halt_error/1 crashes when called more than once

[pkoppstein]

The bash script output below shows the jq version numbers and the crash report:

function json {
cat <<EOF
{"err": 1}
{}
{}
{"err": 1}
EOF
}

for jq in jq1.6 jqMaster
do
  $jq --version
  json | $jq -r '
    if(.err==null) then 0
    else "Failed to process " | halt_error(1) 
    end'
  echo
done

Output:

jq-1.6
Failed to process 0
0
Failed to process jq(55072,0x110bef5c0) malloc: *** error for object 0x7fcc38c03d70: pointer being freed was not allocated
jq(55072,0x110bef5c0) malloc: *** set a breakpoint in malloc_error_break to debug
./halt_error.bug: line 26: 55071 Done                    json
     55072 Abort trap: 6           | $jq -r '
  if(.err==null) 
  then 0
  else "Failed to process " | halt_error(1) 
  end'

jq-1.6-129-g80052e5-dirty
jq: error: Failed to process 0
0
jqMaster(55076,0x1100485c0) malloc: *** error for object 0x3: pointer being freed was not allocated
jqMaster(55076,0x1100485c0) malloc: *** set a breakpoint in malloc_error_break to debug
./halt_error.bug: line 26: 55075 Done                    json
     55076 Abort trap: 6           | $jq -r '
  if(.err==null) 
  then 0
  else "Failed to process " | halt_error(1) 
  end'

Environment: Mac OS

[itchyny]

I cannot reproduce the issue on macOS 13.4.1.

[emanuele6]

On my Linux 6.1.35-lts, I get no crash for jq master:

$ json | ./jq -r '
    if(.err==null) then 0
    else "Failed to process " | halt_error(1) 
    end'
Failed to process 0
0
Failed to process 

But I get a crash for jq 1.6: (note, also the 0s were coloured as numbers in jq master, but not in jq 1.6)

$ json | jq -r '
    if(.err==null) then 0
    else "Failed to process " | halt_error(1) 
    end'
Failed to process 0
0
Failed to process malloc_consolidate(): unaligned fastbin chunk detected
Aborted (core dumped)

So it must have been fixed.

---

But I think there is still a behaviour problem, because I think halt_error(1) is supposed to exit immediately with exit status 1 instead of going to the next input

I don't think it should be possible to "call halt_error more than once"

$ json | jqExpectedBehaviour -r '
    if(.err==null) then 0
    else "Failed to process " | halt_error(1) 
    end'
Failed to process 
$ echo "$?"
1

[emanuele6]

But I think there is still a behaviour problem, because I think halt_error(1) is supposed to exit immediately with exit status 1 instead of going to the next input

This patch would fix that:

diff --git a/src/main.c b/src/main.c
index 1cc6542..462c62a 100644
--- a/src/main.c
+++ b/src/main.c
@@ -681,6 +681,8 @@ int main(int argc, char* argv[]) {
         ret = process(jq, value, jq_flags, dumpopts);
         if (ret <= 0 && ret != JQ_OK_NO_OUTPUT)
           last_result = (ret != JQ_OK_NULL_KIND);
+        if (jq_halted(jq))
+          break;
         continue;
       }
 

bash-5.1$ # jq 1.6
bash-5.1$ <<< '{} {}' jq '1, ("halt!\n" | halt_error(10)), 3'; printf 'ret: %s\n' "$?"
1
halt!
1
halt!
ret: 10
bash-5.1$ # jq patched
bash-5.1$ <<< '{} {}' ./jq '1, ("halt!\n" | halt_error(10)), 3'; printf 'ret: %s\n' "$?"
1
halt!
ret: 10

---

Or maybe it's just the man page that is worded incorrectly:

   halt_error, halt_error(exit_code)
       Stops the jq program with no further outputs. The input will be printed
       on stderr as raw output (i.e., strings will  not  have  double  quotes)
       with no decoration, not even a newline.

       The given exit_code (defaulting to 5) will be jq´s exit status.

       For example, "Error: somthing went wrong\n"|halt_error(1).

From the man page, it sounds like the program should stop as it does with my patch, and that is how i always expected it work.

gojq behaves like jq on master; and fq behaves like my patched version:

bash-5.1$ <<< '{} {}' gojq '1, ("halt!\n" | halt_error(10)), 3'; printf 'ret: %s\n' "$?"
1
halt!
1
halt!
ret: 10

bash-5.1$ <<< '{} {}' fq ...

bash-5.1$ <<< '{} {}' fq '1, ("halt!\n" | halt_error(10)), 3'; printf 'ret: %s\n' "$?"
1
error: halt!

ret: 10

I think the correct behaviour is that the entire program stops like the documentation says, and that requiring users to do jq -n 'inputs | ... | halt_error(1)' to make halt_error actually stop the entire program is an unnecessary annoyance, but since gojq and jq master works this way, maybe we could just change the documenation if we prefer.

EDIT: fq only works that way because it implicitly slurps if there are multiple elements in the inputs, so the code is only ran once with [{},{}].

[itchyny]

The reported bug seems to be fixed by cf4b48c. Not sure where in the patch is related, but I double checked that cf4b48c~1 crashes but cf4b48c does not. Since the original bug is already fixed, closing this issue.

@emanuele6 Good find. I agree the behavior is confusing, but I think this is an intended behavior. And already reported by #2368.