Report from Tavis Ormandy, Google Security Team:

An audit of ncompress version 4.2.4 uncovered a serious security flaw, this loop
in decompress() (~1749, compress42.c) performs no bounds checking, allowing a
specially crafted datastream to underflow a .bss buffer with attacker controlled
data. Some research reveals that the lzw decompressors from gzip and openbsd
(both derived from the same public domain implementation) have already corrected
this flaw, however ncompress shipped by (at least) gentoo, debian, fedora and
suse seem to still be vulnerable.

              while ((cmp_code_int)code >= (cmp_code_int)256)
              { /* Generate output characters in reverse order */
                  *--stackp = tab_suffixof(code);
                  code = tab_prefixof(code);
              }

In my test environment I've been able to successfully overwirte .got and .dtors
with controlled data. The most simple testcase would be:

$ perl -e 'print "\x1f\x9d\x90","\x01"x"2048"' | compress -d

My suggested fix would be adding  `&& stackp >= htabof(0)` to the loop condition.