487162 - ckfw/capi build failure on windows

Status: RESOLVED FIXED

(NSS :: Libraries, defect, P2)

Description

[Kai Engert (:KaiE:)]

When landing NSS 3.12.3 rc 0 we got a build failure when building Mozilla.

Reason is that directory ckfw/capi only gets built in the Mozilla client configuration.

As a quick solution I disabled building capi, r=nelson

We need to fix the compiler error in staticobj.c and re-enable the build.

Comment 1

[Kai Engert (:KaiE:)]

Build log: 
http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1239070965.1239072544.11625.gz

Build failure:
e:/builds/moz2_slave/mozilla-central-win32-unittest/build/security/nss/lib/ckfw/capi/staticobj.c(71) : error C4047: 'initializing' : 'CK_OBJECT_CLASS' differs in levels of indirection from 'void *'

Comment 2

[Kai Engert (:KaiE:)]

Can we change ckfw/Makefile to build capi by default when doing NSS builds?

Comment 3

[Nelson Bolyard (seldom reads bugmail)]

The last few lines of lib/ckfw/cpi/staticobj.c are

> ckcapiInternalObject nss_ckcapi_data[] = {
>  { ckcapiRaw, { 5, nss_ckcapi_types_1, nss_ckcapi_items_1} , {((void *)0)} },
> };
> 
> const PRUint32 nss_ckcapi_nObjects = 1;

The type ckcapiInternalObject contains a union.  The code above is trying 
to initialize one of the members of that union with a static initializer.  
That's a no-no.

Comment 4

[Nelson Bolyard (seldom reads bugmail)]

IMO, it doesn't belong in NSS.  
It uses NSS.  It uses lib/ckfw.  It's not part of either one.
It doesn't belong under lib/ckfw any more than programs that use 
libc belong in the libc source tree.

Comment 5

[Nelson Bolyard (seldom reads bugmail)]

lib/ckfw/Makefile contains these lines:

> ifdef MOZILLA_CLIENT
> NSS_BUILD_CAPI = 1
> endif

But NONE of the NSS team members builds routinely with MOZILLA_CLIENT defined.
So, when Mozilla takes a snapshot of NSS, they find themselves building things
that the NSS does not build and has not build for months, if ever.

I believe Kai removed those lines from Mozilla's snapshot of the Makefile.
(That was my suggestion, and I have r+ to that patch, sight unseen. :)
I'd say we should do that in the CVS tree, too.

Comment 6

[Nelson Bolyard (seldom reads bugmail)]

The code was always wrong, I believe.  NSS just catches it now with its
more strict error checking during compilation.

I believe the only portable solutions to this are:

a) stop using a static initializer, or 
b) change this object (structure) type to remove all unions.

I'll let Bob carry it from here.

Comment 7

[Nelson Bolyard (seldom reads bugmail)]

Attachment: patch, silence MSVC

I was mistaken about the cause of the error.  It wasn't the initialization 
of the union.  It was the attempt to initialize a member of type 
CK_OBJECT_CLASS with the value {NULL}.  This patch fixes it, and makes 
MSVC happy.  I think think it's an error to use a static initializer for 
a union. But the immediate issue is to pacify MSVC.

Comment 8

[Kai Engert (:KaiE:)]

(In reply to comment #5)
> lib/ckfw/Makefile contains these lines:
> 
> > ifdef MOZILLA_CLIENT
> > NSS_BUILD_CAPI = 1
> > endif
> 
> ...
> 
> I believe Kai removed those lines from Mozilla's snapshot of the Makefile.

yes

Comment 9

[Kai Engert (:KaiE:)]

http://hg.mozilla.org/mozilla-central/rev/b5f6a4b00c7f

Comment 10

[[e-mail]]

(In reply to comment #7)
>I think think it's an error to use a static initializer for a union.
My hazy memory suggests that you initialise it as if it was its first member.

Comment 11

[Nelson Bolyard (seldom reads bugmail)]

Checking in staticobj.c; new revision: 1.3; previous revision: 1.2