C codegen: preparations for different seq and string implementations

compiler/ccgexprs.nim

@@ -30,12 +30,6 @@ proc intLiteral(i: BiggestInt): Rope =
   else:
     result = ~"(IL64(-9223372036854775807) - IL64(1))"
 
-proc getStrLit(m: BModule, s: string): Rope =
-  discard cgsym(m, "TGenericSeq")
-  result = getTempName(m)
-  addf(m.s[cfsData], "STRING_LITERAL($1, $2, $3);$n",
-       [result, makeCString(s), rope(len(s))])
-
 proc genLiteral(p: BProc, n: PNode, ty: PType): Rope =
   if ty == nil: internalError(n.info, "genLiteral: ty is nil")
   case n.kind
@@ -67,19 +61,9 @@ proc genLiteral(p: BProc, n: PNode, ty: PType): Rope =
   of nkStrLit..nkTripleStrLit:
     case skipTypes(ty, abstractVarRange).kind
     of tyNil:
-      result = ropecg(p.module, "((#NimStringDesc*) NIM_NIL)", [])
+      result = genNilStringLiteral(p.module, n.info)
     of tyString:
-      if n.strVal.isNil:
+      result = genStringLiteral(p.module, n)
-        result = ropecg(p.module, "((#NimStringDesc*) NIM_NIL)", [])
-      else:
-        let id = nodeTableTestOrSet(p.module.dataCache, n, p.module.labels)
-        if id == p.module.labels:
-          # string literal not found in the cache:
-          result = ropecg(p.module, "((#NimStringDesc*) &$1)",
-                          [getStrLit(p.module, n.strVal)])
-        else:
-          result = ropecg(p.module, "((#NimStringDesc*) &$1$2)",
-                          [p.module.tmpBase, rope(id)])
     else:
       if n.strVal.isNil: result = rope("NIM_NIL")
       else: result = makeCString(n.strVal)
@@ -823,16 +807,16 @@ proc genFieldCheck(p: BProc, e: PNode, obj: Rope, field: PSym) =
     genInExprAux(p, it, u, v, test)
     let id = nodeTableTestOrSet(p.module.dataCache,
                                newStrNode(nkStrLit, field.name.s), p.module.labels)
-    let strLit = if id == p.module.labels: getStrLit(p.module, field.name.s)
+    let strLit = if id == p.module.labels: genStringLiteralDataOnly(p.module, field.name.s, e.info)
                  else: p.module.tmpBase & rope(id)
     if op.magic == mNot:
       linefmt(p, cpsStmts,
-              "if ($1) #raiseFieldError(((#NimStringDesc*) &$2));$n",
+              "if ($1) #raiseFieldError($2);$n",
-              rdLoc(test), strLit)
+              rdLoc(test), genStringLiteralFromData(p.module, strLit, e.info))
     else:
       linefmt(p, cpsStmts,
-              "if (!($1)) #raiseFieldError(((#NimStringDesc*) &$2));$n",
+              "if (!($1)) #raiseFieldError($2);$n",
-              rdLoc(test), strLit)
+              rdLoc(test), genStringLiteralFromData(p.module, strLit, e.info))
 
 proc genCheckedRecordField(p: BProc, e: PNode, d: var TLoc) =
   if optFieldCheck in p.options:

compiler/ccgliterals.nim

@@ -0,0 +1,81 @@
+#
+#
+#           The Nim Compiler
+#        (c) Copyright 2018 Andreas Rumpf
+#
+#    See the file "copying.txt", included in this
+#    distribution, for details about the copyright.
+#
+
+## This include file contains the logic to produce constant string
+## and seq literals. The code here is responsible that
+## ``const x = ["a", "b"]`` works without hidden runtime creation code.
+## The price is that seqs and strings are not purely a library
+## implementation.
+
+template detectVersion(field, corename) =
+  if m.g.field == 0:
+    let core = getCompilerProc(corename)
+    if core == nil or core.kind != skConst:
+      m.g.field = 1
+    else:
+      m.g.field = int ast.getInt(core.ast)
+  result = m.g.field
+
+proc detectStrVersion(m: BModule): int =
+  detectVersion(strVersion, "nimStrVersion")
+
+proc detectSeqVersion(m: BModule): int =
+  detectVersion(seqVersion, "nimSeqVersion")
+
+# ----- Version 1: GC'ed strings and seqs --------------------------------
+
+proc genStringLiteralDataOnlyV1(m: BModule, s: string): Rope =
+  discard cgsym(m, "TGenericSeq")
+  result = getTempName(m)
+  addf(m.s[cfsData], "STRING_LITERAL($1, $2, $3);$n",
+       [result, makeCString(s), rope(len(s))])
+
+proc genStringLiteralV1(m: BModule; n: PNode): Rope =
+  if s.isNil:
+    result = ropecg(m, "((#NimStringDesc*) NIM_NIL)", [])
+  else:
+    let id = nodeTableTestOrSet(m.dataCache, n, m.labels)
+    if id == m.labels:
+      # string literal not found in the cache:
+      result = ropecg(m, "((#NimStringDesc*) &$1)",
+                      [genStringLiteralDataOnlyV1(m, n.strVal)])
+    else:
+      result = ropecg(m, "((#NimStringDesc*) &$1$2)",
+                      [m.tmpBase, rope(id)])
+
+# ------ Version 2: destructor based strings and seqs -----------------------
+
+proc genStringLiteralDataOnlyV2(m: BModule, s: string): Rope =
+  discard "to implement"
+
+proc genStringLiteralV2(m: BModule; n: PNode): Rope =
+  discard "to implement"
+
+# ------ Version selector ---------------------------------------------------
+
+proc genStringLiteralDataOnly(m: BModule; s: string; info: TLineInfo): Rope =
+  case detectStrVersion(m)
+  of 0, 1: result = genStringLiteralDataOnlyV1(m, s)
+  of 2: result = genStringLiteralDataOnlyV2(m, s)
+  else:
+    localError(info, "cannot determine how to produce code for string literal")
+
+proc genStringLiteralFromData(m: BModule; data: Rope; info: TLineInfo): Rope =
+  result = ropecg(m, "((#NimStringDesc*) &$1)",
+                [data])
+
+proc genNilStringLiteral(m: BModule; info: TLineInfo): Rope =
+  result = ropecg(m, "((#NimStringDesc*) NIM_NIL)", [])
+
+proc genStringLiteral(m: BModule; n: PNode): Rope =
+  case detectStrVersion(m)
+  of 0, 1: result = genStringLiteralV1(m, n)
+  of 2: result = genStringLiteralV2(m, n)
+  else:
+    localError(n.info, "cannot determine how to produce code for string literal")

compiler/ccgtypes.nim

@@ -267,10 +267,6 @@ proc addAbiCheck(m: BModule, t: PType, name: Rope) =
   if isDefined("checkabi"):
     addf(m.s[cfsTypeInfo], "NIM_CHECK_SIZE($1, $2);$n", [name, rope(getSize(t))])
 
-proc getTempName(m: BModule): Rope =
-  result = m.tmpBase & rope(m.labels)
-  inc m.labels
-
 proc ccgIntroducedPtr(s: PSym): bool =
   var pt = skipTypes(s.typ, typedescInst)
   assert skResult != s.kind
@@ -316,6 +312,11 @@ proc getSimpleTypeDesc(m: BModule, typ: PType): Rope =
   of tyPointer:
     result = typeNameOrLiteral(m, typ, "void*")
   of tyString:
+    case detectStrVersion(m)
+    of 2:
+      discard cgsym(m, "string")
+      result = typeNameOrLiteral(m, typ, "NimStringV2")
+    else:
       discard cgsym(m, "NimStringDesc")
       result = typeNameOrLiteral(m, typ, "NimStringDesc*")
   of tyCString: result = typeNameOrLiteral(m, typ, "NCSTRING")

compiler/cgen.nim

@@ -238,7 +238,12 @@ proc genProc(m: BModule, prc: PSym)
 template compileToCpp(m: BModule): untyped =
   gCmd == cmdCompileToCpp or sfCompileToCpp in m.module.flags
 
-include "ccgtypes.nim"
+proc getTempName(m: BModule): Rope =
+  result = m.tmpBase & rope(m.labels)
+  inc m.labels
+
+include ccgliterals
+include ccgtypes
 
 # ------------------------------ Manager of temporaries ------------------
 
@@ -551,11 +556,13 @@ proc loadDynamicLib(m: BModule, lib: PLib) =
       for i in countup(0, high(s)):
         inc(m.labels)
         if i > 0: add(loadlib, "||")
-        appcg(m, loadlib, "($1 = #nimLoadLibrary((#NimStringDesc*) &$2))$n",
+        let n = newStrNode(nkStrLit, s[i])
-              [tmp, getStrLit(m, s[i])])
+        n.info = lib.path.info
+        appcg(m, loadlib, "($1 = #nimLoadLibrary($2))$n",
+              [tmp, genStringLiteral(m, n)])
       appcg(m, m.s[cfsDynLibInit],
-            "if (!($1)) #nimLoadLibraryError((#NimStringDesc*) &$2);$n",
+            "if (!($1)) #nimLoadLibraryError($2);$n",
-            [loadlib, getStrLit(m, lib.path.strVal)])
+            [loadlib, genStringLiteral(m, lib.path)])
     else:
       var p = newProc(nil, m)
       p.options = p.options - {optStackTrace, optEndb}

compiler/cgendata.nim

@@ -116,6 +116,7 @@ type
     breakpoints*: Rope # later the breakpoints are inserted into the main proc
     typeInfoMarker*: TypeCache
     config*: ConfigRef
+    strVersion*, seqVersion*: int # version of the string/seq implementation to use
 
   TCGen = object of TPassContext # represents a C source file
     s*: TCFileSections        # sections of the C file

compiler/pragmas.nim

@@ -65,7 +65,7 @@ const
     wGensym, wInject, wCodegenDecl, wGuard, wGoto, wExportNims, wUsed}
   constPragmas* = {wImportc, wExportc, wHeader, wDeprecated, wMagic, wNodecl,
     wExtern, wImportCpp, wImportObjC, wError, wGensym, wInject, wExportNims,
-    wIntDefine, wStrDefine, wUsed}
+    wIntDefine, wStrDefine, wUsed, wCompilerProc, wCore}
   letPragmas* = varPragmas
   procTypePragmas* = {FirstCallConv..LastCallConv, wVarargs, wNosideeffect,
                       wThread, wRaises, wLocks, wTags, wGcSafe}

lib/core/seqs.nim

@@ -15,11 +15,11 @@ type
     len, cap: int
     data: ptr UncheckedArray[T]
 
+const nimSeqVersion {.core.} = 2
+
 template frees(s) = dealloc(s.data, s.cap * sizeof(T))
 
 # XXX make code memory safe for overflows in '*'
-proc nimSeqLiteral[T](x: openArray[T]): seq[T] {.core.} =
-  seq[T](len: x.len, cap: x.len, data: x)
 
 when defined(nimHasTrace):
   proc `=trace`[T](s: seq[T]; a: Allocator) =

lib/core/strs.nim

@@ -12,12 +12,11 @@
 import allocators
 
 type
-  string {.core.} = object
+  string {.core, exportc: "NimStringV2".} = object
     len, cap: int
     data: ptr UncheckedArray[char]
 
-proc nimStringLiteral(x: cstring; len: int): string {.core.} =
+const nimStrVersion {.core.} = 2
-  string(len: len, cap: len, data: x)
 
 template frees(s) = dealloc(s.data, s.cap + 1)
 
@@ -80,7 +79,7 @@ proc newString*(len: int): string =
   if len > 0:
     result.data = alloc0(len+1)
 
-converter toCString(x: string): cstring {.core.} =
+converter toCString(x: string): cstring {.core, inline.} =
   if x.len == 0: cstring"" else: cast[cstring](x.data)
 
 proc newStringOfCap*(cap: int): string =