From 4e690c0e336a4e360ba96f4edf9e8da2c6d2b343 Mon Sep 17 00:00:00 2001 From: Frotty Date: Sat, 15 Aug 2026 03:39:46 +0200 Subject: [PATCH 1/7] WIP: FastHashMap as a compiler-side proof MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Exercises type class bounds through the container they were added for. Six of the seven cases pass on both backends with no compiler change: int keys, tuple keys on Jass, a user class key, two specialisations coexisting, and two instances of one specialisation. tupleKeyLua fails and is a real, pre-existing backend bug. Method names become Lua table keys, but luaMethod.initFor passes the name through raw while every sibling (luaVar, luaFunc, luaClassVar) sanitises via uniqueName. Names are valid identifiers in ordinary code, so nothing hit it until a class method was specialised for Lua with more than one type argument: specializeMethod builds name + "_specialized_" + generics.makeName(), and makeName joins arguments with ", ". Two simple arguments give "get_specialized_integer, integer", which emits "Class.get_specialized_integer, integer = impl" -- valid Lua that assigns to two targets and quietly writes a junk global. A tuple argument gives "⦅integer, integer⦆" and fails the syntax check outright. --- .../wurstscript/tests/FastHashMapTests.java | 207 ++++++++++++++++++ 1 file changed, 207 insertions(+) create mode 100644 de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/FastHashMapTests.java diff --git a/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/FastHashMapTests.java b/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/FastHashMapTests.java new file mode 100644 index 000000000..50becb1f3 --- /dev/null +++ b/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/FastHashMapTests.java @@ -0,0 +1,207 @@ +package tests.wurstscript.tests; + +import org.testng.annotations.Test; + +/** + * Exercises type class bounds through the container they were added for: a hash map whose key type + * is bounded by {@code Hashable}, so it can hash and compare keys without erasing them to int. + *

+ * This is deliberately a whole working container rather than a minimal repro. It combines things + * nothing else tests together: two type parameters where only the first is bounded, static arrays of + * a bounded type parameter, and a bound used from a private method reached through a public one. + */ +public class FastHashMapTests extends WurstScriptTest { + + /** + * Storage follows ArrayList: one array per specialisation, carved into a section per instance. + * Collision handling is linear probing, so a lookup calls both requirements of the bound. + */ + private static String[] fastHashMap(String... extra) { + String[] head = { + "package test", + "native testSuccess()", + "constant int CAPACITY = 8", + "interface Hashable", + " function hash(T x) returns int", + " function equals(T a, T b) returns boolean", + "class FastHashMap", + " private static K array keys", + " private static V array values", + " private static boolean array used", + " private static int nextFree = 0", + " private int base", + " private int count = 0", + " construct()", + " base = nextFree", + " nextFree += CAPACITY", + " private function slotFor(K key) returns int", + " var i = K.hash(key) mod CAPACITY", + " if i < 0", + " i += CAPACITY", + " while used[base + i] and not K.equals(keys[base + i], key)", + " i = (i + 1) mod CAPACITY", + " return base + i", + " function put(K key, V value)", + " let s = slotFor(key)", + " if not used[s]", + " used[s] = true", + " keys[s] = key", + " count++", + " values[s] = value", + " function get(K key) returns V", + " return values[slotFor(key)]", + " function has(K key) returns boolean", + " return used[slotFor(key)]", + " function size() returns int", + " return count", + }; + String[] all = new String[head.length + extra.length]; + System.arraycopy(head, 0, all, 0, head.length); + System.arraycopy(extra, 0, all, head.length, extra.length); + return all; + } + + private static final String[] INT_INSTANCE = { + "implements Hashable", + " function hash(int x) returns int", + " return x", + " function equals(int a, int b) returns boolean", + " return a == b", + }; + + private static String[] program(String[]... parts) { + int size = 0; + for (String[] part : parts) { + size += part.length; + } + String[] all = new String[size]; + int at = 0; + for (String[] part : parts) { + System.arraycopy(part, 0, all, at, part.length); + at += part.length; + } + return all; + } + + /** Keys 1 and 9 land in the same slot at capacity 8, so the probe path is taken. */ + private static final String[] USE_WITH_COLLISION = { + "init", + " let m = new FastHashMap()", + " m.put(1, 10)", + " m.put(9, 90)", + " m.put(2, 20)", + " if m.get(1) == 10 and m.get(9) == 90 and m.get(2) == 20", + " if m.size() == 3 and not m.has(3)", + " testSuccess()", + }; + + @Test + public void fastHashMapRuntime() { + testAssertOkLines(true, program(fastHashMap(), INT_INSTANCE, USE_WITH_COLLISION)); + } + + @Test + public void fastHashMapRuntimeLua() { + test().testLua(true).executeProg().lines(program(fastHashMap(), INT_INSTANCE, USE_WITH_COLLISION)); + } + + /** + * A tuple key, which is the case old generics cannot serve at all: a tuple has no int + * representation to cast to, so the only way to key a map by one is to say how it hashes. + */ + private static final String[] TUPLE_INSTANCE = { + "tuple pos(int x, int y)", + "implements Hashable", + " function hash(pos p) returns int", + " return p.x * 31 + p.y", + " function equals(pos a, pos b) returns boolean", + " return a.x == b.x and a.y == b.y", + }; + + @Test + public void tupleKey() { + testAssertOkLines(true, program(fastHashMap(), TUPLE_INSTANCE, new String[]{ + "init", + " let m = new FastHashMap()", + " m.put(pos(1, 2), 12)", + " m.put(pos(2, 1), 21)", + " if m.get(pos(1, 2)) == 12 and m.get(pos(2, 1)) == 21", + " if m.has(pos(1, 2)) and not m.has(pos(9, 9))", + " testSuccess()" + })); + } + + @Test + public void tupleKeyLua() { + test().testLua(true).executeProg().lines(program(fastHashMap(), TUPLE_INSTANCE, new String[]{ + "init", + " let m = new FastHashMap()", + " m.put(pos(1, 2), 12)", + " m.put(pos(2, 1), 21)", + " if m.get(pos(1, 2)) == 12 and m.get(pos(2, 1)) == 21", + " testSuccess()" + })); + } + + /** A key type of the user's own, with the instance declared in the same package as the type. */ + @Test + public void classKeyWithUserInstance() { + testAssertOkLines(true, program(fastHashMap(), new String[]{ + "class Item", + " int id", + " construct(int id)", + " this.id = id", + "implements Hashable", + " function hash(Item i) returns int", + " return i.id", + " function equals(Item a, Item b) returns boolean", + " return a.id == b.id", + "init", + " let m = new FastHashMap()", + " let a = new Item(1)", + " let b = new Item(2)", + " m.put(a, \"a\")", + " m.put(b, \"b\")", + " if m.get(a) == \"a\" and m.get(b) == \"b\" and m.size() == 2", + " testSuccess()" + })); + } + + /** + * Two specialisations live at once. Storage is static per specialisation, so this is what would + * break if the arrays of one instantiation were shared with another. + */ + @Test + public void twoSpecialisationsCoexist() { + testAssertOkLines(true, program(fastHashMap(), INT_INSTANCE, new String[]{ + // Every string hashes alike, so this also exercises the probe path on every lookup. + "implements Hashable", + " function hash(string s) returns int", + " return 0", + " function equals(string a, string b) returns boolean", + " return a == b", + "init", + " let ints = new FastHashMap()", + " let strs = new FastHashMap()", + " ints.put(1, 100)", + " strs.put(\"a\", 200)", + " if ints.get(1) == 100 and strs.get(\"a\") == 200", + " if ints.size() == 1 and strs.size() == 1", + " testSuccess()" + })); + } + + /** Two maps of the same instantiation must not share storage either. */ + @Test + public void twoInstancesOfOneSpecialisation() { + testAssertOkLines(true, program(fastHashMap(), INT_INSTANCE, new String[]{ + "init", + " let a = new FastHashMap()", + " let b = new FastHashMap()", + " a.put(1, 10)", + " b.put(1, 20)", + " if a.get(1) == 10 and b.get(1) == 20 and a.size() == 1 and b.size() == 1", + " testSuccess()" + })); + } +} From 3ccbba245f652eb548ebfc047b6d0759ff79f7f3 Mon Sep 17 00:00:00 2001 From: Frotty Date: Sat, 15 Aug 2026 03:44:47 +0200 Subject: [PATCH 2/7] Add a working backlog for the type class bound follow-ups --- BACKLOG.md | 70 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 70 insertions(+) create mode 100644 BACKLOG.md diff --git a/BACKLOG.md b/BACKLOG.md new file mode 100644 index 000000000..753288587 --- /dev/null +++ b/BACKLOG.md @@ -0,0 +1,70 @@ +# Language feature backlog + +Working notes for ongoing work on type class bounds for `T:` generics +(shipped in #1226, #1228, #1229). Ordered: take the top unblocked item. + +Keep this file current. It is the only memory that survives between sessions. +When an item is finished, move it to Done with one line on what actually +happened. When something is learned that would have saved time, write it under +Notes rather than leaving it in a commit message. + +## Todo + +1. **Lua method names reach the backend unsanitised.** `LuaTranslator.luaMethod.initFor` + passes `a.getName()` raw; `luaVar`, `luaFunc` and `luaClassVar` all sanitise via + `uniqueName`. Method names become Lua table keys, so they must be valid identifiers. + `EliminateGenerics.specializeMethod` builds `name + "_specialized_" + generics.makeName()` + on the Lua path, and `makeName` joins type arguments with `", "` — so a class method + specialised with two type arguments emits `Class.get_specialized_integer, integer = impl`, + which is valid Lua assigning to two targets, and a tuple argument emits `⦅⦆` and fails + the syntax check. Repro: `FastHashMapTests.tupleKeyLua`. Commas are already visible in + `test-output/lua/FastHashMapTests_fastHashMapRuntimeLua.lua`. + Overriding methods must keep landing in the same slot, so normalise per distinct original + name, not per method node. + +2. **`slotFor` looks bound to `get`'s implementation** in the same emitted Lua. May be a real + mis-binding in `specializeMethod`/`adaptSubmethods`, may be an artefact of item 1 mangling + the output. Diagnose only after item 1, from freshly emitted Lua. + +3. **Finish the FastHashMap proof.** `FastHashMapTests` is the first real use of bounds. + Add `remove` with tombstones, and an assertion that the emitted code stays cheap: no + dispatch node, no instance dictionary, and no WC3 hashtable natives — array access only, + which is the whole point versus `HashMap extends Table`. + +4. **Lua dispatch inside a closure.** Works on Jass since #1229. On Lua the specialised class + is built correctly but nothing calls it, because the closure is reached through its + interface and `specializeMethod` renames the method out of its dispatch slot. + `TypeClassTests.dispatchInsideClosureIsRejectedForLua` pins the current diagnostic and + should become a success test. Related to item 1; AGENTS.md flags this machinery. + +5. **Lua dispatch inside the constructor** of a bounded generic class. Works on Jass. + +6. **Module bounds.** `module M` is rejected with a clear message today. Needs + receiver rewriting during expansion, or type parameters on `ModuleInstanciation`. + +7. **One `ImTypeVar` per type parameter.** Name-tolerant lookups remain in + `EliminateGenerics.indexOfTypeVar`, `inheritTypeClassBinding` and + `ProgramState.getCurrentTypeArgument`, compensating for several nodes standing for one + source parameter. Making the node canonical lets all three compare by identity and removes + a class of silent wrong dispatch. Mechanical, well covered by the suite. + +8. **Jass temp counter is not reset between compilations.** Two runs of the same commit emit + different `.j` (`temp151` vs `temp8`) because the counter is JVM-wide and depends on how + many tests ran before. Not wrong for compiling one map, but it means `.j` cannot be diffed + across runs to validate a change — only `.lua` can. Fixing it would make Jass diffable. + +## Done + +- Substitution now carries the type class binding with the type (#1229). Also fixed the + type-variable reference on `ImTypeVarDispatch`, which a walk over types alone missed. + +## Notes + +- `%` is real modulo in Wurst; `mod` is integer modulo. `int % 8` types as `real`. +- Emitted Lua must be byte-identical for identical input (AGENTS.md §8). It is the only + emitted output that can be diffed across runs — see item 8. +- Tests run five Jass configurations plus the interpreter, then the Lua target separately. + `testAssertOkLines(true, ...)` covers both the pre-transform interpreter and full + monomorphisation, so it is a stronger check than it looks. +- The stdlib copy under `de.peeeq.wurstscript/temp/WurstStdlib2` is a fetched artefact for + tests. Real stdlib changes belong in the WurstStdlib2 repo, not here. From 6fac1f027073e4225f76280351a3e7dbb65a996d Mon Sep 17 00:00:00 2001 From: Frotty Date: Sat, 15 Aug 2026 03:48:38 +0200 Subject: [PATCH 3/7] Expand the backlog with a recurrence guard, design blockers and scope --- BACKLOG.md | 59 +++++++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 49 insertions(+), 10 deletions(-) diff --git a/BACKLOG.md b/BACKLOG.md index 753288587..286550bec 100644 --- a/BACKLOG.md +++ b/BACKLOG.md @@ -22,36 +22,75 @@ Notes rather than leaving it in a commit message. Overriding methods must keep landing in the same slot, so normalise per distinct original name, not per method node. -2. **`slotFor` looks bound to `get`'s implementation** in the same emitted Lua. May be a real +2. **Nothing checks that emitted Lua identifiers are valid.** The luac syntax check catches a + hard break, but not the silent case: `Class.get_specialized_integer, integer = impl` parses + fine and quietly assigns to two targets. That is how item 1 shipped unnoticed. Assert in the + Lua test harness that every emitted name — variable, function, class, method slot, field — + matches `[A-Za-z_][A-Za-z0-9_]*`, so this whole class of bug fails loudly at the point it is + introduced. Do this alongside item 1; it is what stops item 1 recurring. + +3. **`slotFor` looks bound to `get`'s implementation** in the same emitted Lua. May be a real mis-binding in `specializeMethod`/`adaptSubmethods`, may be an artefact of item 1 mangling the output. Diagnose only after item 1, from freshly emitted Lua. -3. **Finish the FastHashMap proof.** `FastHashMapTests` is the first real use of bounds. +4. **Finish the FastHashMap proof.** `FastHashMapTests` is the first real use of bounds. Add `remove` with tombstones, and an assertion that the emitted code stays cheap: no dispatch node, no instance dictionary, and no WC3 hashtable natives — array access only, which is the whole point versus `HashMap extends Table`. -4. **Lua dispatch inside a closure.** Works on Jass since #1229. On Lua the specialised class +5. **Lua dispatch inside a closure.** Works on Jass since #1229. On Lua the specialised class is built correctly but nothing calls it, because the closure is reached through its interface and `specializeMethod` renames the method out of its dispatch slot. `TypeClassTests.dispatchInsideClosureIsRejectedForLua` pins the current diagnostic and should become a success test. Related to item 1; AGENTS.md flags this machinery. -5. **Lua dispatch inside the constructor** of a bounded generic class. Works on Jass. +6. **Lua dispatch inside the constructor** of a bounded generic class. Works on Jass. -6. **Module bounds.** `module M` is rejected with a clear message today. Needs +7. **Module bounds.** `module M` is rejected with a clear message today. Needs receiver rewriting during expansion, or type parameters on `ModuleInstanciation`. -7. **One `ImTypeVar` per type parameter.** Name-tolerant lookups remain in +8. **`MOD_INT`/`DIV_INT` return the left operand's type** rather than `int` + (`AttrExprType.java`, the `case MOD_INT` branch), where `caseMathOperation` returns + `WurstTypeInt.instance()` for `+`, `-`, `*`. Only observable if something is a proper + subtype of int, so it may be harmless — establish whether it is reachable, then either fix + it or leave a comment saying why the asymmetry is intended. Small. + +9. **Keep `WURST_LANGUAGE.md` and `CHANGELOG.md` current** as items land. The bounds section + says nothing about closures, which now work on Jass. Fold this into whichever item changes + the behaviour rather than doing it as a separate pass. + +10. **One `ImTypeVar` per type parameter.** Name-tolerant lookups remain in `EliminateGenerics.indexOfTypeVar`, `inheritTypeClassBinding` and `ProgramState.getCurrentTypeArgument`, compensating for several nodes standing for one source parameter. Making the node canonical lets all three compare by identity and removes a class of silent wrong dispatch. Mechanical, well covered by the suite. -8. **Jass temp counter is not reset between compilations.** Two runs of the same commit emit - different `.j` (`temp151` vs `temp8`) because the counter is JVM-wide and depends on how - many tests ran before. Not wrong for compiling one map, but it means `.j` cannot be diffed - across runs to validate a change — only `.lua` can. Fixing it would make Jass diffable. +11. **Jass temp counter is not reset between compilations.** Two runs of the same commit emit + different `.j` (`temp151` vs `temp8`) because the counter is JVM-wide and depends on how + many tests ran before. Not wrong for compiling one map, but it means `.j` cannot be diffed + across runs to validate a change — only `.lua` can. Fixing it would make Jass diffable. + +12. **Standing item, never finished.** When nothing above is left, find the next thing worth + doing and add it here rather than stopping. Good sources, in order: a test that would have + caught a bug already found; a place where two mechanisms do the same job and disagree; a + comment claiming something the code no longer does; a path where a wrong result is silent + rather than loud. Add what is found as a numbered item and start on it. + +## Blocked on a decision + +- **Eliminating the remaining `castTo int`.** The motivating case is timer data attachment + (`ClosureTimers.wurst`), and the containers behind it: `Table` has 81 casts, `HashList` 13, + `HashSet` 6, `HashMap` 4. None can adopt bounds as things stand, because an instance is + declared one type at a time and these accept any type. It needs a way to give an instance + for a whole family — every class type, or every handle type — which is a language design + question: what the syntax is, where such an instance may be declared under the orphan rule, + and whether a specific instance always beats a family one. Do not start this autonomously. + +## Out of scope + +- The stdlib itself. `de.peeeq.wurstscript/temp/WurstStdlib2` is a fetched artefact for tests; + editing it changes nothing real. `FastHashMap` ships from the WurstStdlib2 repo once the + compiler-side proof is complete, and that is a separate decision. ## Done From e4da96d7d5d032d07da28a51fc9be608f6b12bed Mon Sep 17 00:00:00 2001 From: Frotty Date: Sat, 15 Aug 2026 03:52:22 +0200 Subject: [PATCH 4/7] Add the overnight loop brief --- .claude/ralph-loop.local.md | 10 ++++++++ LOOP.md | 48 +++++++++++++++++++++++++++++++++++++ 2 files changed, 58 insertions(+) create mode 100644 .claude/ralph-loop.local.md create mode 100644 LOOP.md diff --git a/.claude/ralph-loop.local.md b/.claude/ralph-loop.local.md new file mode 100644 index 000000000..305f01377 --- /dev/null +++ b/.claude/ralph-loop.local.md @@ -0,0 +1,10 @@ +--- +active: true +iteration: 1 +session_id: 0413fc86-e9a2-4954-bc47-1620ed81d95e +max_iterations: 0 +completion_promise: null +started_at: "2026-08-15T01:51:05Z" +--- + +Work on the WurstScript compiler at c:Usersrun-eDocumentsGitHubWurstScript, on the branch feat/fasthashmap-proof. Read AGENTS.md and BACKLOG.md before doing anything. diff --git a/LOOP.md b/LOOP.md new file mode 100644 index 000000000..ed3234d69 --- /dev/null +++ b/LOOP.md @@ -0,0 +1,48 @@ +# Overnight loop brief + +Work on the WurstScript compiler in this repository, on the branch +`feat/fasthashmap-proof`. Read `AGENTS.md` and `BACKLOG.md` before doing anything. + +## Each iteration + +1. Read `BACKLOG.md`. Take the top item that isn't blocked. +2. Do it. Root-cause it — adjust the underlying system rather than patching a symptom. + If it turns out to be bigger than one iteration, split it in the backlog and do the + first part. +3. Verify. Targeted tests while iterating; the full suite before any commit that touches + main source: + + cd de.peeeq.wurstscript && ./gradlew test + + It must be green. Emitted Lua must stay byte-identical unless the change is meant to + alter it — compare two runs to check. Do not diff `.j` across runs; it is not stable, + and backlog item 11 explains why. +4. Commit and push. Small commits, one concern each. Never end an iteration with + uncommitted work. +5. Update `BACKLOG.md`: move finished items to Done with one line on what actually + happened, add anything learned to Notes, reorder if something more urgent turned up. + +## Rules + +- Never force-push, rewrite history, merge, push to master, or open a PR. The branch gets + reviewed in the morning. +- Commits are authored as the repository owner. No AI, assistant, or co-author references + anywhere in commit messages or code comments. +- Never stop to ask. If something needs a decision from the owner, write the question into + `BACKLOG.md` under that item, mark it blocked, and move to the next item. +- If the full suite goes red and it can't be fixed within the iteration, revert the change, + note why in `BACKLOG.md`, and move on. Leave the branch green. +- Comments explain why, not what. Match the surrounding style. +- Do not touch `de.peeeq.wurstscript/temp/WurstStdlib2` — a fetched test artefact, editing + it changes nothing real. + +## Keeping going + +- Finishing an item is not the end of the run. Go back to step 1 and take the next one. Do + not stop to summarise, and do not treat a green suite as a finish line. +- At most one item may be marked blocked per iteration, and only if it genuinely needs a + decision rather than more work. If everything left looks blocked, that is wrong about at + least one of them — re-read and start the one that can be moved furthest. +- If an item goes three iterations without landing a commit, split it in the backlog and + move on. Don't spend the whole run on one thing. +- Backlog item 12 is a standing item that never completes. There is always a next item. From cc994da12d2235f0f04fcdd5f7944cc0fdfd3222 Mon Sep 17 00:00:00 2001 From: Frotty Date: Sat, 15 Aug 2026 04:15:00 +0200 Subject: [PATCH 5/7] Sanitise Lua method names where they are assigned Method names become Lua table keys, so they must be identifiers. A method specialised with two type arguments was named after them, commas included, and emitted `Class.get_specialized_integer, integer = impl` - valid Lua that quietly assigns to two targets; a tuple argument produced characters luac rejects outright. normalizeMethodNames is the pass that gives one name to a whole dispatch group, so it sanitises before uniquing: two names that differed only in characters Lua has no place for still get a slot each. The backend maps every slot key and every LuaMethod name through the same function, so call sites and class tables keep agreeing. Lua's identifier rule now has one home. The luac check never caught this, because the broken output parses. Assert instead on the names themselves: every emitted function, method, variable, field and call-by-name must be an identifier, checked for every testLua compile. --- BACKLOG.md | 52 ++++++++------ .../imtranslation/LuaDispatchPreparation.java | 5 +- .../lua/translation/LuaAssertions.java | 70 +++++++++++++++++++ .../lua/translation/LuaIdentifiers.java | 54 ++++++++++++++ .../lua/translation/LuaTranslator.java | 45 +++++------- .../wurstscript/tests/WurstScriptTest.java | 2 + 6 files changed, 177 insertions(+), 51 deletions(-) create mode 100644 de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java diff --git a/BACKLOG.md b/BACKLOG.md index 286550bec..ba7ec03be 100644 --- a/BACKLOG.md +++ b/BACKLOG.md @@ -10,28 +10,22 @@ Notes rather than leaving it in a commit message. ## Todo -1. **Lua method names reach the backend unsanitised.** `LuaTranslator.luaMethod.initFor` - passes `a.getName()` raw; `luaVar`, `luaFunc` and `luaClassVar` all sanitise via - `uniqueName`. Method names become Lua table keys, so they must be valid identifiers. - `EliminateGenerics.specializeMethod` builds `name + "_specialized_" + generics.makeName()` - on the Lua path, and `makeName` joins type arguments with `", "` — so a class method - specialised with two type arguments emits `Class.get_specialized_integer, integer = impl`, - which is valid Lua assigning to two targets, and a tuple argument emits `⦅⦆` and fails - the syntax check. Repro: `FastHashMapTests.tupleKeyLua`. Commas are already visible in - `test-output/lua/FastHashMapTests_fastHashMapRuntimeLua.lua`. - Overriding methods must keep landing in the same slot, so normalise per distinct original - name, not per method node. - -2. **Nothing checks that emitted Lua identifiers are valid.** The luac syntax check catches a - hard break, but not the silent case: `Class.get_specialized_integer, integer = impl` parses - fine and quietly assigns to two targets. That is how item 1 shipped unnoticed. Assert in the - Lua test harness that every emitted name — variable, function, class, method slot, field — - matches `[A-Za-z_][A-Za-z0-9_]*`, so this whole class of bug fails loudly at the point it is - introduced. Do this alongside item 1; it is what stops item 1 recurring. - -3. **`slotFor` looks bound to `get`'s implementation** in the same emitted Lua. May be a real - mis-binding in `specializeMethod`/`adaptSubmethods`, may be an artefact of item 1 mangling - the output. Diagnose only after item 1, from freshly emitted Lua. +Numbering is stable: finished items leave a gap rather than shifting the ones below, +because `LOOP.md` refers to items by number. + +3. **`slotFor` is bound to `get`'s implementation** in the emitted Lua — confirmed real, and + not an artefact of item 1: it survives sanitisation unchanged. Diagnosed, not yet fixed. + The alias sets in `LuaDispatchPreparation` decide which slots a method claims, and + `sharesSemanticName` accepts a match on *either* of two names: the source name (`get`, + `slotFor`) or `semanticNameFromMethodName`, which is the substring after the last + underscore. For a specialised method that substring is a type-argument fragment — + `FastHashMap_get_specialized__integer__integer___integer` and the `slotFor` one both yield + `integer` — so two unrelated methods count as sharing a name. They also share a dispatch + signature here (`(pos) returns int` both), which is the other half of the guard, so `get` + claims `slotFor`'s slot. Fix: when both methods have a real source name, that should decide; + the substring heuristic is a fallback for when there is no trace to ask, not an alternative. + Watch the closure and bridge cases in `TypeClassTests`/`LuaBackendAuditTests` — they are what + the loose match was presumably widened for. 4. **Finish the FastHashMap proof.** `FastHashMapTests` is the first real use of bounds. Add `remove` with tombstones, and an assertion that the emitted code stays cheap: no @@ -94,6 +88,14 @@ Notes rather than leaving it in a commit message. ## Done +- 1 + 2. Lua method names are sanitised where they are assigned, not where they are printed. + `LuaDispatchPreparation.normalizeMethodNames` is the pass that gives one name to a whole + dispatch group, so it now sanitises before uniquing — two names differing only in characters + Lua has no place for still get a slot each. `LuaTranslator` maps every slot key and every + `LuaMethod` name through the same function, so call sites and class tables agree. Lua's + identifier rule now lives in one place, `LuaIdentifiers`. `LuaAssertions.assertNamesAreValidIdentifiers` + walks the emitted Lua and fails on any name that is not an identifier; it runs for every + `testLua` compile, so the silent two-target-assignment case cannot come back. - Substitution now carries the type class binding with the type (#1229). Also fixed the type-variable reference on `ImTypeVarDispatch`, which a walk over types alone missed. @@ -101,7 +103,11 @@ Notes rather than leaving it in a commit message. - `%` is real modulo in Wurst; `mod` is integer modulo. `int % 8` types as `real`. - Emitted Lua must be byte-identical for identical input (AGENTS.md §8). It is the only - emitted output that can be diffed across runs — see item 8. + emitted output that can be diffed across runs — see item 11. +- Method names are not what the frontend called them. `LuaDispatchPreparation` renames a whole + dispatch group to one name and attaches alias sets, and only then does the backend run. A + question about which Lua slot something lands in is a question about that pass, not about + `LuaTranslator`. - Tests run five Jass configurations plus the interpreter, then the Lua target separately. `testAssertOkLines(true, ...)` covers both the pre-transform interpreter and full monomorphisation, so it is a stronger check than it looks. diff --git a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/imtranslation/LuaDispatchPreparation.java b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/imtranslation/LuaDispatchPreparation.java index 662c812b9..cbed8db1b 100644 --- a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/imtranslation/LuaDispatchPreparation.java +++ b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/imtranslation/LuaDispatchPreparation.java @@ -11,6 +11,7 @@ import de.peeeq.wurstscript.jassIm.ImProg; import de.peeeq.wurstscript.jassIm.ImType; import de.peeeq.wurstscript.jassIm.ImVars; +import de.peeeq.wurstscript.translation.lua.translation.LuaIdentifiers; import java.util.ArrayList; import java.util.Collection; @@ -99,7 +100,9 @@ private static void normalizeMethodNames(ImProg prog, List allMethods) continue; } group.sort(Comparator.comparing(LuaDispatchPreparation::methodSortKey)); - String name = uniqueName(group.get(0).getName(), usedNames); + // The name is about to become a Lua table key. Sanitising before uniquing means two + // names that only differed in characters Lua has no place for still get one slot each. + String name = uniqueName(LuaIdentifiers.toIdentifier(group.get(0).getName()), usedNames); for (ImMethod method : group) { method.setName(name); } diff --git a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaAssertions.java b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaAssertions.java index 82d003d60..4c1ed5298 100644 --- a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaAssertions.java +++ b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaAssertions.java @@ -1,9 +1,19 @@ package de.peeeq.wurstscript.translation.lua.translation; +import de.peeeq.wurstscript.luaAst.Element; +import de.peeeq.wurstscript.luaAst.LuaCompilationUnit; +import de.peeeq.wurstscript.luaAst.LuaExprFieldAccess; +import de.peeeq.wurstscript.luaAst.LuaExprFunctionCallByName; +import de.peeeq.wurstscript.luaAst.LuaFunction; +import de.peeeq.wurstscript.luaAst.LuaMethod; +import de.peeeq.wurstscript.luaAst.LuaTableNamedField; +import de.peeeq.wurstscript.luaAst.LuaVariable; + import java.util.ArrayList; import java.util.HashSet; import java.util.List; import java.util.Set; +import java.util.TreeSet; /** * Static assertion helpers for the Lua backend. @@ -15,6 +25,66 @@ public class LuaAssertions { private LuaAssertions() {} + /** + * Asserts that every name the backend emits is a Lua identifier. + * + *

A name that is not one usually breaks the syntax check, but not always: a table key + * containing a comma parses as an assignment to two targets and quietly stores the value in + * the wrong place. Checking the names themselves catches that case at the point it is + * introduced, rather than as a wrong result at runtime. + */ + public static void assertNamesAreValidIdentifiers(LuaCompilationUnit luaCode) { + Set invalid = new TreeSet<>(); + luaCode.accept(new Element.DefaultVisitor() { + private void check(String kind, String name) { + // A vararg parameter is the one name that is legal without being an identifier. + if (!LuaIdentifiers.isValid(name) && !LuaIdentifiers.VARARG.equals(name)) { + invalid.add(kind + " '" + name + "'"); + } + } + + @Override + public void visit(LuaFunction f) { + super.visit(f); + check("function", f.getName()); + } + + @Override + public void visit(LuaMethod m) { + super.visit(m); + check("method", m.getName()); + } + + @Override + public void visit(LuaVariable v) { + super.visit(v); + check("variable", v.getName()); + } + + @Override + public void visit(LuaExprFieldAccess fa) { + super.visit(fa); + check("field", fa.getFieldName()); + } + + @Override + public void visit(LuaTableNamedField f) { + super.visit(f); + check("field", f.getFieldName()); + } + + @Override + public void visit(LuaExprFunctionCallByName call) { + super.visit(call); + check("call to", call.getFuncName()); + } + }); + if (!invalid.isEmpty()) { + throw new RuntimeException("Wurst Lua backend assertion failed: emitted names are not Lua identifiers: " + + String.join(", ", invalid)); + } + } + /** * Asserts that every emitted call to {@code __wurst_GetHandleId} has a helper definition. * diff --git a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java new file mode 100644 index 000000000..95cde0a8b --- /dev/null +++ b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java @@ -0,0 +1,54 @@ +package de.peeeq.wurstscript.translation.lua.translation; + +/** + * Lua's rule for what a generated name may look like, kept in one place. + * + *

Names from the intermediate language are not constrained to Lua's identifier syntax; + * specialised generics, for example, are named after their type arguments. Sanitising in the + * backend keeps that rule where it belongs rather than requiring every earlier pass to know + * about Lua. Any collisions the mapping introduces are resolved by the usual uniquing. + */ +public final class LuaIdentifiers { + + /** Lua's vararg parameter, which is a legal parameter name but not an identifier. */ + public static final String VARARG = "..."; + + /** Whether {@code name} can be used as-is as a Lua identifier or table key. */ + public static boolean isValid(String name) { + if (name == null || name.isEmpty() || isDigit(name.charAt(0))) { + return false; + } + for (int i = 0; i < name.length(); i++) { + if (!isIdentifierPart(name.charAt(i))) { + return false; + } + } + return true; + } + + /** Maps any name onto a Lua identifier, leaving names that already are one untouched. */ + public static String toIdentifier(String name) { + if (isValid(name)) { + return name; + } + StringBuilder sb = new StringBuilder(name.length()); + for (int i = 0; i < name.length(); i++) { + char c = name.charAt(i); + sb.append(isIdentifierPart(c) ? c : '_'); + } + if (sb.length() == 0 || isDigit(sb.charAt(0))) { + sb.insert(0, '_'); + } + return sb.toString(); + } + + private static boolean isIdentifierPart(char c) { + return c == '_' || (c < 128 && Character.isLetterOrDigit(c)); + } + + private static boolean isDigit(char c) { + return c >= '0' && c <= '9'; + } + + private LuaIdentifiers() {} +} diff --git a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaTranslator.java b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaTranslator.java index 4e729a863..f00242efa 100644 --- a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaTranslator.java +++ b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaTranslator.java @@ -135,7 +135,9 @@ public LuaFunction initFor(ImFunction a) { @Override public LuaMethod initFor(ImMethod a) { LuaExpr receiver = LuaAst.LuaExprVarAccess(luaClassVar.getFor(a.attrClass())); - return LuaAst.LuaMethod(receiver, a.getName(), LuaAst.LuaParams(), LuaAst.LuaStatements()); + // A method name is a table key, so it must be an identifier - but unlike a variable + // it must not be uniqued: every override has to keep landing in the same slot. + return LuaAst.LuaMethod(receiver, dispatchSlotName(a.getName()), LuaAst.LuaParams(), LuaAst.LuaStatements()); } }; @@ -196,28 +198,8 @@ public LuaTranslator(ImProg prog, ImTranslator imTr) { luaModel = LuaAst.LuaCompilationUnit(); } - /** - * Makes an intermediate-language name usable as a Lua identifier. - *

- * Names from the IM are not constrained to Lua's identifier syntax; specialised generics, for - * example, are named after their type arguments. Sanitising here keeps that rule where it - * belongs, in the backend, rather than requiring every earlier pass to know about Lua. Any - * collisions the mapping introduces are resolved by the usual uniquing. - */ - private static String toLuaIdentifier(String name) { - StringBuilder sb = new StringBuilder(name.length()); - for (int i = 0; i < name.length(); i++) { - char c = name.charAt(i); - sb.append(c == '_' || Character.isLetterOrDigit(c) && c < 128 ? c : '_'); - } - if (sb.length() == 0 || Character.isDigit(sb.charAt(0))) { - sb.insert(0, '_'); - } - return sb.toString(); - } - protected String uniqueName(String rawName) { - String name = toLuaIdentifier(rawName); + String name = LuaIdentifiers.toIdentifier(rawName); Integer nextIndex = uniqueNameCounters.get(name); if (nextIndex == null) { uniqueNameCounters.put(name, 1); @@ -462,7 +444,7 @@ private void collectMethodNames(ImClass c, Set methodNames, Set } visited.add(c); for (ImMethod method : c.getMethods()) { - methodNames.add(method.getName()); + methodNames.add(dispatchSlotName(method.getName())); } for (ImClassType sc : c.getSuperClasses()) { collectMethodNames(sc.getClassDef(), methodNames, visited); @@ -914,7 +896,7 @@ private void createMethods(ImClass c, LuaVariable classVar) { ImMethod chosen = chosenByGroup.get(groupMethods); Set memberNames = new HashSet<>(); for (ImMethod m : groupMethods) { - memberNames.add(m.getName()); + memberNames.add(dispatchSlotName(m.getName())); } Set slotNames = collectDispatchSlotNames(c, groupMethods); for (String slotName : slotNames) { @@ -934,7 +916,7 @@ private void createMethods(ImClass c, LuaVariable classVar) { ImMethod chosen = chosenByGroup.get(groupMethods); Set memberNames = new HashSet<>(); for (ImMethod m : groupMethods) { - memberNames.add(m.getName()); + memberNames.add(dispatchSlotName(m.getName())); } for (String slotName : collectDispatchSlotNames(c, groupMethods)) { if (memberNames.contains(slotName)) { @@ -973,6 +955,15 @@ && implArity(chosen) != implArity(current)) { } } + /** + * The Lua table key a dispatch slot is emitted under. Aliases and class-qualified names are + * built from IM names, which may contain characters Lua has no place for; the mapping has to + * be the same one call sites go through, so that a slot is still found under its new name. + */ + private String dispatchSlotName(String rawName) { + return LuaIdentifiers.toIdentifier(rawName); + } + private Set collectDispatchSlotNames(ImClass receiverClass, List groupMethods) { Set slotNames = new TreeSet<>(); Set semanticNames = new TreeSet<>(); @@ -982,7 +973,7 @@ private Set collectDispatchSlotNames(ImClass receiverClass, List collectDispatchSlotNames(ImClass receiverClass, List()); for (String className : classNames) { for (String semanticName : semanticNames) { - slotNames.add(className + "_" + semanticName); + slotNames.add(dispatchSlotName(className + "_" + semanticName)); } } } diff --git a/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/WurstScriptTest.java b/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/WurstScriptTest.java index 3498cec3f..545f28c4d 100644 --- a/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/WurstScriptTest.java +++ b/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/WurstScriptTest.java @@ -25,6 +25,7 @@ import de.peeeq.wurstscript.jassprinter.JassPrinter; import de.peeeq.wurstscript.luaAst.LuaCompilationUnit; import de.peeeq.wurstscript.luaAst.*; +import de.peeeq.wurstscript.translation.lua.translation.LuaAssertions; import de.peeeq.wurstscript.translation.imtranslation.ImTranslator; import de.peeeq.wurstscript.translation.imtranslation.RecycleCodeGeneratorQueue; import de.peeeq.wurstscript.utils.Utils; @@ -500,6 +501,7 @@ private void translateAndTestLua(String name, boolean executeProg, WurstGui gui, compiler.runCompiletime(WurstProjectConfigData.empty(), false, false); LuaCompilationUnit luaCode = compiler.transformProgToLua(); + LuaAssertions.assertNamesAreValidIdentifiers(luaCode); checkLuaRootPurity(luaCode); StringBuilder sb = new StringBuilder(); luaCode.print(sb, 0); From 7871731728d95877eb41daef0e15654ead7c5e53 Mon Sep 17 00:00:00 2001 From: Frotty Date: Sun, 16 Aug 2026 08:37:19 +0200 Subject: [PATCH 6/7] Reject Lua keywords when validating a generated name Wurst and Lua reserve different words, so a method can be declared repeat or goto and reach the backend under that name. Method names survive it, because the pass that assigns them uniques against the reserved set. A closure does not: it adds the name it implements as a dispatch alias directly, so the alias arrives as a bare keyword and is emitted as a table key. luac rejects that, so it was loud rather than wrong, but the check added alongside it accepted the name - and catching this before the syntax check is the whole point of having it. isValid now rejects keywords and toIdentifier maps them out of the way. Underscores rather than a counter, so a keyword maps to the same name wherever it is derived: call sites and class tables have to agree without consulting each other. --- .../lua/translation/LuaIdentifiers.java | 17 ++++++++++++++-- .../wurstscript/tests/TypeClassTests.java | 20 +++++++++++++++++++ 2 files changed, 35 insertions(+), 2 deletions(-) diff --git a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java index 95cde0a8b..b0da4381f 100644 --- a/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java +++ b/de.peeeq.wurstscript/src/main/java/de/peeeq/wurstscript/translation/lua/translation/LuaIdentifiers.java @@ -13,7 +13,14 @@ public final class LuaIdentifiers { /** Lua's vararg parameter, which is a legal parameter name but not an identifier. */ public static final String VARARG = "..."; - /** Whether {@code name} can be used as-is as a Lua identifier or table key. */ + /** + * Whether {@code name} can be used as-is as a Lua identifier or table key. + * + *

A keyword is spelled like an identifier and is not one. Wurst reserves a different set, so + * a method can be declared {@code repeat} or {@code goto} and reach the backend under that + * name; emitted as a table key it is a syntax error rather than a wrong result, but this is the + * check that is supposed to catch it first. + */ public static boolean isValid(String name) { if (name == null || name.isEmpty() || isDigit(name.charAt(0))) { return false; @@ -23,7 +30,7 @@ public static boolean isValid(String name) { return false; } } - return true; + return !LuaReservedNames.LUA_KEYWORDS.contains(name); } /** Maps any name onto a Lua identifier, leaving names that already are one untouched. */ @@ -39,6 +46,12 @@ public static String toIdentifier(String name) { if (sb.length() == 0 || isDigit(sb.charAt(0))) { sb.insert(0, '_'); } + // A trailing underscore rather than a counter, so the name a keyword maps to is the same + // wherever it is derived - call sites and class tables have to agree without consulting + // each other. + while (LuaReservedNames.LUA_KEYWORDS.contains(sb.toString())) { + sb.append('_'); + } return sb.toString(); } diff --git a/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/TypeClassTests.java b/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/TypeClassTests.java index 6e0ba186c..cd8429a4e 100644 --- a/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/TypeClassTests.java +++ b/de.peeeq.wurstscript/src/test/java/tests/wurstscript/tests/TypeClassTests.java @@ -121,6 +121,26 @@ public void dispatchInsideClosureIsRejectedForLua() { ); } + /** + * Wurst and Lua reserve different words, so a method can be declared {@code repeat} and reach + * the backend under that name. A closure adds the name it implements as a dispatch alias + * directly, without the uniquing that protects method names, so the alias arrives as a bare + * keyword and is emitted as a table key — {@code expected near 'repeat'} from luac. + */ + @Test + public void closureImplementingALuaKeywordName() { + test().testLua(true).executeProg().lines( + "package test", + "native testSuccess()", + "interface Producer", + " function repeat() returns int", + "init", + " Producer p = () -> 42", + " if p.repeat() == 42", + " testSuccess()" + ); + } + /** Each type argument picks its own instance, so one generic serves several types. */ @Test public void twoInstancesOfOneClass() { From a587e86c960dafc8054eb7631ced79911dd353be Mon Sep 17 00:00:00 2001 From: Frotty Date: Sun, 16 Aug 2026 09:05:33 +0200 Subject: [PATCH 7/7] Keep local session files out of the repository Neither of these belongs in a branch that gets merged. The loop configuration is machine-local state - an active session id, an iteration count, a path and a branch name from one checkout - and another checkout picking it up would inherit an automation session that has nothing to do with it. The brief beside it is one run's instructions, not documentation of anything. Both came in with the branch this work was based on rather than with the work. Ignored as well, so they do not come back the next time either is written. --- .claude/ralph-loop.local.md | 10 -------- .gitignore | 2 ++ LOOP.md | 48 ------------------------------------- 3 files changed, 2 insertions(+), 58 deletions(-) delete mode 100644 .claude/ralph-loop.local.md delete mode 100644 LOOP.md diff --git a/.claude/ralph-loop.local.md b/.claude/ralph-loop.local.md deleted file mode 100644 index 305f01377..000000000 --- a/.claude/ralph-loop.local.md +++ /dev/null @@ -1,10 +0,0 @@ ---- -active: true -iteration: 1 -session_id: 0413fc86-e9a2-4954-bc47-1620ed81d95e -max_iterations: 0 -completion_promise: null -started_at: "2026-08-15T01:51:05Z" ---- - -Work on the WurstScript compiler at c:Usersrun-eDocumentsGitHubWurstScript, on the branch feat/fasthashmap-proof. Read AGENTS.md and BACKLOG.md before doing anything. diff --git a/.gitignore b/.gitignore index 7b7bf5ad3..fad3ad29c 100644 --- a/.gitignore +++ b/.gitignore @@ -60,3 +60,5 @@ de.peeeq.wurstscript/output.txt /HelperScripts/.gradle /.gradle-user-home /gradle-home-temp +/.claude/ +/LOOP.md diff --git a/LOOP.md b/LOOP.md deleted file mode 100644 index ed3234d69..000000000 --- a/LOOP.md +++ /dev/null @@ -1,48 +0,0 @@ -# Overnight loop brief - -Work on the WurstScript compiler in this repository, on the branch -`feat/fasthashmap-proof`. Read `AGENTS.md` and `BACKLOG.md` before doing anything. - -## Each iteration - -1. Read `BACKLOG.md`. Take the top item that isn't blocked. -2. Do it. Root-cause it — adjust the underlying system rather than patching a symptom. - If it turns out to be bigger than one iteration, split it in the backlog and do the - first part. -3. Verify. Targeted tests while iterating; the full suite before any commit that touches - main source: - - cd de.peeeq.wurstscript && ./gradlew test - - It must be green. Emitted Lua must stay byte-identical unless the change is meant to - alter it — compare two runs to check. Do not diff `.j` across runs; it is not stable, - and backlog item 11 explains why. -4. Commit and push. Small commits, one concern each. Never end an iteration with - uncommitted work. -5. Update `BACKLOG.md`: move finished items to Done with one line on what actually - happened, add anything learned to Notes, reorder if something more urgent turned up. - -## Rules - -- Never force-push, rewrite history, merge, push to master, or open a PR. The branch gets - reviewed in the morning. -- Commits are authored as the repository owner. No AI, assistant, or co-author references - anywhere in commit messages or code comments. -- Never stop to ask. If something needs a decision from the owner, write the question into - `BACKLOG.md` under that item, mark it blocked, and move to the next item. -- If the full suite goes red and it can't be fixed within the iteration, revert the change, - note why in `BACKLOG.md`, and move on. Leave the branch green. -- Comments explain why, not what. Match the surrounding style. -- Do not touch `de.peeeq.wurstscript/temp/WurstStdlib2` — a fetched test artefact, editing - it changes nothing real. - -## Keeping going - -- Finishing an item is not the end of the run. Go back to step 1 and take the next one. Do - not stop to summarise, and do not treat a green suite as a finish line. -- At most one item may be marked blocked per iteration, and only if it genuinely needs a - decision rather than more work. If everything left looks blocked, that is wrong about at - least one of them — re-read and start the one that can be moved furthest. -- If an item goes three iterations without landing a commit, split it in the backlog and - move on. Don't spend the whole run on one thing. -- Backlog item 12 is a standing item that never completes. There is always a next item.