Re: [Nano-devel] interrupting an open() does work with gcc when flags are zeroed

[Benno Schulenberg]

Op 23-05-19 om 14:26 schreef Brand Huntsman:
> On Thu, 23 May 2019 11:28:04 +0200 Benno Schulenberg <address@hidden>
> wrote:
>> With attached demo patch, ^C will interrupt the open() call for me, *when* I
>> use clang or tcc as compiler.  But with gcc... nothing happens.  :|
> 
> newaction.sa_handler = noop; newaction.sa_flags = 0;
> 
> The patch segfaults without the sa_flags line, because it contains garbage. ^C
> now works fine in gcc 8.3 when properly set to zero.

Yeah, I started suspecting an uninitialized value when moving the
declaration of oldaction and newaction caused different behavior
with tcc.

> It might be a good idea to zero the flags for other code that uses
> sigaction().

Will do.  (I thought that the initial 'sigaction(SIGINT, NULL, &newaction)'
that is used in other places would initialize the whole struct, but this
appears not to be the case.)

Benno

signature.asc
Description: OpenPGP digital signature