* Normalisation.

src/eval.cc

@@ -5,6 +5,7 @@
 #include <sys/stat.h>
 #include <sys/wait.h>
 #include <unistd.h>
+#include <fcntl.h>
 
 #include "eval.hh"
 #include "globals.hh"
@@ -91,7 +92,7 @@ static Hash computeDerived(Hash sourceHash, string targetName,
                 }
 #endif
 
-            build:
+//             build:
 
                 /* Fill in the environment.  We don't bother freeing
                    the strings, since we'll exec or die soon
@@ -210,13 +211,14 @@ Hash hashExpr(Expr e)
 /* Evaluate an expression; the result must be a string. */
 static string evalString(Expr e)
 {
-    e = evalValue(e);
+    e = whNormalise(e);
     char * s;
     if (ATmatch(e, "Str(<str>)", &s)) return s;
     else throw badTerm("string value expected", e);
 }
 
 
+#if 0
 /* Evaluate an expression; the result must be a value reference. */
 static Hash evalHash(Expr e)
 {
@@ -225,8 +227,10 @@ static Hash evalHash(Expr e)
     if (ATmatch(e, "Hash(<str>)", &s)) return parseHash(s);
     else throw badTerm("value reference expected", e);
 }
+#endif
 
 
+#if 0
 /* Evaluate a list of arguments into normal form. */
 void evalArgs(ATermList args, ATermList & argsNF, Environment & env)
 {
@@ -255,6 +259,7 @@ void evalArgs(ATermList args, ATermList & argsNF, Environment & env)
 
     argsNF = ATreverse(argsNF);
 }
+#endif
 
 
 Expr substExpr(string x, Expr rep, Expr e)
@@ -302,19 +307,13 @@ Expr substExpr(string x, Expr rep, Expr e)
 }
 
 
+#if 0
 Expr evalValue(Expr e)
 {
     char * s;
     Expr eBuildPlatform, eProg, e2, e3, e4;
     ATermList args;
 
-    /* Normal forms. */
-    if (ATmatch(e, "Str(<str>)", &s) ||
-        ATmatch(e, "Bool(True)") || 
-        ATmatch(e, "Bool(False)") ||
-        ATmatch(e, "Lam(<str>, <term>)", &s, &e2))
-        return e;
-
     /* Value references. */
     if (ATmatch(e, "Hash(<str>)", &s)) {
         parseHash(s); /* i.e., throw exception if not valid */
@@ -329,14 +328,6 @@ Expr evalValue(Expr e)
         return evalValue(e3);
     }
 
-    /* Application. */
-    if (ATmatch(e, "App(<term>, <term>)", &e2, &e3)) {
-        e2 = evalValue(e2);
-        if (!ATmatch(e2, "Lam(<str>, <term>)", &s, &e4))
-            throw badTerm("expecting lambda", e2);
-        return evalValue(substExpr(s, e3, e4));
-    }
-
     /* Execution primitive. */
 
     if (ATmatch(e, "Exec(<term>, <term>, [<list>])",
@@ -376,3 +367,120 @@ Expr evalValue(Expr e)
     /* Barf. */
     throw badTerm("invalid expression", e);
 }
+#endif
+
+
+Expr whNormalise(Expr e)
+{
+    char * s;
+    Expr e2, e3, e4, e5;
+
+    /* Normal forms. */
+    if (ATmatch(e, "Str(<str>)", &s) ||
+        ATmatch(e, "Bool(True)") || 
+        ATmatch(e, "Bool(False)") ||
+        ATmatch(e, "Lam(<str>, <term>)", &s, &e2) ||
+        ATmatch(e, "File(<str>, <term>, <term>)", &s, &e2, &e3) ||
+        ATmatch(e, "Derive(<term>, <term>, <term>, <term>)", &e2, &e3, &e4, &e5))
+        return e;
+
+    /* Application. */
+    if (ATmatch(e, "App(<term>, <term>)", &e2, &e3)) {
+        e2 = whNormalise(e2);
+        if (!ATmatch(e2, "Lam(<str>, <term>)", &s, &e4))
+            throw badTerm("expecting lambda", e2);
+        return whNormalise(substExpr(s, e3, e4));
+    }
+
+    throw badTerm("invalid expression", e);
+}
+
+
+Expr dNormalise(Expr e)
+{
+    e = whNormalise(e);
+    /* !!! todo */
+    return e;
+}
+
+
+Expr fNormalise(Expr e)
+{
+    e = dNormalise(e);
+
+    char * s;
+    Expr e2, e3;
+
+    if (ATmatch(e, "File(<str>, <term>, [<list>])", &s, &e2, &e3)) {
+
+        ATermList refs = (ATermList) e3, refs2 = ATempty;
+        while (!ATisEmpty(refs)) {
+            ATerm ref = ATgetFirst(refs);
+            refs2 = ATinsert(refs2, fNormalise(ref));
+            refs = ATgetNext(refs);
+        }
+        refs2 = ATreverse(refs2);
+
+        return ATmake("File(<str>, <term>, <term>)", s, e2, refs2);
+
+    }
+
+    else return e;
+}
+
+
+void writeContent(string path, Content content)
+{
+    char * s;
+
+    if (ATmatch(content, "Regular(<str>)", &s)) {
+
+        int fd; /* !!! close on exception */
+        fd = open(path.c_str(), O_CREAT | O_EXCL | O_WRONLY, 0666);
+        if (fd == -1)
+            throw SysError("creating file " + path);
+
+        int len = strlen(s);
+        if (write(fd, s, len) != len)
+            throw SysError("writing file " + path);
+
+        close(fd);
+    }
+    
+    else throw badTerm("ill-formed content", content);
+}
+
+
+struct RStatus
+{
+    /* !!! the comparator of this hash should match the semantics of
+       the file system */
+    map<string, Hash> paths;
+};
+
+
+static void realise2(RStatus & status, Expr e)
+{
+    char * s;
+    Content content;
+    ATermList refs;
+    
+    if (!ATmatch(e, "File(<str>, <term>, [<list>])", &s, &content, &refs))
+        throw badTerm("not f-normalised", e);
+
+    string path(s);
+
+    while (!ATisEmpty(refs)) {
+        realise2(status, ATgetFirst(refs));
+        refs = ATgetNext(refs);
+    }
+    
+    writeContent(path, content);
+}
+
+
+void realise(Expr e)
+{
+    RStatus status;
+    realise2(status, e);
+}

src/eval.hh

@@ -10,14 +10,14 @@ extern "C" {
 using namespace std;
 
 
-/* Abstract syntax of Nix expressions.
+/* \section{Abstract syntax of Nix expressions}
 
    An expression describes a (partial) state of the file system in a
    referentially transparent way.  The operational effect of
    evaluating an expression is that the state described by the
    expression is realised.
 
-     File : String * Content * [Expr] -> Expr
+     File : Path * Content * [Expr] -> Expr
 
    File(path, content, refs) specifies a file object (its full path
    and contents), along with all file objects referenced by it (that
@@ -25,12 +25,13 @@ using namespace std;
    self-referential.  This prevents us from having to deal with
    cycles.
 
-     Derive : String * Expr * [Expr] * [String] -> Expr
+     Derive : String * Path * [Expr] * [Expr] * [Expr] -> Expr
 
-   Derive(platform, builder, ins, outs) specifies the creation of new
+   Derive(platform, builder, ins, outs, env) specifies the creation of
-   file objects (in paths declared by `outs') by the execution of a
+   new file objects (in paths declared by `outs') by the execution of
-   program `builder' on a platform `platform'.  This execution takes
+   a program `builder' on a platform `platform'.  This execution takes
-   place in a file system state and in an environment given by `ins'.
+   place in a file system state given by `ins'.  `env' specifies a
+   mapping of strings to strings.
 
      Str : String -> Expr
 
@@ -40,39 +41,73 @@ using namespace std;
 
    Tuples of expressions.
 
+     [ !!! NOT IMPLEMENTED 
        Regular : String -> Content
        Directory : [(String, Content)] -> Content
-     Hash : String -> Content
+       (this complicates unambiguous normalisation)
+     ]
+     CHash : Hash -> Content
 
-   File content, given either explicitly or implicitly through a cryptographic hash.
+   File content, given either in situ, or through an external reference
+   to the file system or url-space decorated with a hash to preserve purity.
 
-   The set of expressions in {\em $f$-normal form} is as follows:
+   DISCUSSION: the idea is that a Regular/Directory is interchangeable
+   with its CHash.  This would appear to break referential
+   transparency, e.g., Derive(..., ..., [...CHash(h)...], ...) can
+   only be reduced in a context were the Regular/Directory equivalent
+   of Hash(h) is known.  However, CHash should be viewed strictly as a
+   shorthand; that is, when we export an expression containing a
+   CHash, we should also export the file object referenced by that
+   CHash.
 
-     File : String * Content * [FExpr] -> FExpr
 
-   These are completely evaluated Nix expressions.
+   \section{Reduction rules}
 
-   The set of expressions in {\em $d$-normal form} is as follows:
+   ...
+
+
+   \section{Normals forms}
+
+   An expression is in {\em weak head normal form} if it is a lambda,
+   a string or boolean value, or a File or Derive value.
+
+   An expression is in {\em $d$-normal form} if it matches the
+   signature FExpr:
 
      File : String * Content * [DExpr] -> DExpr
-     Derive : String * DExpr * [Tup] * [String] -> DExpr
+     Derive : String * Path * [Tup] * [Tup2] -> DExpr
 
      Tup : Str * DExpr -> Tup
      Tup : Str * Str -> Tup
 
+     Tup : Str * Str -> Tup2
+
      Str : String -> Str
 
    These are Nix expressions in which the file system result of Derive
    expressions has not yet been computed.  This is useful for, e.g.,
    querying dependencies.
 
+   An expression is in {\em $f$-normal form} if it matches the
+   signature FExpr:
+
+     File : String * Content * [FExpr] -> FExpr
+
+   These are completely evaluated Nix expressions.
+
 */
 
 typedef ATerm Expr;
+typedef ATerm Content;
 
 
-/* Evaluate an expression. */
+/* Expression normalisation. */
-Expr evalValue(Expr e);
+Expr whNormalise(Expr e);
+Expr dNormalise(Expr e);
+Expr fNormalise(Expr e);
+
+/* Realise a $f$-normalised expression in the file system. */
+void realise(Expr e);
 
 /* Return a canonical textual representation of an expression. */
 string printExpr(Expr e);

src/test.cc

@@ -11,13 +11,27 @@
 #include "globals.hh"
 
 
-void evalTest(Expr e)
+typedef Expr (* Normaliser) (Expr);
+
+
+void eval(Normaliser n, Expr e)
 {
-    e = evalValue(e);
+    e = n(e);
     cout << (string) hashExpr(e) << ": " << printExpr(e) << endl;
 }
 
 
+void evalFail(Normaliser n, Expr e)
+{
+    try {
+        e = n(e);
+        abort();
+    } catch (Error e) {
+        cout << "error (expected): " << e.what() << endl;
+    }
+}
+
+
 struct MySink : DumpSink
 {
     virtual void operator () (const unsigned char * data, unsigned int len)
@@ -90,19 +104,48 @@ void runTests()
 
     /* Expression evaluation. */
 
-    evalTest(ATmake("Str(\"Hello World\")"));
+    eval(whNormalise,
-    evalTest(ATmake("Bool(True)"));
+        ATmake("Str(\"Hello World\")"));
-    evalTest(ATmake("Bool(False)"));
+    eval(whNormalise,
-    evalTest(ATmake("App(Lam(\"x\", Var(\"x\")), Str(\"Hello World\"))"));
+        ATmake("Bool(True)"));
-    evalTest(ATmake("App(App(Lam(\"x\", Lam(\"y\", Var(\"x\"))), Str(\"Hello World\")), Str(\"Hallo Wereld\"))"));
+    eval(whNormalise,
-    evalTest(ATmake("App(Lam(\"sys\", Lam(\"x\", [Var(\"x\"), Var(\"sys\")])), Str(\"i686-suse-linux\"))"));
+        ATmake("Bool(False)"));
+    eval(whNormalise,
+        ATmake("App(Lam(\"x\", Var(\"x\")), Str(\"Hello World\"))"));
+    eval(whNormalise,
+        ATmake("App(App(Lam(\"x\", Lam(\"y\", Var(\"x\"))), Str(\"Hello World\")), Str(\"Hallo Wereld\"))"));
+    eval(whNormalise,
+        ATmake("App(Lam(\"sys\", Lam(\"x\", [Var(\"x\"), Var(\"sys\")])), Str(\"i686-suse-linux\"))"));
 
+    evalFail(whNormalise,
+        ATmake("Foo(123)"));
+
+    string builder1fn = absPath("./test-builder-1.sh");
+    Hash builder1h = hashFile(builder1fn);
+
+    string fn1 = nixValues + "/builder-1.sh";
+    Expr e1 = ATmake("File(<str>, ExtFile(<str>, <str>), [])", 
+        fn1.c_str(),
+        builder1h.c_str(),
+        builder1fn.c_str());
+    eval(fNormalise, e1);
+
+    string fn2 = nixValues + "/refer.txt";
+    Expr e2 = ATmake("File(<str>, Regular(<str>), [<term>])",
+        fn2.c_str(),
+        ("I refer to " + fn1).c_str(),
+        e1);
+    eval(fNormalise, e2);
+
+    realise(e2);
+
+#if 0
     Hash builder1 = addValue("./test-builder-1.sh");
 
     Expr e1 = ATmake("Exec(Str(<str>), Hash(<str>), [])",
         thisSystem.c_str(), ((string) builder1).c_str());
 
-    evalTest(e1);
+    eval(e1);
 
     Hash builder2 = addValue("./test-builder-2.sh");
 
@@ -110,14 +153,15 @@ void runTests()
         "Exec(Str(<str>), Hash(<str>), [Tup(Str(\"src\"), <term>)])",
         thisSystem.c_str(), ((string) builder2).c_str(), e1);
 
-    evalTest(e2);
+    eval(e2);
 
     Hash h3 = addValue("./test-expr-1.nix");
     Expr e3 = ATmake("Deref(Hash(<str>))", ((string) h3).c_str());
 
-    evalTest(e3);
+    eval(e3);
 
     deleteValue(h3);
+#endif
 }