lkml.org 
[lkml]   [2012]   [Mar]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] perf: Incorrect use of snprintf results in SEGV

* Peter Zijlstra <peterz@infradead.org> wrote:

> On Wed, 2012-03-07 at 21:37 +0100, Ingo Molnar wrote:
> > size_needed = snprintf_size(...);
>
> This would require 3 passes over the fmt+args, first to find
> the allocated size is insufficient, 2nd to compute the size,
> 3rd to fill buffer.

No. The 1% case would use this separate API with its quirky
return value. snprintf_size() works like today's snprintf, it's
just *named* clearly to signal that it returns something not
quite intuitive and results in bugs even in code that *tries* to
be aware of the corner cases.

> Whereas with the current "creative" API only 2 passes are
> needed.
>
> I can imagine that back in the day of small memory and small
> CPU this was deemed important enough.
>
> Anyway, its all moot, this API exists and has been out in the
> wild for several decades now, its not like we can actually
> change it :-)

Of course it is moot - I am not arguing for a change in the API.

But the self-justification, as outlined in the mail I replied
to, is brutally wrong, and nowhere in this discussion did I see
the important notion mentioned that the *common case matters* -
so maybe reading this will keep others from committing the same
mistake, with newly introduced APIs.

Thanks,

Ingo


\
 
 \ /
  Last update: 2012-03-08 08:37    [from the cache]
©2003-2011 Jasper Spaans