != 44) {
|
|
value[i] = c;
|
|
c = *buffer++;
|
|
i++;
|
|
}
|
|
value[i] = 0;
|
|
digiTalkieOffset = atoi(value);
|
|
i = 0;
|
|
c = *buffer++;
|
|
while (c != charsetCode) {
|
|
value[i] = c;
|
|
c = *buffer++;
|
|
i++;
|
|
}
|
|
value[i] = 0;
|
|
digiTalkieLength = atoi(value);
|
|
((SoundHE *)_sound)->playVoice(_localizer ? _localizer->mapTalk(digiTalkieOffset) : digiTalkieOffset, digiTalkieLength);
|
|
break;
|
|
case 104:
|
|
_haveMsg = 0;
|
|
_keepText = true;
|
|
endLoop = endText = true;
|
|
break;
|
|
case 110:
|
|
c = 13; // new line
|
|
endLoop = true;
|
|
break;
|
|
case 116:
|
|
i = 0;
|
|
memset(value, 0, sizeof(value));
|
|
c = *buffer++;
|
|
while (c != charsetCode) {
|
|
value[i] = c;
|
|
c = *buffer++;
|
|
i++;
|
|
}
|
|
value[i] = 0;
|
|
((SoundHE *)_sound)->playVoiceFile(value);
|
|
break;
|
|
case 119:
|
|
_haveMsg = 0xFF;
|
|
_keepText = false;
|
|
endLoop = endText = true;
|
|
break;
|
|
default:
|
|
error("handleNextCharsetCode: invalid code %d", c);
|
|
}
|
|
}
|
|
_charsetBufPos = buffer - _charsetBuffer;
|
|
*code = c;
|
|
return (endText == 0);
|
|
}
|
|
#endif
|
|
|
|
bool ScummEngine::newLine() {
|
|
_nextLeft = _string[0].xpos;
|
|
if (_charset->_center) {
|
|
int stringWidth = _charset->getStringWidth(0, _charsetBuffer + _charsetBufPos);
|
|
_nextLeft -= stringWidth / 2;
|
|
|
|
if (_nextLeft < 0)
|
|
// The commented out part of the next line was meant as a fix for Kanji text glitches in DIG.
|
|
// But these glitches couldn't be reproduced in recent tests. So the underlying issue might
|
|
// have been taken care of in a different manner. And the fix actually caused other text glitches
|
|
// (FT/German, if you look at the sign on the container at game start). After counterchecking
|
|
// the original code it seems that setting _nextLeft to 0 is the right thing to do here.
|
|
_nextLeft = /*_game.version >= 6 ? _string[0].xpos :*/ 0;
|
|
|
|
// See CHARSET_1() for more context about the following Sega CD code.
|
|
if (_game.platform == Common::kPlatformSegaCD) {
|
|
// Clip 16 pixels away from the right
|
|
if (_nextLeft + stringWidth > (_screenWidth - 16)) {
|
|
_nextLeft -= (_nextLeft + stringWidth) - (_screenWidth - 16);
|
|
}
|
|
|
|
// Clip 16 pixels away from the left
|
|
if (_nextLeft < 16) {
|
|
_nextLeft = 16;
|
|
}
|
|
}
|
|
} else if (_isRTL) {
|
|
if (_game.id == GID_MANIAC || _game.heversion >= 72 || ((_game.id == GID_MONKEY || _game.id == GID_MONKEY2) && _charset->getCurID() == 4)) {
|
|
_nextLeft = _screenWidth - _charset->getStringWidth(0, _charsetBuffer + _charsetBufPos) - _nextLeft;
|
|
} else if (_game.id == GID_MONKEY2 && _charset->getCurID() == 5) {
|
|
_nextLeft += _screenWidth - 210 - _charset->getStringWidth(0, _charsetBuffer + _charsetBufPos);
|
|
}
|
|
}
|
|
if (_game.version == 0) {
|
|
return false;
|
|
} else if (!(_game.platform == Common::kPlatformFMTowns) && _string[0].height) {
|
|
_nextTop += _string[0].height;
|
|
} else {
|
|
if ((_game.platform == Common::kPlatformSegaCD || _isIndy4Jap) && _useCJKMode) {
|
|
_nextTop += _charset->getFontHeight();
|
|
} else {
|
|
bool useCJK = _useCJKMode;
|
|
// SCUMM5 FM-Towns doesn't use the height of the ROM font here.
|
|
if (_game.platform == Common::kPlatformFMTowns && _game.version == 5)
|
|
_useCJKMode = false;
|
|
_nextTop += _charset->getFontHeight();
|
|
_useCJKMode = useCJK;
|
|
}
|
|
}
|
|
if (_game.version > 3) {
|
|
// FIXME: is this really needed?
|
|
_charset->_disableOffsX = true;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
#ifdef ENABLE_HE
|
|
void ScummEngine_v72he::fakeBidiString(byte *ltext, bool ignoreVerb, int ltextSize) const {
|
|
if (*ltext == 0x7F) {
|
|
ltext++;
|
|
while (*(ltext++) != 0x7F);
|
|
}
|
|
byte *loc = ltext;
|
|
byte tmp = 0;
|
|
while (1) {
|
|
while (*loc && *loc != 13) {
|
|
loc++;
|
|
}
|
|
tmp = *loc;
|
|
*loc = 0;
|
|
Common::strcpy_s(reinterpret_cast<char *>(ltext), ltextSize, Common::convertBiDiString((const char *)ltext, Common::kWindows1255).c_str());
|
|
*loc = tmp;
|
|
loc++;
|
|
ltext = loc;
|
|
if (!tmp) {
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
#endif
|
|
|
|
void ScummEngine::fakeBidiString(byte *ltext, bool ignoreVerb, int ltextSize) const {
|
|
// Provides custom made BiDi mechanism.
|
|
// Reverses texts on each line marked by control characters (considering different control characters used in verbs panel)
|
|
// While preserving original order of numbers (also negative numbers and comma separated)
|
|
int32 ll = 0;
|
|
if ((_game.id == GID_INDY4 && ltext[ll] == 0x7F) || (_game.id == GID_MONKEY2 && ltext[ll] == 0x07)) {
|
|
ll++;
|
|
}
|
|
while (ltext[ll] == 0xFF) {
|
|
ll += 4;
|
|
}
|
|
|
|
if (_game.id == GID_MONKEY2 && ltext[0] == 0x07) {
|
|
for (int i = 1; i < ll; i++)
|
|
ltext[i - 1] = ltext[i];
|
|
ltext[--ll] = 0x07;
|
|
}
|
|
|
|
int32 ipos = 0;
|
|
int32 start = 0;
|
|
byte *text = ltext + ll;
|
|
byte *current = text;
|
|
|
|
int32 bufferSize = 384;
|
|
byte * const buff = (byte *)calloc(sizeof(byte), bufferSize);
|
|
assert(buff);
|
|
byte * const stack = (byte *)calloc(sizeof(byte), bufferSize);
|
|
assert(stack);
|
|
|
|
while (1) {
|
|
if (*current == 0x0D || *current == 0 || *current == 0xFF || *current == 0xFE) {
|
|
|
|
// ignore the line break for verbs texts
|
|
if (ignoreVerb && *current && (*(current + 1) == 8)) {
|
|
*(current + 1) = *current;
|
|
*current = 0x08;
|
|
ipos += 2;
|
|
current += 2;
|
|
continue;
|
|
}
|
|
|
|
memset(buff, 0, bufferSize);
|
|
memset(stack, 0, bufferSize);
|
|
|
|
// Reverse string on current line (between start and ipos).
|
|
int32 sthead = 0;
|
|
byte last = 0;
|
|
for (int32 j = 0; j < ipos; j++) {
|
|
byte *curr = text + start + ipos - j - 1;
|
|
// Special cases to preserve original ordering (numbers).
|
|
if (Common::isDigit(*curr) ||
|
|
(*curr == (byte)',' && j + 1 < ipos && Common::isDigit(*(curr - 1)) && Common::isDigit(last)) ||
|
|
(*curr == (byte)'-' && (j + 1 == ipos || Common::isSpace(*(curr - 1))) && Common::isDigit(last))) {
|
|
++sthead;
|
|
stack[sthead] = *curr;
|
|
} else {
|
|
while (sthead > 0) {
|
|
buff[j - sthead] = stack[sthead];
|
|
--sthead;
|
|
}
|
|
buff[j] = *curr;
|
|
}
|
|
last = *curr;
|
|
}
|
|
while (sthead > 0) {
|
|
buff[ipos - sthead] = stack[sthead];
|
|
--sthead;
|
|
}
|
|
memcpy(text + start, buff, ipos);
|
|
start += ipos + 1;
|
|
ipos = -1;
|
|
if (*current == 0xFF || *current == 0xFE) {
|
|
current++;
|
|
if (*current == 0x03 || *current == 0x02) {
|
|
break;
|
|
}
|
|
if (*current == 0x0A || *current == 0x0C) {
|
|
start += 2;
|
|
current += 2;
|
|
}
|
|
start++;
|
|
ipos++;
|
|
current++;
|
|
continue;
|
|
}
|
|
}
|
|
if (*current) {
|
|
ipos++;
|
|
current++;
|
|
continue;
|
|
}
|
|
break;
|
|
}
|
|
if (!ignoreVerb) {
|
|
if (_game.id == GID_INDY4 && ltext[0] == 0x7F) {
|
|
ltext[start + ipos + ll] = 0x80;
|
|
ltext[start + ipos + ll + 1] = 0;
|
|
} else if (_game.id == GID_MONKEY2) {
|
|
// add non-printable character to end to avoid space trimming
|
|
ltext[start + ipos + ll] = '@';
|
|
ltext[start + ipos + ll + 1] = 0;
|
|
}
|
|
}
|
|
|
|
free(buff);
|
|
free(stack);
|
|
}
|
|
|
|
void ScummEngine_v2::drawSentence() {
|
|
Common::Rect sentenceline;
|
|
const byte *temp;
|
|
int slot = getVerbSlot(VAR(VAR_SENTENCE_VERB), 0);
|
|
|
|
if (!((_userState & USERSTATE_IFACE_SENTENCE) ||
|
|
(_game.platform == Common::kPlatformNES && (_userState & USERSTATE_IFACE_ALL))))
|
|
return;
|
|
|
|
if (getResourceAddress(rtVerb, slot))
|
|
_sentenceBuf = (char *)getResourceAddress(rtVerb, slot);
|
|
else
|
|
return;
|
|
|
|
if (VAR(VAR_SENTENCE_OBJECT1) > 0) {
|
|
temp = getObjOrActorName(VAR(VAR_SENTENCE_OBJECT1));
|
|
if (temp) {
|
|
_sentenceBuf += " ";
|
|
_sentenceBuf += (const char *)temp;
|
|
}
|
|
|
|
// For V1 games, the engine must compute the preposition.
|
|
// In all other Scumm versions, this is done by the sentence script.
|
|
if ((_game.id == GID_MANIAC && _game.version == 1 && !(_game.platform == Common::kPlatformNES)) && (VAR(VAR_SENTENCE_PREPOSITION) == 0)) {
|
|
if (_verbs[slot].prep == 0xFF) {
|
|
byte *ptr = getOBCDFromObject(VAR(VAR_SENTENCE_OBJECT1));
|
|
assert(ptr);
|
|
VAR(VAR_SENTENCE_PREPOSITION) = (*(ptr + 12) >> 5);
|
|
} else
|
|
VAR(VAR_SENTENCE_PREPOSITION) = _verbs[slot].prep;
|
|
}
|
|
}
|
|
|
|
if (0 < VAR(VAR_SENTENCE_PREPOSITION) && VAR(VAR_SENTENCE_PREPOSITION) <= 4) {
|
|
drawPreposition(VAR(VAR_SENTENCE_PREPOSITION));
|
|
}
|
|
|
|
if (VAR(VAR_SENTENCE_OBJECT2) > 0) {
|
|
temp = getObjOrActorName(VAR(VAR_SENTENCE_OBJECT2));
|
|
if (temp) {
|
|
_sentenceBuf += " ";
|
|
_sentenceBuf += (const char *)temp;
|
|
}
|
|
}
|
|
|
|
_string[2].charset = 1;
|
|
_string[2].ypos = _virtscr[kVerbVirtScreen].topline;
|
|
_string[2].xpos = 0;
|
|
_string[2].right = _virtscr[kVerbVirtScreen].w - 1;
|
|
if (_game.platform == Common::kPlatformNES) {
|
|
_string[2].xpos = 16;
|
|
_string[2].color = 0;
|
|
} else if (_game.platform == Common::kPlatformC64) {
|
|
_string[2].color = 16;
|
|
} else {
|
|
_string[2].color = 13;
|
|
}
|
|
|
|
byte string[80];
|
|
const char *ptr = _sentenceBuf.c_str();
|
|
int i = 0, len = 0;
|
|
|
|
// Maximum length of printable characters
|
|
int maxChars = (_game.platform == Common::kPlatformNES) ? 60 : 40;
|
|
while (*ptr) {
|
|
if (*ptr != '@')
|
|
len++;
|
|
if (len > maxChars) {
|
|
break;
|
|
}
|
|
|
|
string[i++] = *ptr++;
|
|
|
|
if (_game.platform == Common::kPlatformNES && len == 30) {
|
|
string[i++] = 0xFF;
|
|
string[i++] = 8;
|
|
}
|
|
}
|
|
string[i] = 0;
|
|
|
|
if (_game.platform == Common::kPlatformNES) {
|
|
sentenceline.top = _virtscr[kVerbVirtScreen].topline;
|
|
sentenceline.bottom = _virtscr[kVerbVirtScreen].topline + 16;
|
|
sentenceline.left = 16;
|
|
sentenceline.right = _virtscr[kVerbVirtScreen].w - 1;
|
|
} else {
|
|
sentenceline.top = _virtscr[kVerbVirtScreen].topline;
|
|
sentenceline.bottom = _virtscr[kVerbVirtScreen].topline + 8;
|
|
sentenceline.left = 0;
|
|
sentenceline.right = _virtscr[kVerbVirtScreen].w - 1;
|
|
}
|
|
restoreBackground(sentenceline);
|
|
|
|
drawString(2, (byte *)string);
|
|
}
|
|
|
|
void ScummEngine::CHARSET_1() {
|
|
Actor *a;
|
|
if (_game.heversion >= 60 && _haveMsg == 3) {
|
|
stopTalk();
|
|
return;
|
|
}
|
|
|
|
if (!_haveMsg)
|
|
return;
|
|
|
|
_force2ByteCharHeight = false;
|
|
|
|
if (_game.version >= 4 && _game.version <= 6) {
|
|
// Do nothing while the camera is moving
|
|
if ((camera._dest.x / 8) != (camera._cur.x / 8) || camera._cur.x != camera._last.x)
|
|
return;
|
|
}
|
|
|
|
a = nullptr;
|
|
if (getTalkingActor() != 0xFF)
|
|
a = derefActorSafe(getTalkingActor(), "CHARSET_1");
|
|
|
|
if (a && _string[0].overhead) {
|
|
int s;
|
|
|
|
_string[0].xpos = a->getPos().x - _virtscr[kMainVirtScreen].xstart;
|
|
_string[0].ypos = a->getPos().y - a->getElevation() - _screenTop - _screenDrawOffset;
|
|
|
|
if (_game.version <= 5) {
|
|
if (VAR(VAR_V5_TALK_STRING_Y) < 0) {
|
|
if (_game.version == 4) {
|
|
_string[0].ypos = (int)VAR(VAR_V5_TALK_STRING_Y) + a->getPos().y + a->getElevation();
|
|
} else {
|
|
s = (a->_scaley * (int)VAR(VAR_V5_TALK_STRING_Y)) / 0xFF;
|
|
_string[0].ypos += (int)(((VAR(VAR_V5_TALK_STRING_Y) - s) / 2) + s);
|
|
}
|
|
} else {
|
|
_string[0].ypos = (int)VAR(VAR_V5_TALK_STRING_Y);
|
|
}
|
|
|
|
} else {
|
|
s = a->_scalex * a->_talkPosX / 0xFF;
|
|
_string[0].xpos += ((a->_talkPosX - s) / 2) + s;
|
|
|
|
s = a->_scaley * a->_talkPosY / 0xFF;
|
|
_string[0].ypos += ((a->_talkPosY - s) / 2) + s;
|
|
|
|
if (_string[0].ypos > _screenHeight - 40)
|
|
_string[0].ypos = _screenHeight - 40;
|
|
}
|
|
|
|
if (_string[0].ypos < 1)
|
|
_string[0].ypos = 1;
|
|
|
|
if (_string[0].xpos < 80)
|
|
_string[0].xpos = 80;
|
|
if (_string[0].xpos > _screenWidth - 80)
|
|
_string[0].xpos = _screenWidth - 80;
|
|
}
|
|
|
|
_charset->_top = _string[0].ypos + _screenTop + _screenDrawOffset;
|
|
_charset->_startLeft = _charset->_left = _string[0].xpos;
|
|
_charset->_right = _string[0].right;
|
|
_charset->_center = _string[0].center;
|
|
_charset->setColor(_charsetColor);
|
|
|
|
if (a && a->_charset)
|
|
_charset->setCurID(a->_charset);
|
|
else
|
|
_charset->setCurID(_string[0].charset);
|
|
|
|
if (_game.version >= 5)
|
|
memcpy(_charsetColorMap, _charsetData[_charset->getCurID()], 4);
|
|
|
|
#ifndef DISABLE_TOWNS_DUAL_LAYER_MODE
|
|
if (_game.platform == Common::kPlatformFMTowns && (_keepText || _haveMsg == 0xFF))
|
|
memcpy(&_charset->_str, &_curStringRect, sizeof(Common::Rect));
|
|
#endif
|
|
|
|
if (_talkDelay)
|
|
return;
|
|
|
|
if ((_game.version <= 6 && _haveMsg == 1) ||
|
|
(_game.version == 7 && _haveMsg != 1)) {
|
|
|
|
if (_game.heversion >= 60) {
|
|
if (_sound->isSoundRunning(HSND_TALKIE_SLOT) == 0)
|
|
stopTalk();
|
|
} else {
|
|
if ((_sound->_digiSndMode & DIGI_SND_MODE_TALKIE) == 0)
|
|
stopTalk();
|
|
}
|
|
return;
|
|
}
|
|
|
|
// The second check is from LOOM DOS EGA disasm. It prevents weird speech animations
|
|
// with empty strings (bug #990). The same code is present in actorTalk(). The FM-Towns
|
|
// versions don't have such code, but I do not get the weird speech animations either.
|
|
// So apparently it is not needed there.
|
|
if (a && !_string[0].no_talk_anim && !(_game.id == GID_LOOM && _game.platform != Common::kPlatformFMTowns && !_charsetBuffer[_charsetBufPos])) {
|
|
a->runActorTalkScript(a->_talkStartFrame);
|
|
_useTalkAnims = true;
|
|
}
|
|
|
|
_talkDelay = (VAR_DEFAULT_TALK_DELAY != 0xFF) ? VAR(VAR_DEFAULT_TALK_DELAY) : 60;
|
|
|
|
if (!_keepText) {
|
|
#ifndef DISABLE_TOWNS_DUAL_LAYER_MODE
|
|
if (_game.platform == Common::kPlatformFMTowns)
|
|
towns_restoreCharsetBg();
|
|
else
|
|
#endif
|
|
restoreCharsetBg();
|
|
_msgCount = 0;
|
|
} else if (_game.version <= 2) {
|
|
_talkDelay += _msgCount * _defaultTextSpeed;
|
|
}
|
|
|
|
if (_game.version > 3) {
|
|
int maxWidth = _charset->_right - _string[0].xpos;
|
|
|
|
maxWidth -= _game.platform == Common::kPlatformSegaCD ? 16 : 1;
|
|
|
|
if (_charset->_center) {
|
|
if (maxWidth > _nextLeft)
|
|
maxWidth = _nextLeft;
|
|
maxWidth *= 2;
|
|
}
|
|
|
|
// If the string is centered and this is MI1 Sega CD, don't add linebreaks right away;
|
|
// we will take care of it in a different way just below ... :-)
|
|
if (_game.platform != Common::kPlatformSegaCD ||
|
|
(_game.platform != Common::kPlatformSegaCD && !_charset->_center)) {
|
|
_charset->addLinebreaks(0, _charsetBuffer + _charsetBufPos, 0, maxWidth);
|
|
}
|
|
}
|
|
|
|
if (_charset->_center) {
|
|
// MI1 Sega CD appears to be doing its own thing in here when
|
|
// the string x coordinate is on the right side of the screen:
|
|
// - Applies some tentative line breaks;
|
|
// - Go line by line and check if the string still overflows
|
|
// on the last 16 pixels of the right side of the screen;
|
|
// - If so, take the original string and apply a stricter final wrapping;
|
|
// - Finally, clip the string final position to 16 pixels from the right
|
|
// and from the left sides of the screen.
|
|
//
|
|
// I agree that this is very convoluted :-) , but it's the only way
|
|
// to display pixel accurate text on both ENG and JAP editions of this
|
|
// version.
|
|
|
|
if (_game.platform == Common::kPlatformSegaCD) {
|
|
int predictionMaxWidth = _charset->_right - _string[0].xpos;
|
|
int predictionNextLeft = _nextLeft;
|
|
|
|
bool useStricterWrapping = (_string[0].xpos > _screenWidth / 2);
|
|
|
|
// Predict if a stricter wrapping is going to be necessary
|
|
if (!useStricterWrapping) {
|
|
if (predictionMaxWidth > predictionNextLeft)
|
|
predictionMaxWidth = predictionNextLeft;
|
|
predictionMaxWidth *= 2;
|
|
|
|
byte predictionString[512];
|
|
|
|
memcpy(predictionString, _charsetBuffer, sizeof(predictionString));
|
|
|
|
// Impose a tentative max string width for the wrapping
|
|
_charset->addLinebreaks(0, predictionString + _charsetBufPos, 0, predictionMaxWidth);
|
|
|
|
int predictionStringWidth = _charset->getStringWidth(0, predictionString + _charsetBufPos);
|
|
predictionNextLeft -= predictionStringWidth / 2;
|
|
|
|
if (predictionNextLeft < 16)
|
|
predictionNextLeft = 16;
|
|
|
|
byte *ptrToCurLine = predictionString + _charsetBufPos;
|
|
byte curChar = *ptrToCurLine;
|
|
|
|
// Go line by line and check if the string overflows
|
|
// on the last 16 pixels on the right side of the screen...
|
|
while (curChar) {
|
|
predictionStringWidth = _charset->getStringWidth(0, ptrToCurLine);
|
|
predictionNextLeft -= predictionStringWidth / 2;
|
|
|
|
if (predictionNextLeft < 16)
|
|
predictionNextLeft = 16;
|
|
|
|
useStricterWrapping |= (predictionNextLeft + predictionStringWidth > (_screenWidth - 16));
|
|
|
|
if (useStricterWrapping)
|
|
break;
|
|
|
|
// Advance to next line, if any...
|
|
do {
|
|
// Control code handling...
|
|
if (curChar == 0xFE || curChar == 0xFF) {
|
|
// Advance to the control code and
|
|
// check if it's a new line instruction...
|
|
ptrToCurLine++;
|
|
curChar = *ptrToCurLine;
|
|
|
|
// Gotcha!
|
|
if (curChar == 1 || (_newLineCharacter && curChar == _newLineCharacter)) {
|
|
ptrToCurLine++;
|
|
curChar = *ptrToCurLine;
|
|
break;
|
|
}
|
|
|
|
// If we're here, we don't need this control code,
|
|
// let's just skip it...
|
|
ptrToCurLine++;
|
|
} else if (_useCJKMode && curChar & 0x80) { // CJK char
|
|
ptrToCurLine++;
|
|
}
|
|
|
|
// Line breaks and string termination
|
|
if (curChar == '\r' || curChar == '\n') {
|
|
ptrToCurLine++;
|
|
curChar = *ptrToCurLine;
|
|
break;
|
|
} else if (curChar == '\0') {
|
|
curChar = *ptrToCurLine;
|
|
break;
|
|
}
|
|
|
|
ptrToCurLine++;
|
|
curChar = *ptrToCurLine;
|
|
} while (true);
|
|
}
|
|
}
|
|
|
|
|
|
// Impose the final line breaks with the correct max string width;
|
|
// this part is practically the default v5 text centering code...
|
|
int finalMaxWidth = _charset->_right - _string[0].xpos;
|
|
finalMaxWidth -= useStricterWrapping ? 16 : 0;
|
|
|
|
if (finalMaxWidth > _nextLeft)
|
|
finalMaxWidth = _nextLeft;
|
|
finalMaxWidth *= 2;
|
|
|
|
_charset->addLinebreaks(0, _charsetBuffer + _charsetBufPos, 0, finalMaxWidth);
|
|
|
|
int finalStringWidth = _charset->getStringWidth(0, _charsetBuffer + _charsetBufPos);
|
|
_nextLeft -= finalStringWidth / 2;
|
|
|
|
// Final additional clippings (these will also be repeated on newLine()):
|
|
|
|
// Clip 16 pixels away from the right
|
|
if (_nextLeft + finalStringWidth > (_screenWidth - 16)) {
|
|
_nextLeft -= (_nextLeft + finalStringWidth) - (_screenWidth - 16);
|
|
}
|
|
|
|
// Clip 16 pixels away from the left
|
|
if (_nextLeft < 16) {
|
|
_nextLeft = 16;
|
|
}
|
|
} else {
|
|
int stringWidth = _charset->getStringWidth(0, _charsetBuffer + _charsetBufPos);
|
|
_nextLeft -= stringWidth / 2;
|
|
|
|
if (_nextLeft < 0)
|
|
// The commented out part of the next line was meant as a fix for Kanji text glitches in DIG.
|
|
// But these glitches couldn't be reproduced in recent tests. So the underlying issue might
|
|
// have been taken care of in a different manner. And the fix actually caused other text glitches
|
|
// (FT/German, if you look at the sign on the container at game start). After counterchecking
|
|
// the original code it seems that setting _nextLeft to 0 is the right thing to do here.
|
|
_nextLeft = /*_game.version >= 6 ? _string[0].xpos :*/ 0;
|
|
}
|
|
|
|
} else if (_isRTL) {
|
|
if (_game.id == GID_MANIAC || _game.heversion >= 72 || ((_game.id == GID_MONKEY || _game.id == GID_MONKEY2) && _charset->getCurID() == 4)) {
|
|
_nextLeft = _screenWidth - _charset->getStringWidth(0, _charsetBuffer + _charsetBufPos) - _nextLeft;
|
|
} else if (_game.id == GID_MONKEY2 && _charset->getCurID() == 5) {
|
|
_nextLeft += _screenWidth - 210 - _charset->getStringWidth(0, _charsetBuffer + _charsetBufPos);
|
|
}
|
|
}
|
|
|
|
_charset->_disableOffsX = _charset->_firstChar = !_keepText;
|
|
|
|
int c = 0;
|
|
|
|
if (_isRTL)
|
|
fakeBidiString(_charsetBuffer + _charsetBufPos, true, sizeof(_charsetBuffer) - _charsetBufPos);
|
|
|
|
bool createTextBox = (_macGui && _game.id == GID_INDY3);
|
|
bool drawTextBox = false;
|
|
|
|
while (handleNextCharsetCode(a, &c)) {
|
|
if (c == 0) {
|
|
// End of text reached, set _haveMsg accordingly
|
|
_haveMsg = 1;
|
|
_keepText = false;
|
|
_msgCount = 0;
|
|
break;
|
|
}
|
|
|
|
if (c == 13) {
|
|
if (!newLine())
|
|
break;
|
|
continue;
|
|
}
|
|
|
|
// Handle line overflow for V3. See also bug #2213.
|
|
if (_game.version == 3 && _nextLeft >= _screenWidth) {
|
|
_nextLeft = _screenWidth;
|
|
}
|
|
// Handle line breaks for V1-V2
|
|
if (_game.version <= 2 && _nextLeft >= _screenWidth) {
|
|
if (!newLine())
|
|
break; // FIXME: Is this necessary? Only would be relevant for v0 games
|
|
}
|
|
|
|
_charset->_left = _nextLeft;
|
|
_charset->_top = _nextTop;
|
|
|
|
if (createTextBox) {
|
|
if (!_keepText)
|
|
_macGui->initTextAreaForActor(a, _charset->getColor());
|
|
createTextBox = false;
|
|
drawTextBox = true;
|
|
}
|
|
|
|
if (c & 0x80 && _useCJKMode) {
|
|
if (is2ByteCharacter(_language, c)) {
|
|
byte *buffer = _charsetBuffer + _charsetBufPos;
|
|
c += *buffer++ * 256; //LE
|
|
_charsetBufPos = buffer - _charsetBuffer;
|
|
|
|
// The JAP Sega CD version of Monkey Island 1 doesn't just calculate
|
|
// the font height, but instead relies on the actual string height.
|
|
// If the string contains at least a 2 byte character, then we signal it with
|
|
// a flag, so that getFontHeight() can yield the correct result.
|
|
// It has been verified on the disasm for Indy4 Japanese DOS/V and Mac that
|
|
// this is the correct behavior for the latter game as well.
|
|
// Monkey Island 1 seems to have an exception for the 0xFAFD character, which
|
|
// is a space character with a two byte character height and width, but those
|
|
// dimensions apparently are never used, and the 0x20 character is used instead.
|
|
if (_game.platform != Common::kPlatformSegaCD || c != 0xFAFD) {
|
|
_force2ByteCharHeight = true;
|
|
}
|
|
}
|
|
}
|
|
if (_game.version <= 3) {
|
|
_charset->printChar(c, false);
|
|
_msgCount += 1;
|
|
} else {
|
|
if (_game.features & GF_16BIT_COLOR) {
|
|
// HE games which use sprites for subtitles
|
|
} else if (_game.heversion >= 60 && !ConfMan.getBool("subtitles") && _sound->isSoundInUse(HSND_TALKIE_SLOT)) {
|
|
// Special case for HE games
|
|
} else if (_game.id == GID_LOOM && !ConfMan.getBool("subtitles") && (_sound->pollCD())) {
|
|
// Special case for Loom (CD), since it only uses CD audio.for sound
|
|
} else if (!ConfMan.getBool("subtitles") && (!_haveActorSpeechMsg || _mixer->isSoundHandleActive(*_sound->_talkChannelHandle))) {
|
|
// Subtitles are turned off, and there is a voice version
|
|
// of this message -> don't print it.
|
|
} else {
|
|
_charset->printChar(c, false);
|
|
}
|
|
}
|
|
_nextLeft = _charset->_left;
|
|
_nextTop = _charset->_top;
|
|
|
|
if (_game.version <= 2) {
|
|
_talkDelay += _defaultTextSpeed;
|
|
VAR(VAR_CHARCOUNT)++;
|
|
} else {
|
|
_talkDelay += (int)VAR(VAR_CHARINC);
|
|
}
|
|
}
|
|
|
|
if (drawTextBox)
|
|
mac_drawIndy3TextBox();
|
|
|
|
#ifndef DISABLE_TOWNS_DUAL_LAYER_MODE
|
|
if (_game.platform == Common::kPlatformFMTowns && (c == 0 || c == 2 || c == 3))
|
|
memcpy(&_curStringRect, &_charset->_str, sizeof(Common::Rect));
|
|
#endif
|
|
}
|
|
|
|
void ScummEngine::drawString(int a, const byte *msg) {
|
|
byte buf[270];
|
|
byte *space;
|
|
int i, c;
|
|
byte fontHeight = 0;
|
|
uint color;
|
|
int code = (_game.heversion >= 80) ? 127 : 64;
|
|
bool isV3Towns = _game.version == 3 && _game.platform == Common::kPlatformFMTowns;
|
|
|
|
// drawString is not used in SCUMM v7 and v8
|
|
assert(_game.version < 7);
|
|
|
|
convertMessageToString(msg, buf, sizeof(buf));
|
|
|
|
if (_isRTL)
|
|
fakeBidiString(buf, false, sizeof(buf));
|
|
|
|
_charset->_top = _string[a].ypos + _screenTop + _screenDrawOffset;
|
|
_charset->_startLeft = _charset->_left = _string[a].xpos;
|
|
_charset->_right = _string[a].right;
|
|
_charset->_center = _string[a].center;
|
|
_charset->setColor(_string[a].color);
|
|
_charset->_disableOffsX = _charset->_firstChar = true;
|
|
_charset->setCurID(_string[a].charset);
|
|
|
|
if (_game.version >= 5)
|
|
memcpy(_charsetColorMap, _charsetData[_charset->getCurID()], 4);
|
|
|
|
fontHeight = _charset->getFontHeight();
|
|
|
|
// Disabled in HE games starting from Freddi1 because
|
|
// of issues when writing a savegame name containing spaces...
|
|
if (_game.version >= 4 && _game.heversion < 70) {
|
|
// trim from the right
|
|
byte *tmp = buf;
|
|
space = nullptr;
|
|
while (*tmp) {
|
|
if (*tmp == ' ') {
|
|
if (!space)
|
|
space = tmp;
|
|
} else {
|
|
space = nullptr;
|
|
}
|
|
tmp++;
|
|
}
|
|
if (space)
|
|
*space = '\0';
|
|
}
|
|
|
|
if (_charset->_center) {
|
|
_charset->_left -= _charset->getStringWidth(a, buf) / 2;
|
|
} else if (_isRTL && _game.id != GID_SAMNMAX && _game.id != GID_MANIAC) {
|
|
// Ignore INDY4 verbs (but allow dialogue)
|
|
if (_game.id != GID_INDY4 || buf[0] == 127) {
|
|
if (_game.id == GID_INDY4)
|
|
buf[0] = 32;
|
|
_charset->_left = _screenWidth - _charset->_startLeft - _charset->getStringWidth(1, buf);
|
|
}
|
|
}
|
|
|
|
if (!buf[0]) {
|
|
if (_game.version >= 5) {
|
|
buf[0] = ' ';
|
|
buf[1] = 0;
|
|
} else {
|
|
_charset->_str.left = _charset->_left;
|
|
_charset->_str.top = _charset->_top;
|
|
_charset->_str.right = _charset->_left;
|
|
_charset->_str.bottom = _charset->_top;
|
|
}
|
|
}
|
|
|
|
for (i = 0; (c = buf[i++]) != 0;) {
|
|
if (_game.heversion >= 72 && c == code) {
|
|
c = buf[i++];
|
|
switch (c) {
|
|
case 110:
|
|
if (_charset->_center) {
|
|
_charset->_left = _charset->_startLeft - _charset->getStringWidth(a, buf + i);
|
|
} else {
|
|
_charset->_left = _charset->_startLeft;
|
|
}
|
|
_charset->_top += fontHeight;
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
} else if ((c == 0xFF || (_game.version <= 6 && c == 0xFE)) && (_game.heversion <= 71)) {
|
|
c = buf[i++];
|
|
switch (c) {
|
|
case 9:
|
|
case 10:
|
|
case 13:
|
|
case 14:
|
|
i += 2;
|
|
break;
|
|
case 1:
|
|
case 8:
|
|
if (_charset->_center) {
|
|
_charset->_left = _charset->_startLeft - _charset->getStringWidth(a, buf + i);
|
|
} else if (_isRTL) {
|
|
_charset->_left = _screenWidth - _charset->_startLeft - _charset->getStringWidth(1, buf + i);
|
|
} else {
|
|
_charset->_left = _charset->_startLeft;
|
|
}
|
|
if (!(_game.platform == Common::kPlatformFMTowns) && _string[0].height) {
|
|
_nextTop += _string[0].height;
|
|
} else {
|
|
_charset->_top += fontHeight;
|
|
}
|
|
break;
|
|
case 12:
|
|
color = buf[i] + (buf[i + 1] << 8);
|
|
i += 2;
|
|
if (color == 0xFF)
|
|
_charset->setColor(_string[a].color);
|
|
else
|
|
_charset->setColor(color);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
} else {
|
|
if (a == 1 && _game.version >= 6) {
|
|
// FIXME: The following code is a bit nasty. It is used for the
|
|
// Highway surfing game in Sam&Max; there, _blitAlso is set to
|
|
// true when writing the highscore numbers. It is also in DOTT
|
|
// for parts the intro and for drawing newspaper headlines. It
|
|
// is also used for scores in bowling mini game in fbear and
|
|
// for names in load/save screen of all HE games. Maybe it is
|
|
// also being used in other places.
|
|
//
|
|
// A better name for _blitAlso might be _imprintOnBackground
|
|
|
|
if (_string[a].no_talk_anim == false) {
|
|
//debug(0, "Would have set _charset->_blitAlso = true (wanted to print '%c' = %d)", c, c);
|
|
_charset->_blitAlso = true;
|
|
}
|
|
}
|
|
if (c & 0x80 && _useCJKMode) {
|
|
if (is2ByteCharacter(_language, c))
|
|
c += buf[i++] * 256;
|
|
}
|
|
|
|
// With the code above, we risk missing the termination character.
|
|
// This happens at least for the restart prompt message on INDY3 Towns JAP.
|
|
if (isV3Towns && i > 1 && buf[i - 1] == 0)
|
|
break;
|
|
|
|
_charset->printChar(c, true);
|
|
_charset->_blitAlso = false;
|
|
}
|
|
}
|
|
|
|
if (a == 0) {
|
|
_nextLeft = _charset->_left;
|
|
_nextTop = _charset->_top;
|
|
}
|
|
|
|
// From disasm: this is used to let a yellow bar appear
|
|
// in the bottom of the screen during dialog choices which
|
|
// are longer than the screen width.
|
|
if (_isIndy4Jap)
|
|
_scummVars[78] = _charset->_left;
|
|
|
|
_string[a].xpos = _charset->_str.right;
|
|
|
|
if (_game.heversion >= 60) {
|
|
_string[a]._default.xpos = _string[a].xpos;
|
|
_string[a]._default.ypos = _string[a].ypos;
|
|
}
|
|
}
|
|
|
|
int ScummEngine::convertMessageToString(const byte *msg, byte *dst, int dstSize) {
|
|
uint num = 0;
|
|
uint32 val;
|
|
byte chr;
|
|
byte lastChr = 0;
|
|
const byte *src;
|
|
byte *start, *end;
|
|
byte transBuf[2048];
|
|
|
|
assert(dst);
|
|
start = dst;
|
|
end = dst + dstSize;
|
|
|
|
if (msg == nullptr) {
|
|
debug(0, "Bad message in convertMessageToString, ignoring");
|
|
return 0;
|
|
}
|
|
|
|
if (_game.version >= 7 || isScummvmKorTarget()) {
|
|
translateText(msg, transBuf, sizeof(transBuf));
|
|
src = transBuf;
|
|
} else {
|
|
src = msg;
|
|
}
|
|
|
|
num = 0;
|
|
|
|
while (1) {
|
|
chr = src[num++];
|
|
if (chr == 0)
|
|
break;
|
|
|
|
if (_game.id == GID_LOOM && _game.platform == Common::kPlatformPCEngine) {
|
|
// Code for TM character
|
|
if (chr == 0x0F && src[num] == 0x20) {
|
|
*dst++ = 0x5D;
|
|
*dst++ = 0x5E;
|
|
continue;
|
|
// Code for (C) character
|
|
} else if (chr == 0x1C && src[num] == 0x20) {
|
|
*dst++ = 0x3E;
|
|
*dst++ = 0x2A;
|
|
continue;
|
|
// Code for " character
|
|
} else if (chr == 0x19) {
|
|
*dst++ = 0x2F;
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if (chr == 0xFF) {
|
|
chr = src[num++];
|
|
|
|
// WORKAROUND for bug #1675, a script bug in Indy3. Apparently,
|
|
// a german 'sz' was encoded incorrectly as 0xFF2E. We replace
|
|
// this by the correct encoding here. See also ScummEngine::resStrLen().
|
|
if (_game.id == GID_INDY3 && chr == 0x2E) {
|
|
*dst++ = 0xE1;
|
|
continue;
|
|
}
|
|
|
|
// WORKAROUND for bug #2715: Yet another script bug in Indy3.
|
|
// Once more a german 'sz' was encoded incorrectly, but this time
|
|
// they simply encoded it as 0xFF instead of 0xE1. Happens twice
|
|
// in script 71.
|
|
if (_game.id == GID_INDY3 && chr == 0x20 && vm.slot[_currentScript].number == 71) {
|
|
num--;
|
|
*dst++ = 0xE1;
|
|
continue;
|
|
}
|
|
|
|
if (chr == 1 || chr == 2 || chr == 3 || chr == 8) {
|
|
// Simply copy these special codes
|
|
*dst++ = 0xFF;
|
|
*dst++ = chr;
|
|
} else {
|
|
val = (_game.version == 8) ? READ_LE_UINT32(src + num) : READ_LE_UINT16(src + num);
|
|
switch (chr) {
|
|
case 4:
|
|
dst += convertIntMessage(dst, end - dst, val);
|
|
break;
|
|
case 5:
|
|
dst += convertVerbMessage(dst, end - dst, val);
|
|
break;
|
|
case 6:
|
|
dst += convertNameMessage(dst, end - dst, val);
|
|
break;
|
|
case 7:
|
|
dst += convertStringMessage(dst, end - dst, val);
|
|
break;
|
|
case 9:
|
|
case 10:
|
|
case 12:
|
|
case 13:
|
|
case 14:
|
|
// Simply copy these special codes
|
|
*dst++ = 0xFF;
|
|
*dst++ = chr;
|
|
*dst++ = src[num+0];
|
|
*dst++ = src[num+1];
|
|
|
|
// WORKAROUND: In some versions of Sam & Max, talking to Evelyn Morrison
|
|
// while wearing the bigfoot costume misattributes some lines to Max even
|
|
// though Sam is the one actually speaking. For example, the French and
|
|
// German releases use `startAnim(8)` while the English release correctly
|
|
// uses `startAnim(7)` for this.
|
|
if (_game.id == GID_SAMNMAX && _currentRoom == 52 && vm.slot[_currentScript].number == 102 &&
|
|
chr == 9 && readVar(0x8000 + 95) != 0 && (VAR(171) == 997 || VAR(171) == 998) &&
|
|
dst[-2] == 8 && enhancementEnabled(kEnhMinorBugFixes)) {
|
|
dst[-2] = 7;
|
|
}
|
|
|
|
if (_game.version == 8) {
|
|
*dst++ = src[num+2];
|
|
*dst++ = src[num+3];
|
|
}
|
|
break;
|
|
default:
|
|
error("convertMessageToString(): string escape sequence %d unknown", chr);
|
|
}
|
|
num += (_game.version == 8) ? 4 : 2;
|
|
}
|
|
} else {
|
|
if ((chr != '@') || (_game.version >= 7 && is2ByteCharacter(_language, lastChr)) ||
|
|
(_language == Common::JA_JPN && checkSJISCode(lastChr))) {
|
|
*dst++ = chr;
|
|
}
|
|
lastChr = chr;
|
|
}
|
|
|
|
// Check for a buffer overflow
|
|
if (dst >= end)
|
|
error("convertMessageToString: buffer overflow");
|
|
}
|
|
|
|
// WORKAROUND bug #12249 (occurs also in original): Missing actor animation in German versions of SAMNMAX
|
|
// Adding the missing startAnim(14) animation escape sequence while copying the text fixes it.
|
|
if (_game.id == GID_SAMNMAX && _currentRoom == 56 && vm.slot[_currentScript].number == 200 &&
|
|
_language == Common::DE_DEU && enhancementEnabled(kEnhMinorBugFixes)) {
|
|
// 0xE5E6 is the CD version, 0xE373 is for the floppy version
|
|
if (vm.slot[_currentScript].offs == 0xE5E6 || vm.slot[_currentScript].offs == 0xE373) {
|
|
*dst++ = 0xFF;
|
|
*dst++ = 0x09;
|
|
*dst++ = 0x0E;
|
|
*dst++ = 0x00;
|
|
}
|
|
}
|
|
|
|
// WORKAROUND: Russian The Dig pads messages with 03. No idea why
|
|
// it does not work as is with our rendering code, thus fixing it
|
|
// with a workaround.
|
|
if (_game.id == GID_DIG) {
|
|
while (dst > start && *(dst - 1) == 0x03)
|
|
dst--;
|
|
}
|
|
*dst = 0;
|
|
|
|
return dstSize - (end - dst);
|
|
}
|
|
|
|
#ifdef ENABLE_HE
|
|
int ScummEngine_v72he::convertMessageToString(const byte *msg, byte *dst, int dstSize) {
|
|
uint num = 0;
|
|
byte chr;
|
|
const byte *src;
|
|
byte *end;
|
|
|
|
assert(dst);
|
|
end = dst + dstSize;
|
|
|
|
if (msg == NULL) {
|
|
debug(0, "Bad message in convertMessageToString, ignoring");
|
|
return 0;
|
|
}
|
|
|
|
src = msg;
|
|
num = 0;
|
|
|
|
while (1) {
|
|
chr = src[num++];
|
|
if (_game.heversion >= 80 && src[num - 1] == '(' && (src[num] == 'p' || src[num] == 'P')) {
|
|
// Filter out the following prefixes in subtitles
|
|
// (pickup4)
|
|
// (PU1)
|
|
// (PU2)
|
|
while (src[num++] != ')')
|
|
;
|
|
continue;
|
|
}
|
|
if ((_game.features & GF_HE_LOCALIZED) && chr == '[') {
|
|
while (src[num++] != ']')
|
|
;
|
|
continue;
|
|
}
|
|
|
|
if (chr == 0)
|
|
break;
|
|
|
|
*dst++ = chr;
|
|
|
|
// Check for a buffer overflow
|
|
if (dst >= end)
|
|
error("convertMessageToString: buffer overflow");
|
|
}
|
|
*dst = 0;
|
|
|
|
return dstSize - (end - dst);
|
|
}
|
|
#endif
|
|
|
|
int ScummEngine::convertIntMessage(byte *dst, int dstSize, int var) {
|
|
int num;
|
|
|
|
num = readVar(var);
|
|
return snprintf((char *)dst, dstSize, "%d", num);
|
|
}
|
|
|
|
int ScummEngine::convertVerbMessage(byte *dst, int dstSize, int var) {
|
|
int num, k;
|
|
|
|
bool isKorVerbGlue = false;
|
|
|
|
if (isScummvmKorTarget() && _useCJKMode && var & (1 << 15)) {
|
|
isKorVerbGlue = true;
|
|
var &= ~(1 << 15);
|
|
}
|
|
|
|
num = readVar(var);
|
|
if (num) {
|
|
for (k = 1; k < _numVerbs; k++) {
|
|
// Fix ZAK FM-TOWNS bug #1734 by emulating exact (inconsistant?) behavior of the original code
|
|
if (num == _verbs[k].verbid && !_verbs[k].type && (!_verbs[k].saveid || (_game.version == 3 && _game.platform == Common::kPlatformFMTowns))) {
|
|
// Process variation of Korean postpositions
|
|
// Used by Korean fan translated games (monkey1, monkey2)
|
|
if (isKorVerbGlue) {
|
|
static const byte code0380[4] = {0xFF, 0x07, 0x03, 0x80}; // "eul/reul"
|
|
static const byte code0480[4] = {0xFF, 0x07, 0x04, 0x80}; // "wa/gwa"
|
|
static const byte codeWalkTo[9] = {0xFF, 0x07, 0x03, 0x80, 0x20, 0xC7, 0xE2, 0xC7, 0xD8}; // "eul/reul hyang-hae"
|
|
byte _transText[10];
|
|
memset(_transText, 0, sizeof(_transText));
|
|
// WORKAROUND: MI Korean verb parser
|
|
if (_game.id == GID_MONKEY_VGA) {
|
|
switch (_verbs[k].verbid) {
|
|
case 1: // Open
|
|
case 2: // Close
|
|
case 3: // Give
|
|
case 4: // Turn on
|
|
case 5: // Turn off
|
|
case 6: // Push
|
|
case 7: // Pull
|
|
case 8: // Use
|
|
case 9: // Look at
|
|
case 10: // Walk to
|
|
memcpy(_transText, codeWalkTo, 9);
|
|
break;
|
|
case 11: // Pick up
|
|
memcpy(_transText, code0380, 4);
|
|
break;
|
|
case 13: // Talk to
|
|
memcpy(_transText, code0480, 4);
|
|
break;
|
|
}
|
|
return convertMessageToString(_transText, dst, dstSize);
|
|
}
|
|
// WORKAROUND: MI2 Korean verb parser
|
|
if (_game.id == GID_MONKEY2) {
|
|
if (_verbs[k].verbid <= 11) {
|
|
switch (_verbs[k].verbid) {
|
|
case 2: // Open
|
|
case 3: // Close
|
|
case 4: // Give
|
|
case 5: // Push
|
|
case 6: // Pull
|
|
case 7: // Use
|
|
case 8: // Look at
|
|
case 9: // Pick up
|
|
memcpy(_transText, code0380, 4);
|
|
break;
|
|
case 10: // Talk to
|
|
memcpy(_transText, code0480, 4);
|
|
break;
|
|
case 11: // Walk to
|
|
memcpy(_transText, codeWalkTo, 9);
|
|
break;
|
|
}
|
|
return convertMessageToString(_transText, dst, dstSize);
|
|
}
|
|
}
|
|
} else {
|
|
const byte *ptr = getResourceAddress(rtVerb, k);
|
|
return convertMessageToString(ptr, dst, dstSize);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
int ScummEngine::convertNameMessage(byte *dst, int dstSize, int var) {
|
|
int num;
|
|
|
|
num = readVar(var);
|
|
if (num) {
|
|
const byte *ptr = getObjOrActorName(num);
|
|
if (ptr) {
|
|
int increment = convertMessageToString(ptr, dst, dstSize);
|
|
// Save the final consonant (jongsung) of the last Korean character
|
|
// Used by Korean fan translated games (monkey1, monkey2)
|
|
if (isScummvmKorTarget() && _useCJKMode) {
|
|
_krStrPost = 0;
|
|
int len = resStrLen(ptr);
|
|
if (len >= 2) {
|
|
for (int i = len; i > 1; i--) {
|
|
byte k1 = ptr[i - 2];
|
|
byte k2 = ptr[i - 1];
|
|
if (checkKSCode(k1, k2)) {
|
|
int jongsung = checkJongsung(k1, k2);
|
|
if (jongsung)
|
|
_krStrPost |= 1;
|
|
if (jongsung == 8) // 'ri-eul' final consonant
|
|
_krStrPost |= (1 << 1);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return increment;
|
|
}
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
int ScummEngine::convertStringMessage(byte *dst, int dstSize, int var) {
|
|
const byte *ptr;
|
|
|
|
if (_game.version <= 2) {
|
|
byte chr;
|
|
int i = 0;
|
|
while ((chr = (byte)_scummVars[var++])) {
|
|
if (chr != '@') {
|
|
*dst++ = chr;
|
|
i++;
|
|
}
|
|
}
|
|
|
|
return i;
|
|
}
|
|
|
|
if (_game.version == 3 || (_game.version >= 6 && _game.heversion < 72))
|
|
var = readVar(var);
|
|
|
|
// Process variation of Korean postpositions
|
|
// Used by Korean fan translated games (monkey1, monkey2)
|
|
if (isScummvmKorTarget() && _useCJKMode && (var & (1 << 15))) {
|
|
int idx;
|
|
static const byte codeIdx[] = {0x00, 0x00, 0xC0, 0xB8, 0x00, 0x00, 0xC0, 0xCC, 0xB0, 0xA1, 0xC0, 0xCC, 0xB8, 0xA6, 0xC0, 0xBB, 0xBF, 0xCD, 0xB0, 0xFA, 0xB4, 0xC2, 0xC0, 0xBA};
|
|
|
|
if ((var & ~(1 << 15)) == 0)
|
|
idx = (2 * (var & ~(1 << 15)) + (_krStrPost & 1) - bool(_krStrPost & 2)) * 2;
|
|
else
|
|
idx = (2 * (var & ~(1 << 15)) + (_krStrPost & 1)) * 2;
|
|
|
|
byte _transText[3];
|
|
const byte byteIdx[] = {codeIdx[idx], codeIdx[idx + 1]};
|
|
|
|
memset(_transText, 0, sizeof(_transText));
|
|
memcpy(_transText, byteIdx, 2);
|
|
|
|
return convertMessageToString(_transText, dst, dstSize);
|
|
} else if (var) {
|
|
ptr = getStringAddress(var);
|
|
if (ptr) {
|
|
int increment = convertMessageToString(ptr, dst, dstSize);
|
|
// Save the final consonant (jongsung) of the last Korean character
|
|
// Used by Korean fan translated games (monkey1, monkey2)
|
|
if (isScummvmKorTarget() && _useCJKMode) {
|
|
_krStrPost = 0;
|
|
for (int i = resStrLen(ptr); i > 1; i--) {
|
|
byte k1 = ptr[i - 2];
|
|
byte k2 = ptr[i - 1];
|
|
if (checkKSCode(k1, k2)) {
|
|
int jongsung = checkJongsung(k1, k2);
|
|
if (jongsung)
|
|
_krStrPost |= 1;
|
|
if (jongsung == 8) // 'ri-eul' final consonant
|
|
_krStrPost |= (1 << 1);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
return increment;
|
|
}
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
|
|
bool ScummEngine::hasLocalizer() {
|
|
return _localizer != nullptr;
|
|
}
|
|
|
|
#pragma mark -
|
|
#pragma mark --- Charset initialisation ---
|
|
#pragma mark -
|
|
|
|
|
|
#ifdef ENABLE_HE
|
|
void ScummEngine_v80he::initCharset(int charsetno) {
|
|
ScummEngine::initCharset(charsetno);
|
|
VAR(VAR_CURRENT_CHARSET) = charsetno;
|
|
}
|
|
#endif
|
|
|
|
void ScummEngine::initCharset(int charsetno) {
|
|
if (_game.id == GID_FT) {
|
|
if (!_res->isResourceLoaded(rtCharset, charsetno))
|
|
loadCharset(charsetno);
|
|
} else {
|
|
if (!getResourceAddress(rtCharset, charsetno))
|
|
loadCharset(charsetno);
|
|
}
|
|
|
|
_string[0]._default.charset = charsetno;
|
|
_string[1]._default.charset = charsetno;
|
|
|
|
memcpy(_charsetColorMap, _charsetData[charsetno], sizeof(_charsetColorMap));
|
|
}
|
|
|
|
|
|
#pragma mark -
|
|
#pragma mark --- Translation/localization code ---
|
|
#pragma mark -
|
|
|
|
|
|
#ifdef ENABLE_SCUMM_7_8
|
|
static int indexCompare(const void *p1, const void *p2) {
|
|
const ScummEngine_v7::LangIndexNode *i1 = (const ScummEngine_v7::LangIndexNode *) p1;
|
|
const ScummEngine_v7::LangIndexNode *i2 = (const ScummEngine_v7::LangIndexNode *) p2;
|
|
|
|
return strcmp(i1->tag, i2->tag);
|
|
}
|
|
|
|
// Create an index of the language file.
|
|
void ScummEngine_v7::loadLanguageBundle() {
|
|
if (isScummvmKorTarget()) {
|
|
// Support language bundle for FT
|
|
ScummEngine::loadLanguageBundle();
|
|
return;
|
|
}
|
|
ScummFile file(this);
|
|
int32 size;
|
|
|
|
if (_game.id == GID_DIG) {
|
|
openFile(file, "language.bnd");
|
|
} else if (_game.id == GID_CMI) {
|
|
openFile(file, "language.tab");
|
|
} else {
|
|
return;
|
|
}
|
|
if (file.isOpen() == false) {
|
|
_existLanguageFile = false;
|
|
return;
|
|
}
|
|
|
|
_existLanguageFile = true;
|
|
|
|
size = file.size();
|
|
_languageBuffer = (char *)calloc(1, size+1);
|
|
file.read(_languageBuffer, size);
|
|
file.close();
|
|
|
|
int32 i;
|
|
char *ptr = _languageBuffer;
|
|
|
|
// Count the number of lines in the language file.
|
|
for (_languageIndexSize = 0; ; _languageIndexSize++) {
|
|
ptr = strpbrk(ptr, "\n\r");
|
|
if (ptr == NULL)
|
|
break;
|
|
while (*ptr == '\n' || *ptr == '\r')
|
|
ptr++;
|
|
}
|
|
|
|
// Fill the language file index. This is just an array of
|
|
// tags and offsets. I did consider using a balanced tree
|
|
// instead, but the extra overhead in the node structure would
|
|
// easily have doubled the memory consumption of the index.
|
|
// And anyway, using qsort + bsearch gives us the exact same
|
|
// O(log(n)) access time anyway ;-).
|
|
|
|
_languageIndex = (LangIndexNode *)calloc(_languageIndexSize, sizeof(LangIndexNode));
|
|
|
|
ptr = _languageBuffer;
|
|
|
|
if (_game.id == GID_DIG) {
|
|
int lineCount = _languageIndexSize;
|
|
const char *baseTag = "";
|
|
byte enc = 0; // Initially assume the language file is not encoded
|
|
|
|
// We'll determine the real index size as we go.
|
|
_languageIndexSize = 0;
|
|
for (i = 0; i < lineCount; i++) {
|
|
if (*ptr == '!') {
|
|
// Don't know what a line with '!' means, just ignore it
|
|
} else if (*ptr == 'h') {
|
|
// File contains Korean text (Hangul). just ignore it
|
|
} else if (*ptr == 'j') {
|
|
// File contains Japanese text. just ignore it
|
|
} else if (*ptr == 'c') {
|
|
// File contains Chinese text. just ignore it
|
|
} else if (*ptr == 'e') {
|
|
// File is encoded!
|
|
enc = 0x13;
|
|
} else if (*ptr == '@') {
|
|
// A new 'base tag'
|
|
baseTag = ptr + 1;
|
|
} else if (*ptr == '#') {
|
|
// Number of subtags following a given basetag. We don't need that
|
|
// information so we just skip it
|
|
} else if (Common::isDigit(*ptr)) {
|
|
int idx = 0;
|
|
// A number (up to three digits)...
|
|
while (Common::isDigit(*ptr)) {
|
|
idx = idx * 10 + (*ptr - '0');
|
|
ptr++;
|
|
}
|
|
|
|
// ...followed by a slash...
|
|
assert(*ptr == '/');
|
|
ptr++;
|
|
|
|
// ...and then the translated message, possibly encoded
|
|
_languageIndex[_languageIndexSize].offset = ptr - _languageBuffer;
|
|
|
|
// Decode string if necessary.
|
|
if (enc) {
|
|
while (*ptr != '\n' && *ptr != '\r')
|
|
*ptr++ ^= enc;
|
|
}
|
|
|
|
// The tag is the basetag, followed by a dot and then the index
|
|
Common::sprintf_s(_languageIndex[_languageIndexSize].tag, "%s.%03d", baseTag, idx);
|
|
|
|
// That was another index entry
|
|
_languageIndexSize++;
|
|
} else {
|
|
error("Unknown language.bnd entry found: '%s'", ptr);
|
|
}
|
|
|
|
// Skip over newlines (and turn them into null bytes)
|
|
ptr = strpbrk(ptr, "\n\r");
|
|
if (ptr == NULL)
|
|
break;
|
|
while (*ptr == '\n' || *ptr == '\r')
|
|
*ptr++ = 0;
|
|
}
|
|
} else {
|
|
for (i = 0; i < _languageIndexSize; i++) {
|
|
// First 8 chars in the line give the string ID / 'tag'
|
|
int j;
|
|
for (j = 0; j < 8 && !Common::isSpace(*ptr); j++, ptr++)
|
|
_languageIndex[i].tag[j] = toupper(*ptr);
|
|
_languageIndex[i].tag[j] = 0;
|
|
|
|
// After that follows a single space which we skip
|
|
assert(Common::isSpace(*ptr));
|
|
ptr++;
|
|
|
|
// Then comes the translated string: we record an offset to that.
|
|
_languageIndex[i].offset = ptr - _languageBuffer;
|
|
|
|
// Skip over newlines (and turn them into null bytes)
|
|
ptr = strpbrk(ptr, "\n\r");
|
|
if (ptr == NULL)
|
|
break;
|
|
while (*ptr == '\n' || *ptr == '\r')
|
|
*ptr++ = 0;
|
|
|
|
// Convert '\n' code to a newline. See also bug #1487.
|
|
char *src, *dst;
|
|
src = dst = _languageBuffer + _languageIndex[i].offset;
|
|
while (*src) {
|
|
if (src[0] == '\\' && src[1] == 'n') {
|
|
*dst++ = '\n';
|
|
src += 2;
|
|
} else {
|
|
*dst++ = *src++;
|
|
}
|
|
}
|
|
*dst = 0;
|
|
}
|
|
}
|
|
|
|
// Sort the index nodes. We'll later use bsearch on it, which is just as efficient
|
|
// as using a binary tree, speed wise.
|
|
qsort(_languageIndex, _languageIndexSize, sizeof(LangIndexNode), indexCompare);
|
|
}
|
|
|
|
void ScummEngine_v7::playSpeech(const byte *ptr) {
|
|
if (_game.id == GID_DIG && (ConfMan.getBool("speech_mute") || VAR(VAR_VOICE_MODE) == 2))
|
|
return;
|
|
|
|
if ((_game.id == GID_DIG || _game.id == GID_CMI) && ptr[0]) {
|
|
Common::String pointerStr((const char *)ptr);
|
|
|
|
// Play speech
|
|
if (!(_game.features & GF_DEMO) && (_game.id == GID_CMI)) // CMI demo does not have .IMX for voice
|
|
pointerStr += ".IMX";
|
|
|
|
_sound->stopTalkSound();
|
|
_imuseDigital->stopSound(kTalkSoundID);
|
|
_imuseDigital->startVoice(kTalkSoundID, pointerStr.c_str(), _actorToPrintStrFor);
|
|
_sound->talkSound(0, 0, DIGI_SND_MODE_TALKIE);
|
|
}
|
|
}
|
|
|
|
void ScummEngine_v7::translateText(const byte *text, byte *trans_buff, int transBufferSize) {
|
|
if (isScummvmKorTarget()) {
|
|
// Support language bundle for FT
|
|
ScummEngine::translateText(text, trans_buff, transBufferSize);
|
|
return;
|
|
}
|
|
LangIndexNode target;
|
|
LangIndexNode *found = NULL;
|
|
int i;
|
|
|
|
trans_buff[0] = 0;
|
|
_lastStringTag[0] = 0;
|
|
|
|
if (_game.version >= 7 && text[0] == '/') {
|
|
// Extract the string tag from the text: /..../
|
|
for (i = 0; (i < 12) && (text[i + 1] != '/'); i++)
|
|
_lastStringTag[i] = toupper(text[i + 1]);
|
|
_lastStringTag[i] = 0;
|
|
}
|
|
|
|
// WORKAROUND for bug #1977.
|
|
if (_game.id == GID_DIG) {
|
|
// Based on the second release of The Dig
|
|
// Only applies to the subtitles and not speech
|
|
if (!strcmp((const char *)text, "faint light"))
|
|
text = (const byte *)"/NEW.007/faint light";
|
|
else if (!strcmp((const char *)text, "glowing crystal"))
|
|
text = (const byte *)"/NEW.008/glowing crystal";
|
|
else if (!strcmp((const char *)text, "glowing crystals"))
|
|
text = (const byte *)"/NEW.009/glowing crystals";
|
|
else if (!strcmp((const char *)text, "pit"))
|
|
text = (const byte *)"/NEW.010/pit";
|
|
else if (!strcmp((const char *)text, "You wish."))
|
|
text = (const byte *)"/NEW.011/You wish.";
|
|
else if (!strcmp((const char *)text, "In your dreams."))
|
|
text = (const byte *)"/NEW.012/In your dreams";
|
|
else if (!strcmp((const char *)text, "left"))
|
|
text = (const byte *)"/CATHPLAT.068/left";
|
|
else if (!strcmp((const char *)text, "right"))
|
|
text = (const byte *)"/CATHPLAT.070/right";
|
|
else if (!strcmp((const char *)text, "top"))
|
|
text = (const byte *)"/CATHPLAT.067/top";
|
|
else if (!strcmp((const char *)text, "exit"))
|
|
text = (const byte *)"/SKY.008/exit";
|
|
else if (!strcmp((const char *)text, "unattached lens"))
|
|
text = (const byte *)"/NEW.013/unattached lens";
|
|
else if (!strcmp((const char *)text, "lens slot"))
|
|
text = (const byte *)"/NEW.014/lens slot";
|
|
else if (!strcmp((const char *)text, "Jonathon Jackson"))
|
|
text = (const byte *)"Aram Gutowski";
|
|
else if (!strcmp((const char *)text, "Brink"))
|
|
text = (const byte *)"/CREVICE.049/Brink";
|
|
else if (!strcmp((const char *)text, "Robbins"))
|
|
text = (const byte *)"/NEST.061/Robbins";
|
|
}
|
|
|
|
|
|
if (_game.version >= 7 && text[0] == '/') {
|
|
// Extract the string tag from the text: /..../
|
|
for (i = 0; (i < 12) && (text[i + 1] != '/'); i++)
|
|
target.tag[i] = toupper(text[i + 1]);
|
|
target.tag[i] = 0;
|
|
text += i + 2;
|
|
|
|
// If a language file was loaded, try to find a translated version
|
|
// by doing a lookup on the string tag.
|
|
if (_existLanguageFile) {
|
|
// HACK: These are used for the object line in COMI when
|
|
// using one object on another. I don't know if the
|
|
// text in the language file is a placeholder or if
|
|
// we're supposed to use it, but at least in the
|
|
// English version things will work so much better if
|
|
// we can't find translations for these.
|
|
|
|
if (*text && strcmp(target.tag, "PU_M001") != 0 && strcmp(target.tag, "PU_M002") != 0)
|
|
found = (LangIndexNode *)bsearch(&target, _languageIndex, _languageIndexSize, sizeof(LangIndexNode), indexCompare);
|
|
}
|
|
}
|
|
|
|
if (found != NULL) {
|
|
Common::strlcpy((char *)trans_buff, _languageBuffer + found->offset, transBufferSize);
|
|
|
|
if (((_game.id == GID_DIG) && !(_game.features & GF_DEMO)) || _game.version == 8) {
|
|
// Replace any '%___' (or '%<var-name>%' for v8) by the corresponding special codes in the source text
|
|
const byte *src = text;
|
|
char *dst = (char *)trans_buff;
|
|
|
|
while ((dst = (_game.version == 8 ? strchr(dst, '%') : strstr(dst, "%___")))) {
|
|
// Search for a special code in the message.
|
|
while (*src && *src != 0xFF) {
|
|
src++;
|
|
}
|
|
|
|
// Replace the %___ (or %<var-name>%) by the special code.
|
|
// Luckily, we can do that in-place.
|
|
if (*src == 0xFF) {
|
|
if (_game.version == 7) {
|
|
memcpy(dst, src, 4);
|
|
src += 4;
|
|
dst += 4;
|
|
} else {
|
|
memcpy(dst, src, 6);
|
|
src += 6;
|
|
dst += 6;
|
|
*dst = '\0';
|
|
}
|
|
} else
|
|
break;
|
|
}
|
|
}
|
|
} else {
|
|
// Default: just copy the string
|
|
memcpy(trans_buff, text, resStrLen(text) + 1);
|
|
}
|
|
}
|
|
|
|
#endif
|
|
|
|
void ScummEngine::loadLanguageBundle() {
|
|
if (!isScummvmKorTarget()) {
|
|
_existLanguageFile = false;
|
|
return;
|
|
}
|
|
|
|
ScummFile file(this);
|
|
openFile(file, "korean.trs");
|
|
|
|
if (!file.isOpen()) {
|
|
_existLanguageFile = false;
|
|
return;
|
|
}
|
|
|
|
_existLanguageFile = true;
|
|
|
|
int size = file.size();
|
|
|
|
uint32 magic1 = file.readUint32BE();
|
|
uint32 magic2 = file.readUint32BE();
|
|
|
|
if (magic1 != MKTAG('S', 'C', 'V', 'M') || magic2 != MKTAG('T', 'R', 'S', ' ')) {
|
|
_existLanguageFile = false;
|
|
return;
|
|
}
|
|
|
|
_numTranslatedLines = file.readUint16LE();
|
|
_translatedLines = new TranslatedLine[_numTranslatedLines];
|
|
_languageLineIndex = new uint16[_numTranslatedLines];
|
|
|
|
// sanity check
|
|
for (int i = 0; i < _numTranslatedLines; i++) {
|
|
_languageLineIndex[i] = 0xffff;
|
|
}
|
|
|
|
for (int i = 0; i < _numTranslatedLines; i++) {
|
|
int idx = file.readUint16LE();
|
|
assert(idx < _numTranslatedLines);
|
|
_languageLineIndex[idx] = i;
|
|
_translatedLines[i].originalTextOffset = file.readUint32LE();
|
|
_translatedLines[i].translatedTextOffset = file.readUint32LE();
|
|
}
|
|
|
|
// sanity check
|
|
for (int i = 0; i < _numTranslatedLines; i++) {
|
|
if (_languageLineIndex[i] == 0xffff) {
|
|
error("Invalid language bundle file");
|
|
}
|
|
}
|
|
|
|
// Room
|
|
byte numTranslatedRoom = file.readByte();
|
|
for (uint32 i = 0; i < numTranslatedRoom; i++) {
|
|
byte roomId = file.readByte();
|
|
|
|
TranslationRoom &room = _roomIndex.getOrCreateVal(roomId);
|
|
|
|
uint16 numScript = file.readUint16LE();
|
|
for (int sc = 0; sc < numScript; sc++) {
|
|
uint32 scrpKey = file.readUint32LE();
|
|
uint16 scrpLeft = file.readUint16LE();
|
|
uint16 scrpRight = file.readUint16LE();
|
|
|
|
room.scriptRanges.setVal(scrpKey, TranslationRange(scrpLeft, scrpRight));
|
|
}
|
|
}
|
|
|
|
int bodyPos = file.pos();
|
|
|
|
for (int i = 0; i < _numTranslatedLines; i++) {
|
|
_translatedLines[i].originalTextOffset -= bodyPos;
|
|
_translatedLines[i].translatedTextOffset -= bodyPos;
|
|
}
|
|
_languageBuffer = new byte[size - bodyPos];
|
|
file.read(_languageBuffer, size - bodyPos);
|
|
file.close();
|
|
|
|
debug(2, "loadLanguageBundle: Loaded %d entries", _numTranslatedLines);
|
|
}
|
|
|
|
const byte *ScummEngine::searchTranslatedLine(const byte *text, const TranslationRange &range, bool useIndex) {
|
|
int textLen = resStrLen(text);
|
|
|
|
int left = range.left;
|
|
int right = range.right;
|
|
|
|
int dbgIterationCount = 0;
|
|
|
|
while (left <= right) {
|
|
dbgIterationCount++;
|
|
debug(8, "searchTranslatedLine: Range: %d - %d", left, right);
|
|
int mid = (left + right) / 2;
|
|
int idx = useIndex ? _languageLineIndex[mid] : mid;
|
|
const byte *originalText = &_languageBuffer[_translatedLines[idx].originalTextOffset];
|
|
int originalLen = resStrLen(originalText);
|
|
int compare = memcmp(text, originalText, MIN(textLen + 1, originalLen + 1));
|
|
if (compare == 0) {
|
|
debug(8, "searchTranslatedLine: Found in %d iteration", dbgIterationCount);
|
|
const byte *translatedText = &_languageBuffer[_translatedLines[idx].translatedTextOffset];
|
|
return translatedText;
|
|
} else if (compare < 0) {
|
|
right = mid - 1;
|
|
} else if (compare > 0) {
|
|
left = mid + 1;
|
|
}
|
|
}
|
|
|
|
debug(8, "searchTranslatedLine: Not found in %d iteration", dbgIterationCount);
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
void ScummEngine::translateText(const byte *text, byte *trans_buff, int transBufferSize) {
|
|
if (_existLanguageFile) {
|
|
if (_currentScript == 0xff) {
|
|
// used in drawVerb(), etc
|
|
debug(7, "translateText: Room=%d, CurrentScript == 0xff", _currentRoom);
|
|
} else {
|
|
// Use series of heuristics to preserve "the context of the conversation",
|
|
// since one English text can be translated differently depending on the context.
|
|
ScriptSlot *slot = &vm.slot[_currentScript];
|
|
debug(7, "translateText: Room=%d, Script=%d, WIO=%d", _currentRoom, slot->number, slot->where);
|
|
|
|
byte roomKey = 0;
|
|
if (slot->where != WIO_GLOBAL) {
|
|
roomKey = _currentRoom;
|
|
}
|
|
|
|
uint32 scriptKey = slot->where << 16 | slot->number;
|
|
if (slot->where == WIO_ROOM) {
|
|
scriptKey = slot->where << 16;
|
|
}
|
|
|
|
// First search by _currentRoom and _currentScript
|
|
Common::HashMap<byte, TranslationRoom>::const_iterator iterator = _roomIndex.find(roomKey);
|
|
if (iterator != _roomIndex.end()) {
|
|
const TranslationRoom &room = iterator->_value;
|
|
TranslationRange scrpRange;
|
|
if (room.scriptRanges.tryGetVal(scriptKey, scrpRange)) {
|
|
const byte *translatedText = searchTranslatedLine(text, scrpRange, true);
|
|
if (translatedText) {
|
|
debug(7, "translateText: Found by heuristic #1");
|
|
memcpy(trans_buff, translatedText, MIN<int>(resStrLen(translatedText) + 1, transBufferSize));
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
// If not found, search for current room
|
|
roomKey = _currentRoom;
|
|
scriptKey = WIO_ROOM << 16;
|
|
iterator = _roomIndex.find(roomKey);
|
|
if (iterator != _roomIndex.end()) {
|
|
const TranslationRoom &room = iterator->_value;
|
|
TranslationRange scrpRange;
|
|
if (room.scriptRanges.tryGetVal(scriptKey, scrpRange)) {
|
|
const byte *translatedText = searchTranslatedLine(text, scrpRange, true);
|
|
if (translatedText) {
|
|
debug(7, "translateText: Found by heuristic #2");
|
|
memcpy(trans_buff, translatedText, MIN<int>(resStrLen(translatedText) + 1, transBufferSize));
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Try full search
|
|
const byte *translatedText = searchTranslatedLine(text, TranslationRange(0, _numTranslatedLines - 1), false);
|
|
if (translatedText) {
|
|
debug(7, "translateText: Found by full search");
|
|
memcpy(trans_buff, translatedText, MIN<int>(resStrLen(translatedText) + 1, transBufferSize));
|
|
return;
|
|
}
|
|
|
|
debug(7, "translateText: Not found");
|
|
}
|
|
|
|
// Default: just copy the string
|
|
memcpy(trans_buff, text, MIN<int>(resStrLen(text) + 1, transBufferSize));
|
|
}
|
|
|
|
bool ScummEngine::reverseIfNeeded(const byte *text, byte *reverseBuf, int reverseBufSize) const {
|
|
if (_language != Common::HE_ISR)
|
|
return false;
|
|
if (_game.id != GID_LOOM && _game.id != GID_ZAK)
|
|
return false;
|
|
Common::strlcpy(reinterpret_cast<char *>(reverseBuf), reinterpret_cast<const char *>(text), reverseBufSize);
|
|
fakeBidiString(reverseBuf, true, reverseBufSize);
|
|
return true;
|
|
}
|
|
|
|
Common::CodePage ScummEngine::getDialogCodePage() const {
|
|
switch (_language) {
|
|
case Common::KO_KOR:
|
|
return Common::kWindows949;
|
|
case Common::JA_JPN:
|
|
return Common::kWindows932;
|
|
case Common::ZH_TWN:
|
|
case Common::ZH_CHN:
|
|
return Common::kWindows950;
|
|
case Common::RU_RUS:
|
|
return Common::kDos866;
|
|
case Common::HE_ISR:
|
|
switch (_game.id) {
|
|
case GID_LOOM:
|
|
case GID_ZAK:
|
|
return Common::kDos862;
|
|
default:
|
|
return Common::kWindows1255;
|
|
}
|
|
default:
|
|
return (_game.version > 7) ? Common::kWindows1252 : Common::kDos850;
|
|
}
|
|
}
|
|
|
|
} // End of namespace Scumm
|