7

I want to extract the strings in Shadowgate for NES. I ran file on the image and then strings, no luck. I found some information about the NES cartridge file format. The docs mention the use of “Name Tables”. Is there a way to disassemble this file and view the strings? I tried

 strings -e -T rom.bin

I also tried :

 objdump -i rom.bin

The processor looks to be an M6502 processor and there are Windows disassemblers available.

  • If you read that document, you will see that "Name Table" has nothing to do with strings. This rather seems to relate to the NES way of storing sprite tiles, as used for scrolling backgrounds. – usr2564301 Nov 2 '13 at 16:30
  • If you say so - "0-3 String "NES^Z" used to recognize .NES files." Unfortunately this was the only reference to 'string' i could find in that document. I'm not an iNES hacker - I presume you are? The C definition of string is "abcd\0" - this in itself tells us nothing about how the string is going to be drawn does it? "Name Tables" is discussed under "PPU Memory" and this is what it says: "A Name Table is very similar to the text mode screen buffer which contains codes of characters to display on the screen." PLEASE READ CAREFULLY "very similar to the text mode screen buffer" – user3094 Nov 2 '13 at 17:02
  • aren't strings written to a text mode screen buffer? – user3094 Nov 2 '13 at 17:04
  • The Name table is for tiles what the text buffer is for text -- that's why they are "very much like", but not equal. As for your original question: strings looks for a very specific type of text: ASCII text only. I scanned your file for non-ASCII strings and found none (useful or otherwise). The strings must be encrypted or compressed, and it may need disassembly to find them. – usr2564301 Nov 2 '13 at 17:20
  • This is what the man page for strings has to say: "-e encoding Select the character encoding of the strings that are to be found. ... Useful for finding wide character strings." Or encoded? – user3094 Nov 3 '13 at 1:05
13

There are several approaches to locating strings in an unknown file. One you already tried: strings. This looks for plain, unencoded ASCII text:

Strings looks for ASCII strings in a binary file [..] A string is any sequence of 4 (the default) or more printing characters ending with a newline or a null. (man strings)

But there are many reasons why this naive approach may fail. First off: not every text in the world is ASCII encoded. In fact, examining your file with a binary editor, you can find graphic images for the font used in the game at offset 0x20010 -- monochrome bitmaps of 8x16 pixels. If you assume the first character (a '0') is numbered zero, then 'A' is at position 31 -- definitely not ASCII text. Of course, it's possible the text drawing routine knows this, and re-orders characters to be printed according to this scheme; but, given the age of this particular game (1987) it is more likely that the textual data is stored according to this weird encoding.

In itself, however, this should not be a problem.

Googling for this game provides a number of screen shots, and you can read some of the texts that may appear -- "The last thing you remember", "Word of your historic quest", etc. --, and a noteworthy point is that all text appears to be in ALL CAPS.

How does that help? Well, if the encoding is remotely 'normal', the character code of an 'A' might be anything, but you can safely assume that code+1 is 'B', code+2 is 'C', and so on. Now let's assume the text "THE" occurs anywhere (a safe assumption). Subtract 'T' from the first byte in the data and note the difference. Subtract this difference from the next byte and test if it is an 'H'; if so, test the same difference on the next byte and see if it is an 'E'. Three times is a charm (in this case), and since the string "THE" ought to come up very frequent, you should see lots of hits with the same difference. Then you can write a custom routine to 'convert' all data bytes according to this scheme, and check again if you find useful strings.

That didn't work for Shadowgate.

Another option is that the text has been deliberately obfuscated. A popular (because fast) option was to XOR text with a constant. That way the text was not readily visible when inspected with a hex viewer, yet could easily be displayed. So I did the same as above, only now with a XOR operation instead of a constant subtraction. It didn't work either.

Next: given that SG is a text adventure, it stands to reason the writers tried to stuff as much as possible text into the poor NES memory. To find real world compression (ZIP, LZW) in such an old game is fairly rare, the compression schemes tended to be quite simple. After all, not only RAM was limited but CPU speed as well. What if every character is stored as a 5-bit sequence? That would save lots of memory -- every 8 characters of text could be stored in just 5 bytes, a compression rate of 62.5%.

Why "5-bit"? We're talking English text here, plus a handful of punctuation characters, plus (maybe) digits '0' to '9'. The alphabet itself is 26 characters long, which leaves another 6 values for anything else -- and, hey, one of the extra codes could mean "for the next character use all 8 bits".

Checking every 5 bits against my test string (which in cryptography is called a "crib"), I found the following:

candidate at 0570, delta is 41 H_A\`THE[TROLL[
candidate at 0670, delta is 41 _H\`ATHE[TROLL[
candidate at 0878, delta is 41 `AN`QTHE[TROLL[
candidate at 09E3, delta is 41 FROM^THE[DEPTHS
candidate at 1380, delta is 41 E[OF[THEM_A[THI
candidate at 13F0, delta is 41 ]NX_ATHE[WORDS[
candidate at 14C0, delta is 41 PD^`QTHE[FLAME[
candidate at 1BBA, delta is 41 UDGE[THEM[BY_A_
candidate at 22E0, delta is 41 ]BX_ATHE[GLASS[
candidate at 230D, delta is 41 ID_A[THE^SIGN[O
candidate at 2375, delta is 41 S[ON[THEM_A\`AB
candidate at 2390, delta is 41 LLOW[THE^VISCOU
candidate at 2528, delta is 41 F]PX_THE[STONE[
candidate at 25E6, delta is 36 @CP=KTHE@?OFHBS
candidate at 27F8, delta is 41 YDP]ATHE[BARK[O
candidate at 2B1E, delta is 41 D_H\]THE[WATER[

.. and many more. You can see it works, because I also decoded a few bytes before and after the test string, and that's recognizable as 'something' as well. The 'delta' shown is the difference between the five-bit code (0..31) and ASCII, and you can see it's 41 for the majority of strings (the only exception seems a false positive).

To assure that this is correct, I tried with another crib: KING (it's a fantasy game):

candidate at 0661, delta is 41 Y[LOOKING[SPEAR
candidate at 23B4, delta is 41 [DRINKING[TAR_A
candidate at 2B5D, delta is 41 [DRINKING_A\`AA
candidate at 8E1B, delta is 43 \XVFDKINGDHEEVE
candidate at 146F9, delta is 34 JL54HKING48A4:D

That seems to work out as well: not the 'king' I was expecting, but nevertheless good results with a delta of 41, random stuff with another delta.

But finding useful strings this way is rather fortunate, because of course there is no guarantee that reading every 5 bits starting at the first byte should display anything useful. There may be lots of other strings in between the ones shown, but they didn't happen to start on a multiple of 5*8 bits. Suppose there was no text at position #0, but there was at position #1, then I cannot see it:

 bits for byte 0,1
 0000.0000 TTTT.T000 (T = Text character bits)
 ---
 reading 1st 5 bits
 1111.1??? ????.????
 2nd 5 bits -- the wrong ones!
 .... .111 11??.????

To properly decode all strings, you'd now take the following route:

  • my list of results contain readable text, but some garbage as well. Find out what the 'garbage' is ([ appears to be a simple space, but THEM_A\'AB needs a closer look).
  • find as much as possible string starts and note down their addresses
  • search the binary for these addresses. After all, if they are 'used', there needs to be some reference to them.
  • Before and after these addresses, there will be more. These are addresses of strings the search algorithm did not find, but still may be valid.
  • Usually, a list of this kind is a contiguous one (although there may be some data associated with each string). Scan the binary up and down for similar addresses, until you found what's sure to be the start and the end.
  • Loop over the list and display everything you can according to the decoding scheme.
  • Sit back and enjoy a job well done.
  • wow! cool! Thanks :) that was very clear! – user3094 Nov 3 '13 at 1:27
  • Here is a time-saver you might want to know about: the input file consists of several parts (I think this mimicks the original physical ROM chips). The first 16 bytes are 'header' and can be ignored. Then follow 65,536 bytes of code, and the rest is graphics data -- uninteresting for your purpose. You only need to scan the code part. – usr2564301 Nov 3 '13 at 2:02
  • hi! did you use a tool to do the scanning?? I'm using python to script manually. – user3094 Nov 3 '13 at 4:59
  • Another ROM oddity: the first half and the second half of the 64K image are swapped! No wonder it took me to 3.00/night to find anything useful. By the way, a list of useful string offsets is at 0x8000 (after correcting the ROM image file); for example, at offset A182: "DOST THOU KNOW? BRING^ME( THE )ANSWER TO MY^RIDDLE AND I SHALL LET^THEE PASS." Oh, and I use plain C. Not that the language matters--as long as you can manipulate bit streams, you're good. – usr2564301 Nov 3 '13 at 12:59
  • Actually, scratch the bit about the ROM halves. The entire thing is definitely not in a logical order, but it's not as easy as I figured. This needs a real NES-ROM hacker to make sense of. Ripping out strings still works, but the 'find offsets' part cannot if you can't get 'actual' addresses. – usr2564301 Nov 3 '13 at 14:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy