ws.md

Translating Wind's Seed

Background

• I've translated one other PC-98 game before, so I got on my feet pretty quick with this one.
• I read Japanese at an N2 level. I kinda know x86 assembly.
• I'm armed with np2debug, wxmedit, Ghidra, and Python. Let's do this!
• I had some old Python code lying around for reading .fdi files into a Python dictionary, so started from that.

Text rendering

• The game stores 早く行こうよ as 早 2 行 4 - P. When it sees an ASCII character, it looks it up in a kana table.
• We want to disable that code so we can actually print English strings without them becoming all kana.
• But now there's another problem: the game only moves the text cursor in full-width (double) increments.
• We change this to single increments, so our text doesn't look L i k e t h i s.
  • https://github.com/46OkuMen/windseed contains a patch that does this much.
• But now we have to double the X coordinate of a bunch text positions to compensate.
• This involves a bunch of assembly patches, and seems to be where 46okumen put things down.
• I found a nice place to insert a shl bl, 1 instruction that fixes most of it in menus.

Text in menus

• In menus, only the last character of text was visible. Hmm...
• Menu text rendering involves opaque backgrounds, unlike the dialogue.
• I learned that when you ask Kanji ROM for an ASCII character, it gives it right-aligned in a 16×16px bitmap, so to speak.
• So if we draw A with an opaque background, and then overlap B 8px to the right, we'll get… B.
  • We didn't notice this in the dialogue boxes because there's no opaque background.
• To remedy this I replaced the code that bolds character bitmaps with code that moves them 8px to the left.
  • I replaced row |= row << 1 (how clever!) with row <<= 8, basically.
• Then I disabled some instructions that draw the right half of a 16×16px background rectangle.

Digits

• The kana compression was in charge of mapping !"#$%^&'() to fullwidth digits ０１２３４５６７８９.
• So when we disabled kana compression, numbers on the screen reverted into crazy punctuation.
• I tracked down some instructions like add al, '!' and replaced them with add al, '0' to get ASCII digits.

Non-story text translation

• I wrote some strange code to try to find strings that look like Japanese all over the disk.
• This worked pretty well! Here is everything it found.
• I'm not sure why I translated the error messages and stuff. I was on a roll!!

Story

Dumping

• I ported celcion's PowerShell code for dumping WSDATA.FCP to Python.
• This is a simple archive format that contains all the game's graphics and text.
• It contains STORY.DAT. This contains all the textbox dialogue in the game.
• Figuring out the control codes (set portrait, newline, next text box) was not that hard.
• The first 0x200 bytes of this file form an array ushort[256] of pointers into the file.
• I dumped it into a human-editable format that's kind of arbitrary but plays nice with HTML syntax highlighting.

= event 0x0090 at 0x2207

<!-- <RB><RS TownLady>おそろしい世の中だねぇ<Next> -->
<!-- <End> -->
<RB><RS TownLady>Such a dreadful world we live in.<Next>
<End>

Reinserting

• I wrote code to parse my own format and overwrite the STORY.DAT region of WSDATA.FCP.
• That code automatically inserts newline commands to fit text in 22 columns.
• There's not too much text in the game, so translating everything took only a few days.
  • There are a couple of really weird in-jokes. Maybe I'll write translator's notes later.

Size troubles

• Uh-oh, the file became way bigger. It was 12,304 bytes, but in English it's almost 21,000 bytes!
  • English is often bigger than Japanese: think 森 (2 bytes in SJIS) vs. forest (6 bytes).
  • With compressed kana, the difference is even larger.
• So, I wrote code to repack WSDATA.FCP to fit a large STORY.DAT in it.
• Then I wrote code to repack the floppy disk to fit a large WSDATA.FCP in it...
  • This was stupid and required me learning how FAT12 works. I should have probably used a library!
• Crap, the game keeps the story in a 0x4000-byte array, but again, we have a little over 0x5100 bytes now.
• I bumped the array size up to 0x6000, and moved a subsequent array further down RAM so they don't overlap.

Miscellaneous romhacking wisdom

• One easy way to find code is to set breakpoints on data you know it reads.
• Another is to guess instructions it probably executes and search for those! (Like add al, '!'.)
• If you don't know what some code does, try replacing it with NOPs and running the game like that.

Testing

• Testing a translation is pretty annoying without some form of cheats.
• I don't want to worry about losing a boss battle if I just want to check if my translation looks good.
• I poked at the save file format, and gave my party 999 HP, 999 attack, 999 defense etc.
• Then I ran through the game, occasionally saving to make a "checkpoint", so I could access areas/scenes easily.