Re: [bug-gawk] Increased RAM memory usage in gawk 5.0.1 compared to gawk 4.1.4

[Andrew J. Schorr]

On Wed, Aug 28, 2019 at 02:56:55PM -0400, Andrew J. Schorr wrote:
> I have to imagine that there's a reference counting issue with the
> $1 field arising from 'mom_version !~ $1'. I see this in the
> valgrind output when not setting fixram (note that purge_record never
> appears in the valgrind log in the fixram=1 case):

So I patched field.c like so:

diff --git a/field.c b/field.c
index feba8fd..94614a6 100644
--- a/field.c
+++ b/field.c
@@ -361,6 +361,7 @@ purge_record()
                        memcpy(r->stptr, save, r->stlen);
                        r->stptr[r->stlen] = '\0';
                        r->flags |= MALLOC;
+                       printf("debug: purge_record saving field %d with refcnt 
%ld\n", i, r->valref);
                }
                unref(r);
                getnode(n);

Then I ran:

bash-4.2$ ./gawk -v fixram=1 -f /extra_disk/tmp/gbug/simple_mem.awk 
/extra_disk/tmp/gbug/sqlRefMap.txt 
RSS per record = 0.13

All OK. But without fixram=1:

RSS per record = 0.13
bash-4.2$ ./gawk -f /tmp/simple_mem.awk /tmp/sqlRefMap.txt 
...
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
debug: purge_record saving field 1 with refcnt 2
RSS per record = 0.13

The test script is attached.

Regards,
Andy

simple_mem.awk
Description: Text document