Re: [RFC 01/47] eal: add BSD version of queue.h

[Bruce Richardson <bruce.richardson@intel.com>]

On Tue, Aug 19, 2025 at 09:03:41AM -0700, Stephen Hemminger wrote:
> On Tue, 19 Aug 2025 16:40:28 +0100
> Bruce Richardson <bruce.richardson@intel.com> wrote:
> 
> > On Tue, Aug 19, 2025 at 05:29:28PM +0200, Thomas Monjalon wrote:
> > > 19/08/2025 17:24, Stephen Hemminger:  
> > > > On Tue, 19 Aug 2025 10:37:27 +0200
> > > > Thomas Monjalon <thomas@monjalon.net> wrote:
> > > >   
> > > > > 19/08/2025 01:27, Stephen Hemminger:  
> > > > > > +++ b/lib/eal/include/bsd_queue.h
> > > > > > @@ -0,0 +1,1075 @@
> > > > > > +/*-
> > > > > > + * SPDX-License-Identifier: BSD-3-Clause
> > > > > > + *
> > > > > > + * Copyright (c) 1991, 1993
> > > > > > + *     The Regents of the University of California.  All rights reserved.
> > > > > > + *
> > > > > > + * This is a copy of sys/queue.h which is used to overcome
> > > > > > + * missing parts in the glibc version (and Windows).
> > > > > > + */
> > > > > > +
> > > > > > +#ifndef _SYS_QUEUE_H_
> > > > > > +#define _SYS_QUEUE_H_    
> > > > > 
> > > > > Shouldn't we rename this include flag? with RTE_ prefix?  
> > > > 
> > > > 
> > > > No, by keeping the same prefix if the user includes both sys/queue.h and bsd_queue.h
> > > > there will be no confilcts.  
> > > 
> > > You mean by checking _SYS_QUEUE_H_ which is the same as sys/queue.h?
> > > So if I include sys/queue.h first, I will have a limited version.
> > > It means we should not include it before any DPDK include.
> > > It looks like a limitation for DPDK users.
> > >   
> > I think we should define two defines for our header, and warn if
> > sys/queue.h is included first.
> > 
> > #ifndef _RTE_BSD_QUEUE_H_
> > #ifdef _SYS_QUEUE_H_
> > #warning ...
> > #endif
> > 
> > #define _RTE_BSD_QUEUE_H_
> > #define _SYS_QUEUE_H_
> > 
> > ...
> > 
> > #endif
> > 
> > This way, sys/queue.h contents will be skipped after our header, but if
> > it's included first, the user gets a warning about incomplete
> > implementation.
> > 
> > /Bruce
> 
> There is a similar issue with kernel headers already. And it lead
> to lots if #ifdef __UAPI_DEF_XXX which I want to avoid.
> 
> Rather just do wholesale replacement and if application includes
> the other header, it will work since the BSD stuff is only used
> as add on.
> 
That level of ifdef checks what my suggestions above is trying to avoid.
Just by having our own header define, as well as defining the standing
sys/queue.h one, we can avoid having queue.h included after us, and also
warn the user if they have already included the incomplete header. [Though
for correctness, I suppose the first endif above should be an #else, so
avoid dup definitions if that occurs]

/Bruce