Re: [bug-gawk] 'watch' in gawk debugger gives wrong NR

[Joep van Delft]

Thanks for looking into this, Arnold.  The suspicious behavior, however,
would be one or two "continue"s further. Please consider the following 
run:


| % gawk --version
| GNU Awk 4.1.3, API: 1.1 (GNU MPFR 3.1.3-p4, GNU MP 6.1.0)
|
| % cat dgawk.commands
| list
| watch c "c>0"
| break 6
| display NR
| display nr
| display c
| display $0
| run
|
| % gawk -D -f script.awk tst.txt
| gawk> source dgawk.commands
| 1       #!/usr/bin/gawk -f
| 2       BEGIN {
| 3           c = 0
| 4       }
| 5
| 6       /apple/ {
| 7           nr=NR
| 8           c++
| 9           #printf "[c, NR] = [%s, %s]\n", c, NR
| 10      }
| 11
| 12      END {
| 13          print c
| 14      }
| Watchpoint 1: c
| Breakpoint 2 set at file `script.awk', line 6
| 1: NR = 0
| 2: nr = untyped variable
| 3: c = untyped variable
| 4: $0 = ""
| Starting program:
| Stopping in Rule ...
| Breakpoint 2, main() at `script.awk':6
| 6       /apple/ {
| 1: NR = 1
| 2: nr = untyped variable
| 3: c = 0
| 4: $0 = "1 1 pear"
|
| gawk> continue
| Breakpoint 2, main() at `script.awk':6
| 6       /apple/ {
| 1: NR = 2
| 2: nr = untyped variable
| 3: c = 0
| 4: $0 = "2 1 apple 4"

Here I would expect that the upcoming halt would be on the watchpoint c,
as $0 matches PATTERN. Not so. Notice that c has already been increased,
but that the next line is already processed (according to NR and 0):

| gawk> continue
| Breakpoint 2, main() at `script.awk':6
| 6       /apple/ {
| 1: NR = 3
| 2: nr = 2
| 3: c = 1
| 4: $0 = "3 2 pear"

While we are already in the next line, the watchpoint gets triggered,
and consequently has a wrong NR and 0:

| gawk> continue
| Watchpoint 1: c
|   Old value: untyped variable
|   New value: 1
| main() at `script.awk':6
| 6       /apple/ {
| 1: NR = 3
| 2: nr = 2
| 3: c = 1
| 4: $0 = "3 2 pear"


Kind regards,

Joep


Aharon Robbins schreef op 2015-11-23 20:05:
> Hi.
>
> Thanks for your note. Here is what I got when I tried your example
> using the gawk-4.1-stable branch in git.  I don't think there's
> a bug.
>
> Run gawk:
>
> | Script started on Mon Nov 23 20:56:29 2015
> | [aahz:pts/6 gawk.git]$ ./gawk -D -f testprog.awk data.in
>
> List:
>
> | gawk> list
> | 1       #! /usr/bin/gawk -f
> | 2       BEGIN {
> | 3            c = 0
> | 4       }
> | 5
> | 6       /apple/ {
> | 7            nr = NR
> | 8            c++
> | 9            # printf "[c, NR] = [%s, %s]\n", c, NR
> | 10      }
> | 11
> | 12      END {
> | 13          print c
> | 14      }
>
> Setup watchpoints, display variables, and breakpoint:
>
> | gawk> watch c
> | Watchpoint 1: c
> | gawk> display nr
> | 1: nr = untyped variable
> | gawk> display NR
> | 2: NR = 0
> | gawk> display c
> | 3: c = untyped variable
> | gawk> break 6
> | Breakpoint 2 set at file `testprog.awk', line 6
>
> Run!
>
> | gawk> run
> | Starting program:
> | Stopping in Rule ...
> | Breakpoint 2, main() at `testprog.awk':6
> | 6       /apple/ {
> | 1: nr = untyped variable
> | 2: NR = 1
> | 3: c = 0
> | gawk> print $0
> | $0 = "1 1 pear"
>
> So far, so good. First record doesn't match, nr has not been assigned
> to, and NR is correct.
>
> | gawk> continue
> | Watchpoint 1: c
> |   Old value: untyped variable
> |   New value: 0
>
> Hmm. This is a little strange. Maybe I need to look at why the
> watchpoint was triggered.
>
> | main() at `testprog.awk':6
> | 6       /apple/ {
> | 1: nr = untyped variable
> | 2: NR = 1
> | 3: c = 0
> | gawk> print $0
> | $0 = "1 1 pear"
>
> So, we're still in the first record. NR is ok.
>
> | gawk> continue
> | Breakpoint 2, main() at `testprog.awk':6
> | 6       /apple/ {
> | 1: nr = untyped variable
> | 2: NR = 2
> | 3: c = 0
> | gawk> print $0
> | $0 = "2 1 apple 4"
>
> Continue on:
>
> | gawk> continue
> | Breakpoint 2, main() at `testprog.awk':6
> | 6       /apple/ {
> | 1: nr = 2
> | 2: NR = 3
> | 3: c = 1
> | gawk> print $0
> | $0 = "3 2 pear"
>
> We're at record 3. It won't match, but NR is displayed correctly.
>
> | gawk> exit
> | error: unknown command - "exit", try help
> | gawk> ^D
>
> I think display of NR is OK.  I probably need to figure out
> why the watchpoint triggered.
>
> Thanks,
>
> Arnold