$ python2.5 Python 2.5.2 (r252:60911, Jun 16 2008, 15:20:47) [GCC 3.3.5 (propolice)] on openbsd4 Type "help", "copyright", "credits" or "license" for more information. >>> def complete(text, state): ... print 'complete %r %d' % (text, state) ... if text == 'i' and state == 0: ... return 'import' ... else: ... return None ... >>> import readline; readline.parse_and_bind("tab: complete") >>> readline.set_completer(complete) >>> i # is press the tab key complete 'i' 0 complete 'i' 1 Segmentation fault (core dumped) The problem is that Python is using a function present in libreadline but not declared in readline/readline.h: completion_matches. Instead of using rl_completion_matches. Therefor the return type of completion_matches was an int which is 32bits on amd64 but the function was supposed to returns a pointer which is 64 bits. So when the pointer had a "high" value, it was truncated. The problem is fixed by adding libcurses to AC_CHECK_LIB when checking for functions in libreadline since libreadline depends on curses. This makes Python use the correct functions declared on readline.h with a correct return type. Patch is attached. (Others versions of Python should also be affected)
Looks like this patch should also be applied to python 2.6 & 3. I did not tested but the patch is trivial enough to be applied without too much fear of breaking something ;)
committed to trunk (2.6) in r66179. This should be back ported to release25-maint and automagically merged into py3k. can someone with OpenBSD confirm that this has indeed fixed the problem? if so i'll do the 25 backport and mark it as closed instead of pending.
I just compiled the latest version of trunk. The problem seems to be fixed. And according to config.log & nm, readline.so is linked with the right function.
