Re: [PATCH 3/4] tools/nolibc: move FD_* definitions to sys/select.h

[Willy Tarreau <w@1wt.eu>]

On Sun, Jun 29, 2025 at 06:27:27PM +0200, Thomas Weißschuh wrote:
> On 2025-06-29 18:19:08+0200, Willy Tarreau wrote:
> > On Sun, Jun 29, 2025 at 06:14:01PM +0200, Thomas Weißschuh wrote:
> > > On 2025-06-29 18:08:23+0200, Willy Tarreau wrote:
> > > > On Sun, Jun 29, 2025 at 05:54:02PM +0200, Thomas Weißschuh wrote:
> > > > > I saw the same issue, but only because of the changes to types.h.
> > > > > And these should not be necessary in the first place.
> > > > 
> > > > You mean fd_set definition ? It's solely a matter of includes ordering
> > > > in fact, since it depends on FD_SETSIZE.
> > > 
> > > No, I mean the '#include "sys/select.h"' in "types.h".
> > 
> > I had already dropped it as well.
> 
> Hm, then I'm not sure where it commes from.

I noticed that you included linux/unistd.h from sys/select.h, which I
didn't have, maybe it plays a role as well.

> > > That breaks the dependency order, as it pulls in all kinds of other
> > > stuff into the beginning of "types.h" which themselves depend on
> > > definitions of "types.h".
> > 
> > It was not just this. I'm pretty sure that what unbroke it for you is
> > keeping fd_set in types.h.
> 
> But I *did* move fd_set to sys/select.h.

Sorry then, I just didn't notice it while trying to comapre the
differences.

> > > > > The below works nicely for me:
> > > > 
> > > > OK. Do you prefer that we take that one or just a stub ? I'm personally
> > > > fine with both.
> > > 
> > > I prefer moving the code.
> > 
> > OK. Do you want me to merge your patch or will you take care of it ?
> 
> As you'll resubmit your series anyways, please pick up my proposal.

OK will do that. Thanks.
willy