Gawk master problems with -fsanitize=undefined

[Paul Eggert]

I tried building Gawk master on Fedora 31 x86-64 as follows:

./configure
make CFLAGS='-g3 -O2 -fsanitize=undefined'

I found two bugs; the first is trivial and has an obvious fix and the 
second I don't offhand know how to fix. There may be more bugs, but I 
stopped after the second bug.

A patch to the first bug is attached. The C Standard doesn't allow 
memcpy (..., NULL, 0), and -fsanitize=undefined reports this as an 
error. Although I don't know of platforms where such a memcpy fails, 
it's better to be safer, as in the attached patch.

The second bug has the following symptoms:

  $ ./gawk -f test/synerr3.awk
  gawk: test/synerr3.awk:1: for (i = ) in foo bar baz
  gawk: test/synerr3.awk:1: ^ syntax error
  gawk: test/synerr3.awk:1: for (i = ) in foo bar baz
  gawk: test/synerr3.awk:1:          ^ syntax error
  awkgram.y:1720:23: runtime error: member access within null pointer 
of type 'struct exp_instruction'

This appears to be related to the following code in awkgram.y:

                $4->opcode = Op_in_array;
                if ($2 == NULL) {       /* error */
                        errcount++;
                        $4->expr_count = 0;
                        $$ = list_merge($5, $4);

Here, $4's opcode is Op_in_array, and yet list_merge (when GAWKDEBUG is 
defined) asserts that $4's opcode must be Op_list. The list_merge sets 
$4->lasti to NULL, and when line 1720 of awkgram.y later dereferences 
lasti it gets into trouble.

0001-Fix-memcpy-issue-found-by-fsanitize-undefined.patch
Description: Text Data