6eaf8dedd
Denis Vlasenko
2008-06-17
Subject
hush: fix "for a in; do echo 'I should never run'; done" bug
Binary
No binary (build failed)
Architecture & Compiler Flags
—
Build Log
(log yok)
Commit Diff GitHub →
shell/hush.c+10 -8
| @@ -2004,7 +2004,7 @@ static int run_list(struct pipe *pi) | |
| || (rpipe->res_word == RES_FOR && rpipe->next->res_word != RES_IN) | |
| ) { | |
| /* TODO: what is tested in the first condition? */ | |
| - | syntax("malformed for"); /* 2nd condition: not followed by IN */ |
| + | syntax("malformed for"); /* 2nd condition: FOR not followed by IN */ |
| debug_printf_exec("run_list lvl %d return 1\n", run_list_level); | |
| return 1; | |
| } | |
| @@ -2898,7 +2898,8 @@ static int done_word(o_string *word, struct p_context *ctx) | |
| ctx->pending_redirect = NULL; | |
| } | |
| #if ENABLE_HUSH_LOOPS | |
| - | if (ctx->res_w == RES_FOR) { /* comment? */ |
| + | /* comment? is it forcing "for" to have just one word (variable name)? */ |
| + | if (ctx->res_w == RES_FOR) { |
| //TESTING | |
| //looks like (word->length == 0 && !word->nonnull) is true here, always | |
| //(due to o_reset). done_word would return at once. Why then? | |
| @@ -2910,8 +2911,8 @@ static int done_word(o_string *word, struct p_context *ctx) | |
| return 0; | |
| } | |
| - | /* The only possible error here is out of memory, in which case |
| - | * xmalloc exits. */ |
| + | /* Command (member of a pipe) is complete. The only possible error here |
| + | * is out of memory, in which case xmalloc exits. */ |
| static int done_command(struct p_context *ctx) | |
| { | |
| /* The child is really already in the pipe structure, so | |
| @@ -2949,7 +2950,6 @@ static int done_command(struct p_context *ctx) | |
| static void done_pipe(struct p_context *ctx, pipe_style type) | |
| { | |
| - | struct pipe *new_p; |
| int not_null; | |
| debug_printf_parse("done_pipe entered, followup %d\n", type); | |
| @@ -2960,9 +2960,11 @@ static void done_pipe(struct p_context *ctx, pipe_style type) | |
| ctx->ctx_inverted = 0; | |
| /* Without this check, even just <enter> on command line generates | |
| * tree of three NOPs (!). Which is harmless but annoying. | |
| - | * IOW: it is safe to do it unconditionally. */ |
| - | if (not_null) { |
| - | new_p = new_pipe(); |
| + | * IOW: it is safe to do it unconditionally. |
| + | * RES_IN case is for "for a in; do ..." (empty IN set) |
| + | * to work. */ |
| + | if (not_null || ctx->pipe->res_word == RES_IN) { |
| + | struct pipe *new_p = new_pipe(); |
| ctx->pipe->next = new_p; | |
| ctx->pipe = new_p; | |
| ctx->child = NULL; /* needed! */ | |
?+0 -0
?+0 -0