site_archiver@lists.apple.com Delivered-To: darwin-dev@lists.apple.com On Aug 30, 2008, at 5:54 PM, Luke Daley wrote: On 31/08/2008, at 3:59 AM, Jordan K. Hubbard wrote: - Jordan _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.appl... And the grand prize goes to Jonas, for correctly intuiting that svn 1.5 is clearly built and running 64 bit, whereas the interpositioning library is probably not built with the same architectural support! I might further speculate that Luke is running a SnowLeopard seed since 64 bit execution is not the default in Leopard, and he would have to either force 64 bit execution or perhaps even have an svn 1.5 binary *without* the right 32 bit section which is now "falling back" to 64 bit, but that's pretty far-fetched. I'm going with SnowLeopard seed. :-) Either way, the solution is clear: Make sure that otool -f displays the same architecture sections (or use use file(1) - it's more human readable) for both the interposing dylib and the svn binary(s). This _is_ on Leopard. :) Yeah, sorry for the misinformation there - I was misremembering the Leopard defaults. So that would be the same? Here is how we build this thing… Well, clearly the 64 bit interpositioning case is the one that is biting you, and if you've already clarified that all the select() variants are STILL converging on your interpositioning routines even in the 64 bit ABI case, then I guess you need to look very carefully at your interpositioning code and see if you have any 32 vs 64 bit size issues anywhere, that's the only thing I can think of (getting a 64 bit value in and truncating it to 32 bits then passing it on to a syscall which expects 64 bits would do it). This email sent to site_archiver@lists.apple.com
participants (1)
-
Jordan K. Hubbard