Re: gnustep-base code freeze

[David Chisnall]

On 23 Mar 2011, at 19:34, Fred Kiefer wrote:

> I just made that small change I suggested on the list. Now the test code 
> passes all the tests for me. Of course this might be different if more 
> formatter tests were added.
> 
>   5544 Passed tests
>     11 Dashed hopes
>      8 Skipped sets

On FreeBSD/x86, I get:

base/NSString/test02.m:
Failed test:     test02.m:277 ... foo->bar relative symlink not expanded by 
stringByResolvingSymlinksInPath

5689 Passed tests
   6 Dashed hopes
   1 Skipped set
   1 Failed test

This failure is a bit strange.  Looking at the test, I can't understand why it 
works anywhere:

It creates a directory /tmp/GNUstepSecure001/bar.  It creates a symbolic link, 
from /tmp/GNUstepSecure001/foo to bar.  It then tries to resolve @foo" as a 
symbolic link - only foo in the current working directory isn't a symbolic link 
at all.

On FreeBSD, this is using the realpath() call, which also expands the path to a 
full absolute path.  From the OS X docs, this sounds like it's the correct 
thing, but if the test is passing on other platforms then it might not be 
(since it expects @"bar" to be returned).

On Linux/x86-64, I get:

./NSConnection/connection.m:
Failed file:     connection.m aborted without running all tests!
   5659 Passed tests
      6 Dashed hopes
      2 Skipped sets
      1 Failed file

That NSConnection test seems to be failing pretty consistently on that machine 
(12 core AMD thing), but from what Richard said it sounds like it's safe to 
ignore.

The stuff that relies on gnutls is skipped on both platforms.  Eventually I 
might get around to implementing this with OpenSSL, but the NSStream API is so 
horrible I probably won't, since I can't imagine ever wanting to actually use 
it.  The GSXML tests are skipped on Linux, because I don't have root on that 
machine and got bored with compiling dependencies from source and installing 
them in ~.

Both are with latest trunk GNUstep and clang, using the non-fragile ABI.  All 
of my clang bug fixes have made it into the 2.9 release branch.  I can probably 
sneak small fixes in if anyone finds any last-minute things, but since all of 
the tests that are remotely compiler / runtime related are passing on both 
Linux and FreeBSD, 32 and 64-bit, I'm fairly happy.  I should be able to add 
ARM Cortex A8 to the list of architectures I test soon...

David


-- Sent from my Apple II