Re: AIX build fails with missing symbol .krb5_c_make_random_key
Ben Huntsman <[email protected]>
| Newsgroups | gmane.comp.file-systems.openafs.devel |
|---|---|
| Message-ID | <MWHPR0701MB3674E3FB47B9D076AF876F86A7699@MWHPR0701MB3674.namprd07.prod.outlook.com> |
Sure, I agree for testing that's fine. But, there is evidence to suggest that this has compiled successfully on AIX 6.1 and 7.2 in the past, and this issue would have occurred there. Or are you suggesting that all successful compiles used Kerberos packages which provided only 32-bit libraries? Thanks. -Ben ________________________________ From: Benjamin Kaduk <[email protected]> Sent: Sunday, August 14, 2022 2:21 PM To: Ben Huntsman <[email protected]> Cc: Ken Hornstein <[email protected]>; [email protected] <[email protected]> Subject: Re: [OpenAFS-devel] AIX build fails with missing symbol .krb5_c_make_random_key On Sun, Aug 14, 2022 at 09:12:53PM +0000, Ben Huntsman wrote: > Hi there- > Yes, on AIX 6.1 we need to be building 64-bit, as there is no 32-bit on 6.1. The 32-bit kernel is available on AIX 5.3 and below. On XL C, the default is -q32. > > So, we should probably tweak something to always build as 64-bit on AIX 6.1... Alternatively, for userland stuff, the krb5-libs package provides both 32-bit and 64-bit libraries... It is interesting that krb5-config isn't doing the right thing for 32-bit... It's reasonable to just pass CFLAGS (and LDFLAGS?) to configure for diagnosis, but I think the long-term answer would be in the src/cf/osconf.m4 stanza I mentioned earlier. I don't see anything in your krb5-config that offers an option to provide 32- vs 64-bit output, and it wouldn't have any way to know intrinsically which bittedness we're asking it for. So maybe there are missing pieces in both what we ask it and what it can tell us... -Ben