Re: Changing $0 after gsub() breaks output

[Yasuhiro Yamada]

Thanks for identifying and fixing the problem.

> +                        * (Test case was gawk 'gusb(/./, "@") && $0=$1'). So 
> we save

I noticed a minor typo by the way.
s/gusb/gsub/

Thanks,

On Sun, Jan 29, 2023 at 7:53 AM <arnold@skeeve.com> wrote:
>
> Hi.
>
> Thank you for the bug report and easy reproducer. The patch
> below fixes the problem. It will be in Git in the next
> day or two.
>
> Arnold
>
> Yasuhiro Yamada <yamada@gr3.ie> wrote:
>
> > Hi.
> > Executing gsub() without an action and making changes to $0 will
> > corrupt the output.
> > This behavior seems like a bug.
> >
> >     $ ./gawk --version
> >     GNU Awk 5.2.1, API 3.2, PMA Avon 8-g1
> >     ...
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1'
> >     q  <===== broken output
> >
> > Interestingly, the result is different for each run.
> >
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1' | od -tx1c
> >     0000000  b0  61  99  0a
> >              260   a 231  \n
> >     0000004
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1' | od -tx1c
> >     0000000  c0  e7  f7  0a
> >             300 347 367  \n
> >     0000004
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1' | od -tx1c
> >     0000000  c0  07  14  0a
> >             300  \a 024  \n
> >     0000004
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1' | od -tx1c
> >     0000000  c0  57  76  0a
> >             300   W   v  \n
> >     0000004
> >
> > Older versions do NOT reproduce this issue.
> > Also, the outputs are as expected and intuitive.
> >
> >     $ ./gawk --version
> >     GNU Awk 4.1.4, API: 1.1
> >     ...
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1'
> >     @@@
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1' | od -tx1c
> >     0000000  40  40  40  0a
> >               @   @   @  \n
> >     0000004
> >
> > This issue seems to occur in v4.2.0 and later.
> >
> >     $ ./gawk --version
> >     GNU Awk 4.2.0, API: 2.0
> >     ...
> >     $ echo abc | ./gawk 'gsub(".","@") && $0=$1' | od -tx1c
> >     0000000  e0  4f  fc  0a
> >             340   O 374  \n
> >     0000004
> >
> > My environment is
> >
> >     $ uname -a
> >     Linux ip-172-31-9-222 5.4.0-1093-aws #102~18.04.2-Ubuntu SMP Wed
> > Dec 7 00:31:59 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux
> >     $ gcc --version
> >     gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
>
> ---------------------------------------------
> diff --git a/interpret.h b/interpret.h
> index 4540d302..5adb229b 100644
> --- a/interpret.h
> +++ b/interpret.h
> @@ -872,16 +874,30 @@ mod:
>                         break;
>
>                 case Op_assign:
> +               {
> +                       NODE *save_lhs;
> +
>                         lhs = POP_ADDRESS();
>                         r = TOP_SCALAR();
> -                       unref(*lhs);
> +                       /*
> +                        * 1/2023:
> +                        * The old NODE pointed to by *lhs has to be freed.
> +                        * But we can't free it too early, in case it's both 
> $0 and $1
> +                        * (Test case was gawk 'gusb(/./, "@") && $0=$1'). So 
> we save
> +                        * the old one, and after the assignment, we free it, 
> since
> +                        * $0 and $1 have the same stptr value but only $0 
> has MALLOC
> +                        * in the flags. Whew!
> +                        */
> +                       save_lhs = *lhs;
>                         if (r->type == Node_elem_new) {
>                                 DEREF(r);
>                                 r = dupnode(Nnull_string);
>                         }
>                         UPREF(r);
>                         UNFIELD(*lhs, r);
> +                       unref(save_lhs);
>                         REPLACE(r);
> +               }
>                         break;
>
>                 case Op_subscript_assign: