Implement $(cmd) subcommand syntax in commands (issue #27)

Add $() for inline subcommand substitution that runs a command and
inserts its stdout (trailing newlines stripped) into the outer command
string. Supports nesting, and mixing with ${expr} and $NAME.

issues/27-subcommand-syntax.md

@@ -1,6 +1,6 @@
 # 27 — Subcommand syntax in commands
 
-**Status:** open
+**Status:** done
 
 ## Problem
 

lcmd.c

@@ -1039,6 +1039,30 @@ int lushCmd_command (lua_State *L) {
 }
 
 
+/* ===== lushCmd_subcmd ===== */
+
+/*
+** Subcommand substitution: run a command, return stdout with trailing
+** newlines stripped.  Used for $(cmd) inside command strings.
+*/
+int lushCmd_subcmd (lua_State *L) {
+  size_t len;
+  const char *s;
+  /* run the command — pushes {code, stdout, stderr} table */
+  lushCmd_command(L);
+  /* extract .stdout from result table */
+  lua_getfield(L, -1, "stdout");
+  lua_remove(L, -2);  /* remove the table, keep stdout string */
+  /* strip trailing newlines */
+  s = lua_tolstring(L, -1, &len);
+  while (len > 0 && s[len - 1] == '\n')
+    len--;
+  lua_pushlstring(L, s, len);
+  lua_remove(L, -2);  /* remove original stdout string */
+  return 1;
+}
+
+
 /* ===== lushCmd_interactive ===== */
 
 int lushCmd_interactive (lua_State *L) {

lcmd.h

@@ -14,11 +14,13 @@
 #define LUSH_OP_INTERACTIVE  1
 #define LUSH_OP_GETENV       2
 #define LUSH_OP_SETENV       3
-#define LUSH_OP_COUNT        4
+#define LUSH_OP_SUBCMD       4
+#define LUSH_OP_COUNT        5
 
 int lushCmd_command (lua_State *L);
 int lushCmd_interactive (lua_State *L);
 int lushCmd_getenv (lua_State *L);
 int lushCmd_setenv (lua_State *L);
+int lushCmd_subcmd (lua_State *L);
 
 #endif

linit.c

@@ -45,7 +45,7 @@ static const luaL_Reg stdlibs[] = {
 
 /*
 ** Store shell functions in a registry table at LUA_RIDX_LUSH.
-** Integer keys 1..4 hold the C functions (accessed by OP_LUSH).
+** Integer keys 1..5 hold the C functions (accessed by OP_LUSH).
 ** String key "builtins" holds the builtins table (set by luaopen_builtins).
 */
 static void opencommand (lua_State *L) {
@@ -58,6 +58,8 @@ static void opencommand (lua_State *L) {
   lua_rawseti(L, -2, LUSH_OP_GETENV + 1);
   lua_pushcfunction(L, lushCmd_setenv);
   lua_rawseti(L, -2, LUSH_OP_SETENV + 1);
+  lua_pushcfunction(L, lushCmd_subcmd);
+  lua_rawseti(L, -2, LUSH_OP_SUBCMD + 1);
   lua_rawseti(L, LUA_REGISTRYINDEX, LUA_RIDX_LUSH);
 }
 

llex.c

@@ -193,6 +193,7 @@ void luaX_setinput (lua_State *L, LexState *ls, ZIO *z, TString *source,
      so they cannot be collected */
   ls->cmd_mode = 0;
   ls->cmd_envvar = NULL;
+  ls->cmd_subcmd = 0;
   ls->envn = luaS_newliteral(L, LUA_ENV);  /* get env string */
   ls->brkn = luaS_newliteral(L, "break");  /* get "break" string */
 #if defined(LUA_COMPAT_GLOBAL)
@@ -516,6 +517,15 @@ static int read_command_body (LexState *ls, SemInfo *seminfo) {
           ls->saved_cmd_mode = ls->cmd_mode;  /* save for readcommandcont */
           return interactive ? TK_INTERACTIVE : TK_COMMAND;
         }
+        else if (ls->current == '(') {
+          next(ls);  /* skip '(' */
+          /* $() subcommand interpolation */
+          seminfo->ts = luaX_newstring(ls, luaZ_buffer(ls->buff),
+                                           luaZ_bufflen(ls->buff));
+          ls->saved_cmd_mode = ls->cmd_mode;
+          ls->cmd_subcmd = 1;
+          return interactive ? TK_INTERACTIVE : TK_COMMAND;
+        }
         else if (lislalpha(ls->current)) {
           /* $NAME envvar expansion */
           seminfo->ts = luaX_newstring(ls, luaZ_buffer(ls->buff),
@@ -532,6 +542,17 @@ static int read_command_body (LexState *ls, SemInfo *seminfo) {
         }
         break;
       }
+      case ')': {
+        if (ls->cmd_mode == 3) {
+          next(ls);  /* skip closing ')' */
+          seminfo->ts = luaX_newstring(ls, luaZ_buffer(ls->buff),
+                                           luaZ_bufflen(ls->buff));
+          ls->cmd_mode = 0;
+          return TK_COMMAND;
+        }
+        save_and_next(ls);  /* literal ')' in non-subcommand mode */
+        break;
+      }
       case '`': {
         if (interactive) {
           /* backtick is literal in interactive mode */

llex.h

@@ -81,8 +81,9 @@ typedef struct LexState {
   TString *brkn;  /* "break" name (used as a label) */
   TString *glbn;  /* "global" name (when not a reserved word) */
   TString *cmd_envvar;       /* envvar name for $NAME in commands */
-  lu_byte cmd_mode;  /* 0=normal, 1=backtick command, 2=interactive command */
+  lu_byte cmd_mode;  /* 0=normal, 1=backtick, 2=interactive, 3=subcommand */
   lu_byte saved_cmd_mode;  /* cmd_mode to restore after interpolation */
+  lu_byte cmd_subcmd;  /* 1 when current interpolation is $() subcommand */
 } LexState;
 
 

lparser.c

@@ -556,23 +556,59 @@ static void singlevar (LexState *ls, expdesc *var) {
 static void codeenvget (LexState *ls, expdesc *v, TString *name);
 
 /*
-** Parse a single interpolation fragment in a command: either ${expr}
+** Parse a single interpolation fragment in a command: ${expr}, $NAME,
-** or $NAME.  Emits tostring(<value>), reads the continuation fragment,
+** or $(cmd).  Emits tostring(<value>), reads the continuation fragment,
 ** and pushes it as a string constant.
 ** For ${expr}: advances past the command token, parses expr, checks '}'.
 ** For $NAME:  the identifier was already collected by the lexer into
 **             cmd_envvar — no tokens to consume, just emit getenv(name).
+** For $(cmd): enters subcommand mode (cmd_mode=3), parses the inner
+**             command with its own interpolation loop, emits subcmd().
 */
 static void parseinterp (LexState *ls, TString *tsname, int *nconcat) {
   FuncState *fs = ls->fs;
   expdesc interp, tostrfn;
   int tostr_base;
   TString *envname = ls->cmd_envvar;
+  lu_byte subcmd = ls->cmd_subcmd;
   ls->cmd_envvar = NULL;
+  ls->cmd_subcmd = 0;
   buildglobal(ls, tsname, &tostrfn);
   luaK_exp2nextreg(fs, &tostrfn);
   tostr_base = tostrfn.u.info;
-  if (envname != NULL) {
+  if (subcmd) {
+    /* $(cmd) — parse inner command as subcommand */
+    expdesc subfunc, cmdstr;
+    int subbase, subnconcat = 0;
+    lu_byte outer_saved = ls->saved_cmd_mode;
+    TString *sub_tsname = NULL;
+    codelushfunc(fs, LUSH_OP_SUBCMD, &subfunc);
+    subbase = subfunc.u.info;
+    /* enter subcommand lexing mode (terminated by ')') */
+    ls->saved_cmd_mode = 3;
+    luaX_readcommandcont(ls);  /* read first fragment of inner command */
+    codestring(&cmdstr, ls->t.seminfo.ts);
+    luaK_exp2nextreg(fs, &cmdstr);
+    subnconcat++;
+    /* inner interpolation loop */
+    while (ls->cmd_mode != 0) {
+      if (sub_tsname == NULL)
+        sub_tsname = luaX_newstring(ls, "tostring", 8);
+      parseinterp(ls, sub_tsname, &subnconcat);
+    }
+    /* concatenate inner fragments if needed */
+    if (subnconcat > 1) {
+      int first = subbase + 1;
+      luaK_codeABC(fs, OP_CONCAT, first, subnconcat, 0);
+      fs->freereg = cast_byte(first + 1);
+    }
+    /* call subcmd(inner_cmd_string) */
+    init_exp(&interp, VCALL, luaK_codeABC(fs, OP_CALL, subbase, 2, 2));
+    fs->freereg = cast_byte(subbase + 1);
+    /* restore outer command context */
+    ls->saved_cmd_mode = outer_saved;
+  }
+  else if (envname != NULL) {
     /* $NAME — emit getenv(name) */
     codeenvget(ls, &interp, envname);
   }

testes/lush/interpolation.lua

@@ -109,4 +109,54 @@ do
   assert(r.stdout == "$HOME\n", "got: " .. r.stdout)
 end
 
+-- $(cmd) subcommand: basic
+do
+  local r = `echo $(echo hello)`
+  assert(r.stdout == "hello\n", "got: " .. r.stdout)
+end
+
+-- $(cmd) strips trailing newline from subcommand output
+do
+  local r = `echo -$(echo world)-`
+  assert(r.stdout == "-world-\n", "got: " .. r.stdout)
+end
+
+-- $(cmd) multiple subcommands
+do
+  local r = `echo $(echo aaa) $(echo bbb)`
+  assert(r.stdout == "aaa bbb\n", "got: " .. r.stdout)
+end
+
+-- $(cmd) nested subcommand
+do
+  local r = `echo $(echo $(echo deep))`
+  assert(r.stdout == "deep\n", "got: " .. r.stdout)
+end
+
+-- $(cmd) with ${expr} inside subcommand
+do
+  local r = `echo $(echo ${1+1})`
+  assert(r.stdout == "2\n", "got: " .. r.stdout)
+end
+
+-- $(cmd) with $NAME inside subcommand
+do
+  local r = `echo $(echo $USER)`
+  assert(r.stdout == $USER .. "\n", "got: " .. r.stdout)
+end
+
+-- $(cmd) mixed with $NAME and ${expr}
+do
+  local x = 42
+  local r = `echo $(echo sub) $USER ${x}`
+  local expected = "sub " .. $USER .. " 42\n"
+  assert(r.stdout == expected, "got: " .. r.stdout)
+end
+
+-- escaped \$(cmd) remains literal
+do
+  local r = `echo \$(pwd)`
+  assert(r.stdout == "$(pwd)\n", "got: " .. r.stdout)
+end
+
 print "OK"