Re: [bug-gawk] Computed regex and getline bug / issue

[Andrew J. Schorr]

Hi Arnold,

On Fri, May 09, 2014 at 11:54:36AM +0300, Aharon Robbins wrote:
> Thanks everyone for their help cutting the test case down, and to Andy for
> setting up some test cases to drop into the test suite.  The fix is below.
> I will push it into the repo shortly.

I'm afraid my test cases were not comprehensive.

> diff --git a/io.c b/io.c
> index 2bb8f28..dab1c07 100644
> --- a/io.c
> +++ b/io.c
> @@ -206,6 +206,7 @@ typedef enum { CLOSE_ALL, CLOSE_TO, CLOSE_FROM } 
> two_way_close_type;
>  #define at_eof(iop)     (((iop)->flag & IOP_AT_EOF) != 0)
>  #define has_no_data(iop)        ((iop)->dataend == NULL)
>  #define no_data_left(iop)    ((iop)->off >= (iop)->dataend)
> +#define buffer_has_all_data(iop) ((iop)->dataend - (iop)->off == 
> (iop)->public.sbuf.st_size)
>  
>  /*
>   * The key point to the design is to split out the code that searches through
> @@ -3474,6 +3475,8 @@ get_a_record(char **out,        /* pointer to pointer 
> to data */
>               iop->flag &= ~IOP_AT_START;
>               if (ret == REC_OK)
>                       break;
> +             if (ret == TERMNEAREND && buffer_has_all_data(iop))
> +                     break;
>  
>               /* need to add more data to buffer */
>               /* shift data down in buffer */
> 

The logic here depends on having access to the st_size.  When reading data
from stdin, this doesn't seem to work.  We need another test case that
reads from stdin instead of from a file.  After rebuilding with your patch,
I see this:

bash-4.2$ ./gawk -f test/rsgetline.awk < test/rsgetline.in
[1] [,]
-1-
[2] [,]
bash-4.2$ cat test/rsgetline.in
1,2,bash-4.2$ 
bash-4.2$ printf '1,2,' | ./gawk -f test/rsgetline.awk
[1] [,]
-0-
[1] [,]
bash-4.2$ 

So the fix works when reading from a regular file, but apparently not
from stdin.

Regards,
Andy