Re: Fwd (potential regression in 5.0.3): Bug#732726: zsh function freeze

[Bart Schaefer <schaefer@brasslantern.com>]

On Dec 20,  6:19pm, Bart Schaefer wrote:
}
} +       * 31549: Src/exec,c, Src/zsh.h: replace ad-hoc subsh_close file
} +       descriptor for pipes with new addfilelist() job-based mechanism.

Yep, that's the one.

I compiled commit 709dbbbda82efde2020d9d67a19687c101b91570 and was
able to reproduce the problem.  The immediately preceding commit
39ab9952e8255cb99e9c0abcc8bbec43158a55d7 does not show the bug.

Non-interactive shells also do not show the bug, which may be why
"make check" didn't flag anything.

The problem seems to be with this hunk specifically:

@@ -1744,8 +1742,6 @@ execpline2(Estate state, wordcode pcode,
 	execpline2(state, *state->pc++, how, pipes[0], output, last1);
 	list_pipe = old_list_pipe;
 	cmdpop();
-	zclose(pipes[0]);
-	subsh_close = -1;
     }
 }
 

The problem is that pipes[0] isn't always added to the list of files
for the job; sometimes it really does need to be closed there.


diff --git a/Src/exec.c b/Src/exec.c
index dccdc2b..4480033 100644
--- a/Src/exec.c
+++ b/Src/exec.c
@@ -1691,6 +1691,7 @@ execpline2(Estate state, wordcode pcode,
 	execcmd(state, input, output, how, last1 ? 1 : 2);
     else {
 	int old_list_pipe = list_pipe;
+	int subsh_close = -1;
 	Wordcode next = state->pc + (*state->pc), pc;
 	wordcode code;
 
@@ -1738,6 +1739,7 @@ execpline2(Estate state, wordcode pcode,
 	} else {
 	    /* otherwise just do the pipeline normally. */
 	    addfilelist(NULL, pipes[0]);
+	    subsh_close = pipes[0];
 	    execcmd(state, input, pipes[1], how, 0);
 	}
 	zclose(pipes[1]);
@@ -1750,6 +1752,8 @@ execpline2(Estate state, wordcode pcode,
 	execpline2(state, *state->pc++, how, pipes[0], output, last1);
 	list_pipe = old_list_pipe;
 	cmdpop();
+	if (subsh_close != pipes[0])
+	    zclose(pipes[0]);
     }
 }