Re: [PATCH v2 next 03/11] tools/nolibc/printf: Add buffering to vfprintf() callback.

[Thomas Weißschuh]

On 2026-02-17 11:51:43+0000, David Laight wrote:
> On Mon, 16 Feb 2026 20:07:35 +0100
> Thomas Weißschuh <linux@xxxxxxxxxxxxxx> wrote:
> 
> > On 2026-02-07 23:36:51+0000, David Laight wrote:
> > > On Sat, 7 Feb 2026 20:29:34 +0100
> > > Willy Tarreau <w@xxxxxx> wrote:
> > >   
> > > > On Fri, Feb 06, 2026 at 07:11:13PM +0000, david.laight.linux@xxxxxxxxx wrote:  
> > > > > From: David Laight <david.laight.linux@xxxxxxxxx>
> > > > > 
> > > > > Add per-call buffering to the vprintf() callback.
> > > > > While this adds some extra code it will speed things up and
> > > > > makes a massive difference to anyone looking at strace output.    
> > > > 
> > > > This patch alone adds more than 200 extra bytes to the smallest binary
> > > > for something that was never expressed as a need by users:
> > > > 
> > > >  $ size hello-patch*
> > > >    text    data     bss     dec     hex filename
> > > >    1859      48      24    1931     78b hello-patch1
> > > >    2071      48      24    2143     85f hello-patch2
> > > > 
> > > > I doubt it would make sense to have a build option to choose this.
> > > > Or alternately one could decide do disable it when __OPTIMIZE_SIZE__
> > > > is defined. I just tried quickly and it does the job:  
> > > 
> > > That probably makes sense.
> > > For anything non-trivial the extra size is noise.
> > > Actually is would be easier to read with a single #if covering the
> > > whole lot.  
> > 
> > While I like the slimmer strace output, the added complexity, including
> > a different implementation for __OPTIMIZE_SIZE__ is a bit scary.
> > Could we leave this out for now and work on it on its own?
> 
> I might move it later in the patch series.

Which advantage would such a move have?
Please leave it out for now.

> I plan to add a #define that can select some of the extra features.
> Basically making bits from 'flags' makes the compiler optimise the
> associated code away - so 0 => no options and -1 => all options.
> But you also set '-' (with the quotes) to get just left alignment
> or '-', '.' to get left alignment and zero pad.

While I like the idea, it seems impossible to test properly.
This will also require a whole other discussion. Let's also skip this for now.

> I can use a character (that misses a flag -eg 'b') to enable buffering.
> 
> I thought about being able to disable features, but the requirement
> is only really to make small programs very small (but they may want
> one extra bit), the 400(ish) bytes that everything adds isn't that
> significant for anything non-trivial.

Agreed.


Thomas