On Wed, Jun 02, 2004 at 08:46:15AM -0400, [EMAIL PROTECTED] wrote: > > > On 2 Jun 2004 [EMAIL PROTECTED] wrote: > > > jorton 2004/06/02 01:27:43 > > > > Modified: test testsock.c > > Log: > > * test/testsock.c (setup_socket): Return NULL if bind fails, fixing > > test suite hang if port 8021 is in use; all callers updated. (come > > back longjmp, all is forgiven). > > No, you were right the first time with regard to longjmp. The trade-off > is that while using longjmp makes it easier to ignore the failure case, it > also means you _can't_ cleanup after yourself in the case of an error.
I was half joking: removing the longjmp was a good long term goal, but doing so *without* rewriting the test suite to take account of it just means the test suite is now totally unreliable, it's hanging/segfaulting in places where it wasn't before, which is quite frustrating. All the _NOT_IMPL calls need to checked to see whether they need to have a "return" too, for instance. joe
