Version fields

The word version names six unrelated things in this codebase, and only one of them is about which commercial release you are running. This document separates them, records exhaustively what each one changes, and pins the release-identity story against what the pressed discs actually contain.

Nothing here is a proposal. It is a record of the original mechanism, plus the places the port has touched it.

Field	Storage	Identifies	Status
DistribVersion	Version key in lba2.cfg, or the data when none declares one	Which publisher's edition this is	Live, 6 branch sites
Version_US	Version_US key in lba2.cfg	Nothing	Read, never used
NumVersion / NUM_VERSION	First byte of a .lba save	Save layout revision	Live, see SAVEGAME.md
SAVE_VERSION	Nothing	Save routine lineage	Declared, never referenced
XPL_HEADER.Version	First field of a .XPL header	Palette/shading file format	Present on disk, never read
Version (the string)	Binary .rodata	The build	Port-added, see RELEASING.md

DistribVersion

Where the value comes from

One read, in ReadConfigFile (PERSO.CPP:2447):

// version distributeur
DistribVersion = DefFileBufferReadValueDefault("Version", UNKNOWN_VERSION);

Two things about that line carry most of the story.

The config key is Version, the global is DistribVersion. The key name says nothing about what it holds, which is why an install assembled by hand tends to lose it silently.

A declaration always wins. That is the original contract, and it holds whatever the assets turn out to be: the installer states the release, WriteConfigFile never writes Version back, and the game cannot change its own release identity. What the port adds is a measurement beside it, so a declaration that contradicts its own data is reported instead of silently drawing one publisher's sprites over another's. See Reading it off the data.

With the key absent the value is UNKNOWN_VERSION unless the data measures as the American master, in which case it is the American arm. Nothing else about an undeclared install changes: 0 already behaves as EA everywhere except the splash, so recognising a re-release alters nothing about how it runs, and a value that came from the assets never brings a publisher splash with it.

The compiled-in default at GLOBAL.CPP:21 is ACTIVISION_VERSION, but ReadConfigFile always overwrites it, so it only applies before the config is read.

The six values

DEFINES.H:200-206:

Value	Constant	Publisher
0	UNKNOWN_VERSION	none declared
1	ACTIVISION_VERSION	Activision
2	ACTIVISION_SUD_VERSION	Activision, sud (French for south)
3	EA_VERSION	Electronic Arts
4	VIRGIN_VERSION	Virgin
5	VIRGIN_ASIA_VERSION	Virgin, Asia

Three publishers, two of them with a territory split. The source never says which territories SUD and ASIA cover, and it does not need to: as the next section shows, no branch in the shipped code distinguishes 2 from 1 or 5 from 4.

Every place it branches

Six switch statements, and this is the complete list. The first covers two branches at once.

Site	Branch	{0, 3}	{1, 2, 4, 5}
PERSO.CPP:3383	CD volume label to look for	LBA2	TWINSEN
PERSO.CPP:3383	MessageNoCD	MESSAGE_NO_CD	MESSAGE_NO_CD_US
CONFIG.CPP:1463 AskForCD	Insert-disc prompt	text id 7	text id 6
MUSIC.CPP:229 InitTabTracks	Track table	TrackCD, first CD track 6	TrackCDUS, first CD track 0
GAMEMENU.CPP:1108 DrawCadreNewGame	New-game panel sprite	11	16
OBJECT.CPP:6122	In-game corner logo sprite	11 at x-103	16 at x-110
GAMEMENU.CPP:4870 DistribLogo	Boot splash	see below	see below

Voices are not on the list. Nothing in SOURCES/ or LIB386/ consults DistribVersion to build a voice path, so the folder read off a disc does not depend on the release.

The two messages are in DEFINES.H:24-25: "You need LBA2 CD, sorry!" and "You need Twinsen's Odyssey CD, sorry!".

All but one ask the same yes/no question: is this UNKNOWN or EA, or is it anything else? That is the European identity, where the game is Little Big Adventure 2 and the disc is labelled LBA2, against the Activision/Virgin identity, where it is Twinsen's Odyssey and the disc is labelled TWINSEN. Every consumer is written as a switch with UNKNOWN_VERSION and EA_VERSION cased together and everything else falling to default.

Only DistribLogo needs more than two answers, and it needs exactly three:

Value	Splash	Resource
1, 2	Activision	PCR_ACTIVISION (72)
3	Electronic Arts	PCR_EA (74)
4, 5	Virgin	PCR_VIRGIN (76)
0	none	no splash at all

Resource indices come in pairs, image then palette, which is why they step by two (COMMON.H:162-164, and ShowLogo loading numscr and numscr + 1).

So the practical effect of the six constants is:

• 2 behaves identically to 1, and 5 identically to 4, everywhere.
• 0 behaves identically to 3 everywhere except the splash, which 0 skips.

The finer constants record a distinction the game was never asked to make. Since the value only ever arrives from an installer-written config, the likeliest reading is that the granularity was for the installers rather than the game.

Which of those you can actually see

Two of them, on this port.

The boot splash is one, and the new-game panel sprite is the other.

The three CD-related branches are computed and then discarded. InitCD passes the volume name to OpenCD, and the SDL backend at CD.CPP:50 ignores the argument entirely and returns a fixed drive letter, so the check always succeeds. The no-CD message, the AskForCD prompt and the retry loop are therefore unreachable. Under the original MILES backend this was real: OpenCD walked drives A to Z, opened each as a Red Book device, and compared the volume label against the name it was given (MILES/CD.CPP:39). That comparison is the one place the engine ever cross-checked DistribVersion against the physical medium.

The music table selection no longer changes what plays: both tables hold the same music numbers and differ only in a JINGLE flag that nothing reads now that PlayMusic routes everything through PlayJingle. The header comment at MUSIC.CPP:218 explains why the tables were kept anyway, which is that they are original data recording which entries the US disc pressed as Red Book audio. See MUSIC.md and DISC_IMAGE_SOURCE.md.

The in-game corner logo is gated on DemoSlide, so in a normal session it never draws. That gate is what makes demo mode the only capture surface where DistribVersion visibly differs, which TESTING.md relies on.

What the shipped releases declare

Verified against retail media, not inferred: config bytes dumped from the disc, volume labels and directory trees read from the ISO 9660 descriptors.

Release	Config header comment	Version	Volume label	Game directory
Retail CD, Activision	(Activision)	1	TWINSEN	/TWINSEN/
Retail CD, Electronic Arts	(Electronic Arts)	3	LBA2	/LBA2/
GOG	(Electronic Arts)	3	n/a	n/a
Steam Classic	no header comment	key absent	n/a	n/a
Demo	no header comment	key absent	n/a	n/a
SOURCES/LBA2.CFG in this repo	no header comment	key absent	n/a	n/a

Two things follow.

The disc layout matches the branch it selects, on every axis the code touches. An Activision config saying 1 sits on a disc stamped TWINSEN, whose game directory is /TWINSEN/, whose voice files are therefore at /TWINSEN/VOX/, which is exactly the path DIRECTORIES.CPP:406 builds for values 1, 2, 4 and 5. An EA config saying 3 sits on a disc stamped LBA2 with /LBA2/ and /LBA2/VOX/, which is what the same line builds for 0 and 3. The engine verifies none of this. The mastering was simply self-consistent, which is why a mechanism that is pure declaration held up in 1997.

Both pressings keep LBA2.CFG inside the game directory next to LBA2.HQR, and both set Version. The Activision one ships it fully populated, with language, volumes and key bindings already filled in; the EA one ships a template with those fields blank.

Nothing tested carries 2, 4 or 5. ACTIVISION_SUD_VERSION, VIRGIN_VERSION and VIRGIN_ASIA_VERSION are attested only in the source.

Why modern installs report unknown

A fresh profile with no config is seeded from the config sitting beside the game assets, so on GOG or a mounted retail image the medium tells the engine what it is. Steam Classic breaks that in two independent ways: it keeps its config under %USERPROFILE%\Saved Games\2point21\tlba2-classic\Settings\lba2.cfg rather than beside the assets, so the seeding never reaches it, and that file has no Version key in any case. It has a SteamLanguage key instead. Either way the answer is 0, which differs from 3 only in the splash.

The same happens to any install assembled by copying assets without the config. See GAME_DATA.md and DISC_IMAGE_SOURCE.md.

Changing it

The distrib console command reports the current value and writes a new one (CONSOLE_CMD.CPP:889). It accepts either the number or the name (unknown, activision, activision_sud, ea, virgin, virgin_asia), and it is the only thing in the tree that writes the Version key.

It takes effect on the next launch, deliberately. DistribVersion is latched at boot into derived state: MessageNoCD is copied, PtrTrackCD and FirstCDTrack are set by InitTabTracks, and the CD voice path is resolved once and cached in a function-local static. Mutating the global at runtime would change some of those and not others.

Editing the Version line in lba2.cfg by hand does the same thing.

Reading it off the data instead

Everything above is what an installer wrote down. scripts/dev/fingerprint_distro.py answers the same question from the payload, for a directory, a disc image or a cue sheet:

scripts/dev/fingerprint_distro.py <path>...      # or --scan <dir> for every child

It reports the master and the pressing separately, because they move independently.

RESS.HQR names the master. It takes one of two values across every release checked, and the split is the one the engine makes: {UNKNOWN, EA} is LBA2, everything else is Twinsen's Odyssey. Three of its fifty entries carry the difference and none of them is branding, so it records which data master the copy was built from rather than who sold it. Both Activision pressings ship it byte for byte despite being different locales.

SCENE.HQR and TEXT.HQR name the pressing, locale included, which is exactly what RESS cannot see.

That separation is what makes an unsampled release degrade usefully. A pressing from a publisher not in the table answers with its master immediately and only the pressing reads as unrecorded, where a single combined key would have answered nothing at all. The six DistribVersion constants are six labels on those two masters, so a Virgin disc is Twinsen's Odyssey and the tool says so.

A declaration that agrees about the master is taken as naming the publisher, which the measurement cannot reach. One that disagrees is reported as a conflict: it means either the rules are wrong or the install was assembled from two sources.

What the engine does with it

The same measurement, on one RESS.HQR and one FileSize, which goes through OpenRead and so sizes a bank inside a mounted disc image from its directory record without reading it. The rules live in DISTRIB.CPP and are pinned by tests/distrib:

Config	Release identity	Splash
declares a Version	as declared	shown
declares nothing, data is LBA2	0	none
declares nothing, data is Twinsen's Odyssey	1	none
declares nothing, data unrecognised	0	none

ShowDistribLogo overrides the splash in either direction and is never written back, so the default keeps deriving from whether a release was declared rather than freezing one install's answer into a profile.

The boot line reports both halves, operative value first:

Release    default (0), data is LBA2
Release    activision (1) from the config, data agrees
Release    activision (1) from the data
Release    ea (3) from the config, data is Twinsen's Odyssey

The last is the case nothing else catches. A size is a weak hash and the sample behind the table is small, so a tree matching no entry reads as unrecognised and keeps the behaviour it would have had anyway.

Version_US

A vestige. Declared at C_EXTERN.H:139, defined as TRUE at GLOBAL.CPP:112, and read from the config at PERSO.CPP:2351:

Version_US = DefFileBufferReadValue("Version_US");

Nothing reads it after that. Not one branch in SOURCES/ or LIB386/.

It is worse than merely unused. DefFileBufferReadValue is the no-default variant, which returns -1 when the key is missing, and no shipped config sets Version_US. So the read reliably replaces the initial TRUE with -1 on every launch, and nothing notices.

It predates LBA2. lba1-classic carries the same extern LONG Version_US; in its C_EXTERN.H with no definition and no use anywhere, so the declaration was already dead when it was copied forward. LBA1 has no distributor mechanism at all; DistribVersion and the six constants are LBA2 additions.

The name invites the assumption that it is the US/EU switch. It is not; that job belongs entirely to DistribVersion.

Other installer-written keys

Three more keys are written by the DOS installer and never read by the engine. They travel with Version and are worth recognising when reading a shipped config.

Key	Written by	Read by
LanguageInstall	installer	nothing
Demo	installer	nothing
PathInstall	installer	nothing

The in-repo template leaves Version out on purpose. It is written into a profile only when the game data ships no config of its own, and a value seeded that way would shadow the install underneath it for good, which is exactly the snapshot the layered read exists to avoid. Absent reads as UNKNOWN_VERSION, which is what it declared before.

Demo: 1 in the demo's config is the clearest case. It looks like the flag that selects demo behaviour, but the demo is a compile-time build (-DDEMO) and the key is inert. The demo config also has no Version key, which is consistent: DistribLogo under DEMO skips the switch entirely and shows resource index 6, which its comment calls the three-distributor bumper.

Version numbers that are not release identity

NUM_VERSION, the save layout revision

The first byte of a .lba save. Low 7 bits are NUM_VERSION (36), the high bit is SAVE_COMPRESS. Constants at COMMON.H:168-170. It versions the serialisation layout and has nothing to do with which release wrote the file. Fully covered in SAVEGAME.md; the engine-version-versus-NUM_VERSION distinction is in RELEASING.md.

The two collide by name and by habitat. NumVersion and DistribVersion are both globals with version in the name, and both are partly config-driven from the same ReadConfigFile, since CompressSave folds into NumVersion a hundred lines above where Version is read. They are unrelated.

SAVE_VERSION

#define SAVE_VERSION 4 at DEFINES.H:209. Nothing references it. The matching comment at SAVEGAME.CPP:516, "Version 4 : LBA II only !!", suggests it records which generation of Adeline's save routine this file implements rather than anything the code decides on.

XPL_HEADER.Version

First field of the .XPL palette and shading header (COMMON.H:76). The header is mapped over the loaded file at AMBIANCE.CPP:457 and its offsets are used, but the Version field is never read. A format version the loader chose not to check.

The engine build version

Version the string, in VERSION.CPP, is the boot banner and the --version output. It is a port addition: the string is composed from LBA2_PRODUCT_NAME and LBA2_VERSION_STRING, both generated by CMake from git describe. See RELEASING.md.

The name collides with the Version config key and with DistribVersion's comment header, which is worth knowing when grepping. The two are kept in separate files for the same reason: VERSION.CPP is the build string and nothing else, and the publisher names live in DISTRIB.CPP.

Build variants

For completeness, the release axes that are compile-time rather than config-driven.

Macro	Set by	Effect
CDROM	SOURCES/CMakeLists.txt:149	Compiles the CD paths, so the volume-label and voice-folder branches are live
DEMO	LBA2_BUILD_DEMO CMake option (default OFF)	The 1997 playable demo: replaces DistribLogo's switch, adds the demo bumper, forces the corner logo on, and swaps free saves for three fixed slots. Needs demo game data; see GAME_DATA.md
DEBUG_TOOLS	CMake option	Skips the whole boot logo sequence, so no splash regardless of DistribVersion
DEBUG_TOOLS, TEST_TOOLS	CMake options	Enable the stale-save fallback to LoadGameOldVersion at OBJECT.CPP:1373
LBA_EDITOR, EDITLBA2	editor build	Compiles the editor paths out of the game build

DEFINES.H still carries the 1997 comments for these as commented-out #defines (DEFINES.H:7-17), including "Cdrom Version si pas define version disk". The build now defines CDROM externally, so the commented-out line in the header is misleading if read on its own.