# The releases whose mods this build must still run, and the differences
# from them that were looked at and accepted (tools/pc/check_mod_abi.py).
#
#   baseline TAG [TARGET]         check this build against release TAG's mods
#                                 for TARGET: i386 (the default, the 32-bit
#                                 games' objects) or x86_64-windows (the 64-bit
#                                 Windows game's, from the first release that
#                                 ships them; until then that game is only held
#                                 to refusing the i386 baselines' code mods by
#                                 name)
#   sha256 TAG SYSTEM DIGEST      the package of TAG the check may unpack and
#                                 run (windows: the .zip, linux: the .tar.gz,
#                                 windows-x64: the 64-bit .zip, which an
#                                 x86_64-windows baseline needs pinned too),
#                                 as GitHub's release API lists its digest:
#                                 gh api repos/Unchiga/Yu-Gi-Oh-Forbidden-Memories-Recompiled/releases/tags/TAG --jq '.assets[] | .name + " " + (.digest | ltrimstr("sha256:"))'
#   accept TAG KIND NAME reason   a difference that breaks no mod in use
baseline v0.1.2
sha256 v0.1.2 windows a6d5ddc86ad1de3fac263298b070224d990684ac221f1b55be60246186faf558
sha256 v0.1.2 linux 2ef326fad001284a6dddbc1cfc33561a16d61d8656a180a586eba42275abc1e9
baseline v0.2.0
sha256 v0.2.0 windows bf1db67397f60f6f12b845c21ce82cef3051e6b67d4f82e284e8c9bf359824ab
sha256 v0.2.0 linux c6dfa734174ef1432ba5cb4b085f3a59a6d4e53bdd4763db249f2e4c897fde68

# The Cheats menu's functions now say whether they did anything (0 before a
# game is loaded); a caller built against v0.1.2's void ignores eax.
accept v0.1.2 function Cheats_GiveAllCards returns int where it returned void
accept v0.1.2 function Cheats_UnlockAllFreeDuelists returns int where it returned void
# The port SDK stand-ins compile against the game's Psy-Q headers: libgte.h
# now declares RotAverageNclip3_nom with the return the resident routine and
# its stand-in always had (the GTE flag word, in v0/eax). A caller built
# against v0.1.2's void ignores eax, which every call may change anyway.
accept v0.1.2 function RotAverageNclip3_nom returns long (the flag word) where it returned void
# PR #105 (the burning card keeps its HD pack image): the recorder gained
# `capture` at its end. Only the port's own OpenGL picture installs a
# recorder; none of v0.1.2's mods calls SoftGpu_SetRecorder.
accept v0.1.2 record SoftGpuRecorder grew by one callback at its end
# The Game list (host actions: Exit game, save states, the other shortcuts):
# every controls profile gained its host bindings at its end, which moves
# what follows a profile. No v0.1.2 mod imports a Controls or
# ControlsRuntime name; the controls file is the port's own and has its own
# version (controls.txt 2).
accept v0.1.2 record ControlsProfile grew by the host action bindings at its end
accept v0.1.2 record ControlsDeviceProfile holds the larger ControlsProfile
accept v0.1.2 record ControlsConfig holds the larger ControlsProfile
# PR #231 (the fusion helper asks a mod's equip bonus): the planner's rules
# gained `bonus` at their end. Only the port's own fusion helper builds a
# FusionRules; none of v0.1.2's mods calls a Fusion_ function.
accept v0.1.2 record FusionRules grew by two callbacks at its end (bonus, bonuses)
# Equips with their own DEF bonus ("bonus_defense"): the duel record's unused
# bytes 0x1A-0x1B hold defense_modifier (offsets and size unchanged; the
# disc never used them), and the planner's card gained defense_modifier at
# its end, which FusionLine holds. None of v0.1.2's mods calls a Fusion_
# function, and a mod reading a record's stat_modifier still reads it.
accept v0.1.2 record DuelCardRecord names two of its pad bytes defense_modifier
accept v0.1.2 record FusionCard grew by defense_modifier at its end
accept v0.1.2 record FusionLine holds the larger FusionCard

# v0.2.0 (PRs #237-#291 since then). The same RotAverageNclip3_nom return
# as v0.1.2 above: v0.2.0's SDK still declared it void.
accept v0.2.0 function RotAverageNclip3_nom returns long (the flag word) where it returned void
# PR #279 (equips with their own DEF bonus), as for v0.1.2 above: the duel
# record's unused bytes 0x1A-0x1B hold defense_modifier, the planner's card
# grew by it at its end, and the rules by `bonuses` at theirs. Only the
# port's own fusion helper builds these; no v0.2.0 mod calls a Fusion_
# function.
accept v0.2.0 record DuelCardRecord names two of its pad bytes defense_modifier
accept v0.2.0 record FusionCard grew by defense_modifier at its end
accept v0.2.0 record FusionLine holds the larger FusionCard
accept v0.2.0 record FusionRules grew by the bonuses callback at its end
# PR #265 ("image_style": "full"): Pack grew by image_style at its end, so a
# mod reading Packs_At() still finds every v0.2.0 field where it was. The
# pack table is the port's own; a mod never allocates a Pack.
accept v0.2.0 record Pack grew by image_style at its end
# PR #297 (the title menus' buttons take a "scale"): the title screen's own
# label maker takes the button's factor, and its config and image tables
# grew. Mods set these through the "title" and "menu" keys; no mod calls
# MenuLabel_Make or reads a TitleConfig, which are the port's title code's.
accept v0.2.0 function MenuLabel_Make takes the button's factor after selected
accept v0.2.0 record TitleConfig holds the larger TitleItem
accept v0.2.0 record TitleItem gained scale before wide
accept v0.2.0 enum TITLE_IMAGES counts the scaled buttons' images
# A win's starchips up to 1000 ("starchip_prize"): the results record's
# starchip_prize byte and the unread pad byte after it are one halfword on
# the PC (offset 0x3A and size 0x40 unchanged). The disc's 1 to 8 read the
# same through the low byte, so a mod reading the old byte still sees them.
accept v0.1.2 record DuelResultDisplayState reads starchip_prize and its pad byte as one halfword
accept v0.2.0 record DuelResultDisplayState reads starchip_prize and its pad byte as one halfword
# American spelling: the title records' `colour` fields are `color` (same
# offset, type and size; a mod built against v0.2.0 still reads them). The
# enumerators TITLE_BACKGROUND_COLOUR and FONT_ART_COLOURS stay as aliases.
accept v0.2.0 record TitleBackground names its colour field color
accept v0.2.0 record TitleLine names its colour field color
# The export lists of v0.1.2 and v0.2.0 (and of every build until the
# per-target lists) held "name": build_mod.py read the C library list out
# of all of mod_libc.c, where F(name) and AS(name, function) are the
# macros' parameters. The game never lent anything under it (Mods_Lookup has no
# such entry), so no mod can have used it.
accept v0.1.2 export name was the C library macros' parameter, never a name the game lent
accept v0.2.0 export name was the C library macros' parameter, never a name the game lent
