Re: [PATCH] Fix memory leak in printf_positional

[Paul Pluzhnikov <ppluzhnikov at google dot com>]

On Mon, Aug 31, 2015 at 3:41 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> On Sat, 29 Aug 2015, Paul Pluzhnikov wrote:
>
>> It doesn't take very long: all I needed is a printf invocation with >=
>> 65536 / 3 / sizeof(void*) arguments.
>>
>> Writing such invocation by hand is of course toublesome, plumbing
>> Makefile to generate it for me, and figuring out why it doesn't work
>> is what takes time :-(
>
> Normally such invocations would be generated with macros, e.g.
>
> #define A a, a, a, a, a, a, a, a, a, a
> #define B A, A, A, A, A, A, A, A, A, A
>
> etc., unless the sort of expansion you require is unsuited to that for
> some reason.

Well, to show the leak requires positional format, like this:

  "%1$s %2$s ..."

However, I was able to use

  "%1$s %s %s ..."

if I disable -Wformat for the test.


>> In addition, there is a GCC regression: compiling a printf call with
>> 2800 arguments takes 4.8.4-2ubuntu1~14.04 0.06s without optimization,
>> 0.86s with -O2. Same numbers for current GCC trunk (@r227321): 0.06s
>> and  4m46s. This is on a very recent and fast PC. I expect there could
>> be PCs in current use where the time will be 3x longer.
>
> To me this suggests building the test with -O0

That doesn't work: glibc requires to be built with optimization :-)

But #pragma GCC optimize does work.

The test case revealed an additional leak, so at least that effort was
not in vain.

Combined patch attached. Tested on Linux/x86_64.

Thanks,

2015-09-01  Paul Eggert  <eggert@cs.ucla.edu>
   Paul Pluzhnikov  <ppluzhnikov@google.com>

        [BZ #18872]
        * stdio-common/Makefile (tst-printf-bz18872): New test.
        (tst-printf-bz18872-mem.out): Likewise.
        * stdio-common/tst-printf-bz18872.c: New test.
        * stdio-common/vfprintf.c: Fix memory leaks.



-- 
Paul Pluzhnikov

Attachment: bz18872.20150901.txt
Description: Text document