Strider in MSHvsSF.

Strider Hiryu is in Marvel Super Heroes vs. Street Fighter now. He's the bottom middle square of the select screen, the one that only showed the game's logo. He plays the way he plays in Marvel vs. Capcom, because it's the same code: the installer takes his code, his sprites and his sounds out of your own mvsc.zip and puts them into your own MSHvsSF zip.

This is the part under the hood. How I found that square is in two earlier posts: Norimaro and the one byte, and Strider's face in the square. Back then the square had his face and nothing else. Now he's playable.

What's in v1

• The square. Bottom row, middle. It's the slot the Japanese game gives Norimaro, slot $22. The punch buttons pick Strider's first colour, the kick buttons his second.
• His moves. His MvC normals and specials, run by his MvC code, with his MvC inputs: Ame no Murakumo (QCF+P), the teleport (B,D,DB+K), Gram (F,D,DF+P), Excalibur (QCF+P in the air), and two kick specials, QCF+K and QCB+K.
• His hypers. Ouroboros (QCF+PP), Legion (QCF+KK) and Ragnarok (F,D,DF+PP), a level each.
• The team moves. Tag, Variable Combination and Variable Counter, the same as for everyone else in MSHvsSF.
• His voice and sounds. 42 sound commands: his voice, his effects and his theme, with 630,613 bytes of sampled sound.
• His name everywhere. The select screen, the life bar, the versus and win screens, continue and attract mode. He wins with his MvC win quotes.
• His ending. His three MvC ending pictures and his last line: "Target terminated. Mission complete."

Same inputs, same Strider

The first thing I wanted to know was whether he plays like MvC's Strider, not something that looks like him. So I recorded the same inputs in both games and compared them frame by frame.

In diagnostic runs (cloud MAME 0.264, headless, not my pinned MAME), I played 47 input sequences into both games. In 46 of them Strider did the same actions in the same order. The one that differs, a QCF+PP, has one extra idle moment before the hyper starts. Player 2 reacted the same way in all 47.

Blocking matches too, in the same diagnostic runs: the guard starts on the same frame as in MvC, and blocking all 47 sequences, at range and up close, throws no error. One pushback differs, and stock Ryu shows the same difference between the two games, so that one is MSHvsSF's own.

His code

In MvC, Strider is his own block of 68000 code and data. The patch moves that block into MSHvsSF instead of rewriting him:

• 16,722 bytes of his MvC code, relocated to $3F1000, near the top of MSHvsSF's program space, in the 10b chip.
• 11 child and victim blocks: the code for the things he spawns, and for the fighters on the other end of his moves.
• 1,302 bytes of bridge code at $3D4000. These are small shims for the places where MvC's code expects something at an MvC address, and MSHvsSF keeps it somewhere else.

CPS-2 programs are encrypted. The board decrypts its program as it runs it, with a key each game has. MAME's sets keep that key in a 20-byte file inside the zip, mvsc.key for MvC. The installer uses the key in your own mvsc.zip to decrypt Strider's code from MvC's 03 and 04 chips, the same way the board does. MSHvsSF's 10b chip is outside its game's encrypted range, so his code goes there as plain bytes. The download carries none of that code.

Inside the game he lives where every MSHvsSF fighter lives: player 1 at $FF3800, player 2 at $FF3C00, the partners at $FF4000 and $FF4400. His sprites go into object RAM at $708000, his colours into palette RAM at $914000, and his sound commands into the QSound chip's shared RAM at $618000.

The crash on block

The candidate before this one crashed when the other player blocked one of Strider's cypher moves. On a blocked hit, MSHvsSF looks up the guard spark in a table of 16 entries at $018C84. His cypher moves use spark numbers past the end of that table, so the game read past it and crashed.

The fix is one more shim, shim_guardspark. It has MSHvsSF's 16 entries plus MvC's guard sparks for the cypher moves. In the diagnostic runs, the crash replay now runs to its end, and all 47 sequences blocked at range and up close throw no error.

His sprites, effects and sounds

• 10,905 sprite tiles, at fixed places in MSHvsSF's graphics chips (13m to 20m): the space Norimaro's art used. His slot is Strider's now, so Norimaro's tiles aren't needed.
• 60 effect scripts and 683 effect tiles for his effects.
• 42 sound commands and 630,613 bytes of sampled sound, in the sound chips (11m and 12m).

On either set, the installer's own log line counts every tile it writes: 12,078. That's his sprites and effects, plus his portraits and pictures for the select screen, the versus and win screens, the life bar and his ending.

His sounds are checked in the build's data only: the cloud MAME I tested with has no sound output. If something sounds wrong, tell me in the poll below.

How the square opens

The USA game hides slot $22 behind a flag in work RAM, $FFFF31. In the first post I turned that flag on with one byte in a table of region defaults. That byte is only read when the game sets up a new EEPROM, the memory where the board keeps its settings. A saved EEPROM from before the patch kept the square locked.

This release doesn't use that byte. Four branch instructions in the select-screen code, at $009E58, $00AB90, $06CEC0 and $06CFB8, skip the flag, so the square is open no matter what the EEPROM says.

The Euro 970625 set has no such flag. It locks the square with a byte of its ROM header, $0000CD, instead. The Euro build forces the same four branches at their Euro addresses, $009E5A, $00AB94, $06CEE8 and $06CFE2, so on both sets the square opens in the code, not in the EEPROM.

That matters for Fightcade. Fightcade's FBNeo never saves or loads the CPS-2 EEPROM (those lines are commented out in its source, marked "@FC"), and at the start of a match it can load a start-up savestate, savestates\mshvsf_fbneo.fs, if the client ships one. With the unlock in the select-screen code, neither matters.

I checked that in Fightcade's FBNeo core, built headless from its source (fightcade-fbneo c959501). This is a diagnostic run of the core, not of Fightcade itself. On both sets, after a fresh boot, and after savestates made on the original game at two points, Strider was selectable and playable, with no exceptions. On the USA set, the build before the fix failed the savestate case. Fightcade's own front end and online play are untested.

MAME is different in one way. Its savestates carry the decrypted program, so a savestate made on the original game, loaded into the patched one, puts the old code back until you restart. Make your savestates on the patched game.

What changes, chip by chip

On either set, the installer's output differs from the clean set in 2,018,896 bytes on 19 of its 21 files. On the USA set, that's the 2,018,888 bytes of the preview build the clips show, plus the eight bytes of the four select-screen branches. The files it changes are the sound program (01, 02), the program (03, 04, 05, 06, 07, 09, 10b), the samples (11m, 12m) and all eight graphics chips.

What the installer does

The installer is one .cmd file for Windows. It holds a PowerShell script and the recipe: which bytes, tiles and samples to copy from which of your files to where, plus the patch's own code and tables. The patch package has the same recipe with a Python patcher, for macOS and Linux. Both write the same bytes.

1. It finds your two zips: the ones you dropped on it, or the ones next to it.
2. It checks every file of both, 21 and 21: size, CRC32 and SHA-1 against the clean sets. A patched, changed or incomplete set is refused with one line that says why.
3. It builds the new set in memory, then checks every file it built against the SHA-1 it should have.
4. It writes mshvsf_strider_v1.zip (or mshvsfu_strider_v1.zip) next to your zip, and a .receipt.json with the sizes and hashes of everything it read and wrote. Your zips are not changed.

The new zip is stored, not compressed. Compression can differ between tools and machines, and stored bytes can't, so every correct output has the same SHA-256 on every PC. The installer prints it, and the install section above lists the one each set should give. Two Fightcade players compare theirs, and if they match, they have the same file.

The download holds no ROM file of either game. Two pieces of MvC art are in it, because the installer can't rebuild them from your zip: his select-screen icon, downscaled, and his three ending pictures, re-coloured into 25 palettes. That's about 24 KB. Everything else of Strider comes from your mvsc.zip.

His name, and the secret menu

His name is right in places players don't usually look, too. In the game's secret test menu, CATCH EDIT lists him as 11 STRIDER HIRYU, and the KAO TEST has his row. In 146 attract-mode shots of the diagnostic runs, Norimaro showed up nowhere.

What got tested

All of these are diagnostic runs, in cloud MAME 0.264, headless and unpinned, unless the line says otherwise:

• Moves: 47 input sequences against MvC with the same inputs: 46 give the same actions, player 2 reacts the same in all 47.
• Blocking: guard starts on MvC's frame, and every move blocked at range and after a walk-in throws no error.
• Team: tag out and back in, Variable Combination with both cut-ins and both hypers, Variable Counter forward and back.
• His name: select screen, life bar, versus and win screens, win quotes, continue and 146 attract shots, with no Norimaro anywhere.
• His ending: his three pictures and his line, through the secret test menu's ending test. I haven't cleared the arcade mode with him yet.
• The Euro set: the same checks on the Euro build, in the same kind of runs: 46 of the 47 sequences the same (the same QCF+PP), blocking, the three hypers, team moves, win screen, ending and CATCH EDIT. Its attract mode matches the stock Euro game frame for frame, 9,000 frames and 141 shots.
• Fightcade's FBNeo core (a diagnostic build from its source, not Fightcade): both sets, fresh boot and savestates made on the original game, Strider selectable and playable.
• The installers: the Windows installer's PowerShell part under PowerShell 7 and the Python patcher write the same output, byte for byte, for both sets.
• Not tested yet: Windows PowerShell 5.1 itself, real Windows, Fightcade's own front end and online play, and real hardware. His audio is checked in the data only.

Known limits

• Strider is never a CPU opponent. Slot $22 isn't in the pool the arcade mode picks opponents from.
• The six colour buttons land on MvC's two Strider colours.
• Up close, right after a forward walk straight into the motion, MSHvsSF reads his QCF+LK, QCF+HK and DP+LK as typed. MvC reads the first two as DP+K and the third as a normal. Stock Ryu reads them the same way in both games, so the difference is in the port. It isn't fixed yet.
• On the USA set, the first attract-mode demo fight starts one frame later than on the stock game, so the demo fights play out differently from there.
• The clips on this page were recorded on the v1 preview build 834ab7b8, on the USA set. The USA release differs from it only in the four select-screen branches above, so the gameplay is the same.

If you play him, the poll below is where I read what to fix next.