Skip to content

Legacy 1.19.2 Estimate

github-actions[bot] edited this page Nov 16, 2026 · 1 revision

Estimate: legacy 1.19.2 Forge mods on NeoForge 1.21.1

Research note, 2026-11-16. Scoping only: no 1.19.2 support exists in the layer. The roadmap (Version-Roadmap) plans 1.19.2 as "renames and diffs over the 1.20.1 pipeline" (stage C of the 1.16 to 1.19 plan) at "1 to 2 days". This page replaces that guess with a measurement: 451 real Forge 1.19.2 jars were run through the 1.20.1 pipeline as it is today, and what was left was classified. Back to Home. Model: Legacy 1.20.1 estimate. Data: tools/dev/agent-tasks/1.19.2-symbols.md (top 300 symbols), 1.19.2-gap-jars.tsv (per jar), 1.19.2-gap-summary.txt (all counts).

Verdict

  • The 1.20.1 pipeline links 50 of 428 Forge 1.19.2 jars with code (11.7%; 35 fully, 15 more except for other mods that are not in the sample). The same pipeline links 141 of 506 (28%) real 1.20.1 jars (ATM 9 run). The difference is the 1.19.2 to 1.20.1 API delta, and it is not renames: of 2,406 distinct unresolved Minecraft/Forge symbols, 165 (7%) are mechanical (one table change), 992 (41%) need real adapters (changed or removed API), 737 are gaps the 1.20.1 layer has too, 283 are Forge API, 196 are unmapped members.
  • Doing all of the 1.19.2 work raises the link rate from 11.7% to 29.7% (127 jars), which is the ceiling the 1.20.1 layer itself has on this kind of sample. Every further jar is 1.20.1 work.
  • The roadmap's "1 to 2 days" holds for stage 1 below (tables only, 50 to 72 jars). Honest estimate: 6 to 8 days wall-clock with about five agents (roughly 22 to 28 agent-days) to reach the 1.20.1 link rate; 10 to 14 days to an Alpha that matches the 1.20.1 "Current state" (server suites reach "Done"); client GUI parity another week. Plus or minus 40%. It depends on 1.20.1 progress, because every 1.20.1 gap is also a 1.19.2 gap (class c).

1. Method

All offline; no game, server or client was started. The default build needs and fetches nothing of 1.19.2; every piece below is opt-in.

  1. Sample. python tools/compat/fetch_probe_mods.py 1.19.2 <dir> --limit 400 --content (Modrinth, Forge files, most downloaded content mods, plus required dependencies, SHA-1 checked), into a folder outside the repository; never committed.
  2. SRG to Mojang table of 1.19.2. ./gradlew :loader:genSrgMojang1192 -Pexperimental.mc1192=true. The same chain as 1.20.1 (MCPConfig joined.tsrg of 1.19.2-20220805.130853 joined with Mojang's client.txt by obfuscated owner, name and descriptor), parameterised by version in loader/build.gradle (srgMojangTables). Mojang's file is fetched at build time and its SHA-1 verified (8e8c9be5dc27802caba47053d4fdea328f7f89bd), SHA-256 fingerprints and the row count are recorded in build/reports/mappings/1.19.2/srg-mojang-provenance.json; neither Mojang's file nor the derived table is committed or added to the loader jar. 59,186 rows (44,929/44,929 methods and 28,897/28,897 fields matched). Without the property nothing is declared, resolved or fetched.
  3. Gap tool. ./gradlew gap1192 -Pexperimental.mc1192=true -PtargetJars=@list -PgapJars=<dir> is Gap1201 registered a second time with the 1.19.2 table (registerGap in build.gradle); -Pgap.srgOverlay=<1.20.1 table> lets a second table win. Three runs: A raw (1.19.2 table, 1.20.1 to 1.21.1 renames, no translation), B like A with the 1.20.1 table overlaid, C ./gradlew gap1201 -Pgap.translate=true, i.e. the real Forge1201Transformer (call rewrites, stand-ins, facades, soft links) over the 1.19.2 jars as if they were 1.20.1 jars. C is the measurement; A and B only show what the tables alone do.
  4. Mechanical diff. tools/dev/mc_member_diff.py joins the Mojang-named class and member lists of 1.19.2 and 1.20.1 by the stable SRG ids (class ids C_n, m_n_, f_n_ are kept by MCPConfig across 1.19.2 and 1.20.1).
  5. Classification. tools/compat/gap1192_analyze.py <detail.tsv> <jars> <diff> <1.19.2 table> <forward/1.20.1> <out> (classes below, shapes, what-if and staged curves). tools/compat/data1192_extra.py and the existing data1201.py scan the jars' data/.

The 1.19.2 mods are in SRG names (m_46859_) with Mojang class names, as 1.20.1 mods are, so no step of the 1.20.1 pipeline is missing: it only knows 1.20.1.

2. Sample

Jars 451 (399 most downloaded content mods that have a 1.19.2 Forge file, 52 required dependencies, 1 already present; 400 projects)
With META-INF/mods.toml (Forge) 447; 55 also carry fabric.mod.json or quilt.mod.json (multi-loader), 10 also a neoforge.mods.toml; 4 jars have no mods.toml (2 Fabric only, 2 with neither marker)
With code 428 analysed after translation (19 are data or resource packs, 4 failed to translate: ASM MethodTooLargeException in huge animation <clinit> methods, Cataclysm, EeeAbsMobs, Friends and Foes: a generic 1.20.1 defect, not 1.19.2)
Mixins 297 jars have a mixin config, 3,605 mixin classes (see risks)
Data pack format pack_format of the jars: 9 (207), 10 (67), 8 (62), 6 (40) and 15 (30: multi-version jars)
Libraries they link against JEI (90 jars), Curios (42), Create (35, with Flywheel 26 and Registrate 26), REI (23), Puzzles Lib (21), Cloth Config (21), Architectury (21), Jade (21), Farmer's Delight (20), Patchouli (10), Blueprint (10), TheOneProbe (9), EMI (7)

Not representative of the "abandoned" set the way the 1.20.1 corpus was (that one counted mods without a 1.21.1 port); this is the most downloaded content mods, so many have a 1.21.1 port. The numbers measure the layer, not the demand.

3. Link numbers

Run What it shows Jars with code Link fully Link but for other mods/libraries Unresolved
A: 1.19.2 table, renames only the tables alone 431 26 10 395
B: A plus the 1.20.1 table overlaid stable ids named as in 1.20.1 431 26 11 394
C: the 1.20.1 layer as it is what a 1.19.2 jar gets today 428 35 15 378
(ATM 9, real 1.20.1 jars, same layer) for comparison 506 141 (28%)

Unresolved symbol occurrences of run C by area (jar times symbol): Minecraft common 7,982, Minecraft client 4,701, Forge API 693, Mixin/ASM 1, other mods and libraries 31,215 (missing from the sample because it is the most-downloaded set, not a closed pack; they say nothing about the layer). Fabric intermediary names in the multi-loader jars (2,176 symbols) are dropped: that code is never loaded.

Do the stable ids carry over? Yes, and it helps. 24,663 of 30,289 methods and 26,779 of 28,897 fields that have a 1.19.2 id are identical in 1.19.2 and 1.20.1; 1,055 methods and 6 fields are renamed under the same id (the 1.20.1 table names them with the 1.20.1 name, so a 1.19.2 jar run through the 1.20.1 table gets the rename for free). Only ids that 1.20.1 does not know need the 1.19.2 table (removed members, and the classes 1.20.1 stopped obfuscating, see class a).

4. Classification of the unresolved Minecraft/Forge symbols (run C, 2,406 distinct, 13,377 jar x symbol)

Class Meaning Distinct symbols Occurrences Is it 1.19.2 work?
a one table row or one rename fixes it 165 1,845 yes, mechanical
b API changed shape or was removed after 1.19.2 992 8,858 yes, the work
c still there, unchanged, in 1.20.1: the 1.20.1 layer lacks it too 737 1,279 no (1.20.1 work)
F Forge API (Forge 43 to 47, Forge to NeoForge) 283 694 partly; not diffed here
u member of a class that exists in both, in neither list (synthesised by the layer or not mapped) 196 437 no, noise
n not a Minecraft member 33 264 no

4.1 Class a: renames and moves between 1.19.2 and 1.20.1 (derived mechanically)

mc_member_diff.py over the two Mojang-named versions (members that have an SRG id; the rest of the file is in members-a/b):

Same Renamed Moved Signature changed Removed Added
Classes 6,537 0 (by id) 0 322 899
Methods 24,663 1,055 152 200 4,219 7,152
Fields 26,779 6 0 50 2,062 4,168
  • Classes are not renamed, they are replaced. The class ids change when a class is rebuilt, so removed and added classes only pair by simple name: 47 pairs, 17 of them top-level, and most of those are just id churn (GLX, RenderSystem, MinecraftServer). Genuine moves found in the sample: AtlasSet, Resource$IoSupplier to IoSupplier, ChunkHolder$FullChunkStatus to FullChunkStatus, Registry$RegistryBootstrap and BuiltinRegistries$RegistryBootstrap to RegistrySetBuilder$RegistryBootstrap, the TextureAtlasSprite inner classes to SpriteContents: about ten rows. The 127 removed top-level classes are 1.19.3 chat signing and preview (ChatPreview*, MessageSigner, ChatSender, 21 classes), Material/MaterialColor, com.mojang.math (Matrix3f, Quaternion, Vector3d...), GuiComponent, LootTables/PredicateManager/ItemModifierManager, UpgradeRecipe, Wearable, the Alternative loot condition: those are class b.
  • The big class-a item is the SRG table itself. 1.19.2 obfuscates com.mojang.blaze3d (and other "official" packages) with SRG ids (RenderSystem.m_157456_ = setShaderTexture); 1.20.1 keeps the plain name, so the 1.20.1 table has no row for it and run C leaves 145 distinct symbols (1,769 occurrences, 193 jars: RenderSystem.setShaderTexture/setShaderColor/setShader/enableBlend..., BufferBuilder, VertexConsumer) as bare ids. Using the 1.19.2 table underneath the 1.20.1 one (the 1.20.1 names win) fixes all of them: the Gap1201 -Pgap.srgOverlay run proves it. This is the cheapest and largest single item.
  • Members renamed or moved under the same id (1,207) are fixed by the 1.20.1 table already when the id is looked up there: the only 1.19.2-specific part is "look up the 1.19.2 table second".

4.2 Class b: changed shape, by shape (distinct symbols / occurrences / jars; run C)

Shape Symbols Occurrences Jars Adapter in the style of v1_20_1/migrate? Size
Registry statics: Registry.ITEM/BLOCK/..., *_REGISTRY keys, BuiltinRegistries (1.19.3) 162 1,821 267 Yes, pure rows: GETSTATIC to BuiltInRegistries/Registries fields of the same name, generated from the class diff 0.5 day
GUI drawing: Screen.render(PoseStack,..), GuiComponent.blit/fill, Font.draw(PoseStack,..), Button(...), RenderSystem.enableTexture (1.19.3 to 1.20) 70 990 186 Yes, but not rows: bridge render(PoseStack,..) overrides to render(GuiGraphics,..) and a thread-local current-GuiGraphics helper for the statics, a Button stand-in over the protected 1.20.1 constructor 2 to 3 days, client-only
Damage sources: DamageSource.X statics, new DamageSource(String), causeXDamage (1.19.4 data driven) 76 870 132 Statics become level.damageSources().x(); new DamageSource(id) needs a generated damage_type data entry per id (the recipes mechanism) 1 to 1.5 days
com.mojang.math Vector3f/Quaternion/Matrix3f/Matrix4f to JOML (1.19.3), PoseStack.mulPose(Quaternion), Pose.pose(), VertexConsumer.vertex(Matrix4f,..) 21 632 209 Yes: legacy-API stand-in classes with conversion at the PoseStack/VertexConsumer boundary (precedent: CompatBufferBuilder) 1 to 1.5 days
Creative tabs: Item.Properties.tab(...), new CreativeModeTab(String), makeIcon, fillItemList (1.19.3 to 1.20) 63 (+126 occurrences of the constructor counted as n) 556 215 Yes: stand-in tab class registered as a real tab, tab(x) calls recorded by a helper and replayed into BuildCreativeModeTabContentsEvent (the layer already replays tab entries for 1.20.1) 1 to 1.5 days
Rendering: ItemTransforms.TransformType to ItemDisplayContext, AtlasSet, TextureAtlasSprite, model baking, TextureStitchEvent 89 689 188 Mixed: enum map (rows) plus a few helpers 1.5 to 2 days, client
Other Minecraft/server: BlockPos.<init>(DDD), BlockPos(Vec3), Registry helpers, constructor to factory 69 545 204 Yes: constructor-to-factory rows 0.5 to 1 day
Entities/attributes/AI 102 299 104 Rows plus a few bridges 1 day
Block/item/state 44 345 113 Rows 0.5 day
Client other (resources, keys, gui components) 68 322 126 Rows, Widget/AbstractWidget stand-ins 1 day
Sound events: SoundEvent.<init>(ResourceLocation) to createVariableRangeEvent 15 162 116 Yes: constructor-to-factory rows 0.25 day
Loot: LootTables, PredicateManager, LootContext.Builder, Alternative condition 47 369 87 Rows plus helpers (1.20.1 to 1.21.1 reload stand-ins of the 1.20.1 layer apply afterwards) 1 day
Level/world/chunks 32 198 95 Rows 0.5 day
Recipes/smithing: UpgradeRecipe, SmithingRecipe, Ingredient 28 277 76 Class stand-in plus data conversion (below) 1 day
Material, MaterialColor, BlockState.getMaterial (1.20) 3 symbols but the constants are fields 302 203 Yes: stand-in Material constants carrying map colour, push reaction, replaceable, flammable; BlockBehaviour.Properties.of(Material) helper 1 to 1.5 days
Worldgen: StructureFeature, Feature, BiomeLoadingEvent, WorldgenRandom 36 170 64 Rows, plus the event as in the 1.7.10 bridge to biome modifiers 1 day
Data generators: DataGenerator.addProvider, RecipeProvider.<init>(DataGenerator), TagsProvider 29 72 35 Cheapest by stripping: a GatherDataEvent listener never runs in game 0.25 day
Explosions, commands (sendSuccess(Component, boolean) to a supplier), network packets, chat (1.19.3 signing) 7 + 9 + 17 + 5 113 + 83 + 34 + 9 46 + 73 + 24 + 6 Rows and a few helpers 0.5 to 1 day

The sum is about 15 to 19 agent-days for class b alone. Chat signing and previews (21 removed classes) and Material are the only areas without a mechanical mapping; both are small in the sample (chat 6 jars).

4.3 Class c: gaps the 1.20.1 layer has too (not 1.19.2 work)

737 distinct symbols, 1,279 occurrences: a long, thin tail (most symbols occur in one or two jars). It overlaps tools/dev/agent-tasks/1.20.1-symbols-leftovers.md (14 of the entries match it by name: static animal food fields, BannerBlockEntity, CauldronInteraction, ClientboundExplodePacket, LootingEnchantFunction...); the larger items are SoundEvents.UI_BUTTON_CLICK (40 jars; now a holder), SkinManager, registryAccess() on LevelAccessor/WorldGenLevel, the custom payload packet constructors (ClientboundCustomPayloadPacket(ResourceLocation, FriendlyByteBuf)), StateSwitchingButton, and 95 data-generator symbols. These do not move with the 1.19.2 work; they are why the ceiling is 29.7% and not higher.

4.4 Forge API (class F: 283 symbols, 694 occurrences)

Small. Top: TextureStitchEvent.Pre (46 jars; removed in 1.19.3, TextureStitchEvent no longer has Pre/Post), RegisterParticleProvidersEvent.register (42 and 13 jars for the two overloads), NeoForgeMod.REACH_DISTANCE/ATTACK_RANGE (now holders), LivingSpawnEvent.CheckSpawn/SpecialSpawn, CapabilityItemHandler/CapabilityFluidHandler/CapabilityEnergy (the capability constants), the getPoseStack() accessors of RenderGuiOverlayEvent/ScreenEvent.Render (which follow the GUI work), BiomeGenerationSettingsBuilder.addFeature, RenderLevelLastEvent, RegistryEvent (8 jars: most 1.19.2 mods already use DeferredRegister). A proper Forge 43 to 47 diff needs the Forge jars, which this scoping did not fetch; it is stage 0 below. Same size as the 1.20.1 Forge facade round for the removed-in-1.19.3 events (about 1 day).

4.5 Data and resources (sample scan of the jars' data/ and assets/)

tools/compat/data1201.py (the 1.20.1 to 1.21.1 formats) over the 451 jars: plural folder names 124,537 files in 303 jars (1.19.2 uses plural folders as 1.20.1 does: already converted), recipe results 33,075 files/203 jars, forge: tags 11,327/162, forge: ids 10,376/106, recipe conditions 9,922/52, item predicates 5,202/116, advancement icons 2,036/89, conditional recipes 1,923/17, loot functions 1,514/99, model loaders 1,033/26, forge: directories 572/84, renamed ids 195/55, NBT in recipes 143/14, loot conditions 108/22. These are what DataConverter1201 converts today; they are the same for 1.19.2 data.

tools/compat/data1192_extra.py (what is new for 1.19.2): old smithing recipes (minecraft:smithing, 250 files in 34 jars; since 1.19.4 smithing_transform needs a template item, only 1 file already uses it): the converter has to pick a template (the netherite upgrade template as vanilla's own conversion does, or a generated invisible one) and keep the recipe usable; damage-type tags/json 36 files in 7 jars (new in 1.19.4: only matter once damage sources are mapped); pack_format 9 and 10 of the jars (what 1.19.2 shipped) against 15 (1.20.1) and 34 (1.21.1): the pack format check is already bypassed for translated jars; alternative loot condition (262 files, 32 jars; minecraft:alternative is any_of since 1.20); no feature flags and no chat types in the sample. Atlas rules (1.19.3: loose textures are invisible unless listed in the atlas) may hide block or item textures of mods that rely on directory stitching; the converter must list them (the 1.12.2 layer does this for code-referenced sprites). Size: 1 to 1.5 days including tests; small next to the 1.20.1 data work because the folder, recipe and forge: conversions already exist.

5. Risks

  1. Mixins. 297 of 451 jars carry a mixin config (3,605 classes), more than the 1.20.1 samples: GUI, Screen, ItemRenderer, PoseStack and Material targets all changed in 1.19.3 to 1.20. The layer's per-injection repair and drop applies, but a dropped GUI mixin silently removes a feature. The 1.20.1 mixin tooling covers 1.20.1 targets only; the 1.19.2 index of targets (a fourth jar to fetch) is part of stage 0.
  2. GUI and rendering parity is visual, cannot be proven offline, and the client path needs real screenshots (a week of client verification, as the 1.12.2 render sweep).
  3. Semantic stand-ins (Material, creative tabs, damage sources) are approximations: Material held behaviours that are now spread over block properties, tabs held an item order, damage sources had no registry. Expect a defect tail per mod family.
  4. Libraries are 1.19.2 builds (Create 0.5, Curios 5, Jade 8, Farmer's Delight 1.2, Patchouli, Architectury 6/8): the 1.20.1 bridge rules apply only where the 1.19.2 API equals the 1.20.1 one; each library needs a check of its 1.19.2 class list (JEI mezz.jei 90 jars and Curios 42 are the big two; the 1.20.1 JEI/Curios bridges are the starting point, not the answer).
  5. Dependence on 1.20.1. The class c tail and the 1.20.1 data converters are shared; progress on 1.20.1 raises the 1.19.2 ceiling for free, and a 1.20.1 regression breaks both.
  6. Transformer version. Every loader output change needs a TRANSFORMER_VERSION bump and invalidates the caches; the 1.19.2 mode must be keyed so a 1.19.2 jar and a 1.20.1 jar never share a cache entry.

6. Staged plan

Curves are measured on the sample: "link" = jars with no unresolved Minecraft/Forge symbol left, "cleared" = jars with no remaining 1.19.2-specific (class a/b) symbol (373 of 428 jars have at least one). Baseline today: 50 link. Ceiling with all stages: 127.

Stage Content Unlocks (cumulative) Days
0. Plumbing LegacyVersion.V1_19_2 detection (mods.toml loaderVersion [43,), minecraft [1.19.2]), a cache key per version, the 1.19.2 table as a build-time resource overlay under the 1.20.1 one (same licence rule as 1.20.1: derived at build time, not committed), Forge 43 to 47 and Forge to NeoForge diff of the Forge facades (TextureStitchEvent, RegistryEvent, capability constants...), 1.19.2 mixin target index nothing visible; class a (SRG table) is part of it: 145 symbols in 193 jars 1.5 to 2
1. Tables registry statics, constructor-to-factory rows (BlockPos, SoundEvent), level/world/block/commands/network/chat rows, the ten real class moves 33 cleared, 72 link 1.5 to 2
2. Stand-ins creative tabs, Material, com.mojang.math, damage sources (with generated damage_type data), entities, loot 107 cleared, 92 link 5 to 6
3. Client GUI (GuiGraphics bridges), rendering, client other 245 cleared, 118 link 5 to 6
4. Data and worldgen smithing conversion, recipes, worldgen rows, datagen stripping, atlas listing 373 cleared, 127 link (the ceiling) 3 to 4
5. Run time server suites with 1.19.2 content mods until "Done", verify.sh suite, mixin triage, first client scenarios what links runs open; 1.20.1's equivalent took about a week of rounds

The order inside stages is by jars unlocked, not by size; stage 2 and 3 parallelise (different agents, different files). ./gradlew gap1192 and gap1192_analyze.py are the progress meter: re-run, compare the staged curve.

7. Effort and pace

Pace of the 1.20.1 work (git log --since=2026-11-05 -i --grep="1.20.1"): 253 commits on 9 active days (2026-11-06: 4, 11-07: 65, 11-08: 62, 11-09: 7, 11-13: 33, 11-14: 29, 11-15: 45, a few days of nothing in between). The migration tables, Forge facades and first server loads came in about two days (131 commits, 11-06 to 11-08); the ATM 9 link rate went from the first measurement to 28% over the rounds of 11-13 to 11-15. The 1.20.1 corpus was a full port from nothing; 1.19.2 reuses the transformer, the gap tool, the data converter, the soft link, the mixin repair, the facades and the library table, so the first link-rate is cheaper than 1.20.1's first week, but it is about 1,000 distinct symbols of changed API, not a table.

Milestone Wall-clock with about 5 agents Agent-days
Stage 0 + 1 (tables; link 50 to 72) 1.5 to 2 days (the roadmap's figure) 3 to 4
Stage 2 (stand-ins; 92 link) +2 days 8 to 10
Stages 3 + 4 (client, data; 127 link, equal to the 1.20.1 layer) +3 to 4 days 12 to 14
Link parity with 1.20.1 6 to 8 days 22 to 28
Alpha (server suites reach "Done", regression suite, mixin triage) +4 to 6 days +10 to 14
Client verification of GUI and rendering +about 1 week, partly parallel +10
Long tail (issue list) open-ended

Plus or minus 40%. This is about 4 to 5 times the roadmap's "1 to 2 days": the 1.20.1 estimate also began with a "renames and diffs" intuition, and the measured API change was larger. Recommendation: do stage 0 and 1 when 1.20.1 stabilises (cheap, shows a real number, and its two days are the roadmap's guess), re-measure with the same scripts, then decide on stage 2 to 4 with the abandoned-mod share of the 1.19.2 corpus (this sample is the most downloaded, not the abandoned set; fetch with --slug for a pack).

8. What 1.18.2 adds on top (Forge 40, Java 17, same as 1.19.2)

Stage B of the plan (1.18.2 to 1.19.2) is a second, independent delta; it was not measured. Expected from the API history: TextComponent/TranslatableComponent to Component factories (constructor-to-factory rows, cheap); Random to RandomSource in most game methods (override bridges both ways, the most error-prone item); the Forge 1.19 event renames (World to Level), ForgeRegistryEntry/setRegistryName removal (a registry-name helper), FluidAttributes to FluidType; registry freezing and holders: 1.19 made vanilla registries intrusive-holder based and froze them earlier, so code that registers late or reads Registry.X.get(...) before the freeze behaves differently (the 1.20.1 layer already has a tolerant registry for this); worldgen: StructureFeature to Structure, configured structures and biome data formats, BiomeLoadingEvent (alive in 1.19.2, gone later); chat: 1.19 signed chat and sendMessage(Component, UUID) to sendSystemMessage. Class names and SRG ids of 1.18.2 are Mojang names and m_ ids, so the id-join should work for 1.18.2 to 1.19.2 as well (to be verified with the 1.18.2 MCPConfig; the 1.16.5 step cannot use it: those ids were renumbered). Expect about 60 to 70% of the 1.19.2 effort again, with a larger share of worldgen and data format conversion and a smaller share of GUI and rendering (the 1.19.3/1.20 GUI delta would already be solved by stage 3), i.e. 4 to 5 days wall-clock after 1.19.2, and the same measure-first rule: fetch the 1.18.2 sample and the 1.18.2 table with the same opt-in switch (srgMojangTables('1.18.2', ...)).

9. Reproducing

export JAVA_HOME="/c/Program Files/Java/jdk-21"
python tools/compat/fetch_probe_mods.py 1.19.2 C:/Users/Kyleb/probe-jars/1.19.2 --limit 400 --content
./gradlew :loader:genSrgMojang1192 :loader:genSrgMojang1201 -Pexperimental.mc1192=true
python tools/dev/mc_member_diff.py loader/build/tmp/mcpconfig1192/joined.tsrg loader/build/mojang-mappings/1.19.2-client.txt \
   loader/build/generated/srg-mojang-1192/legacyconnector/mappings/1.19.2/srg-mojang.tsv loader/build/tmp/mcpconfig1201/joined.tsrg \
   loader/build/mojang-mappings/1.20.1-client.txt loader/build/generated/srg-mojang/legacyconnector/mappings/1.20.1/srg-mojang.tsv <diff-dir>
./gradlew gap1201 -PtargetJars=@<targets.txt> -PgapJars=<jar-dir> -PgapReport=<out>/transC.txt -Pgap.translate=true      # run C (targets: probe.py gap1201_targets)
./gradlew gap1192 -Pexperimental.mc1192=true -PtargetJars=@<targets.txt> -PgapJars=<jar-dir> -PgapReport=<out>/rawA.txt  # run A (add -Pgap.srgOverlay=<1.20.1 table> for B)
python -I tools/compat/gap1192_analyze.py <out>/transC.txt.detail.tsv <jar-dir> <diff-dir> <1.19.2 table> loader/src/main/resources/legacyconnector/forward/1.20.1 <out>/ana
python -I tools/compat/data1201.py <jar-dir>; python -I tools/compat/data1192_extra.py <jar-dir>

Run C takes about 25 minutes for 451 jars (single JVM, 3 GB). docs/data/ is untouched by all of this.

Clone this wiki locally