Re: emacs' turn: remove useless if-before-free tests

[Jim Meyering]

Jim Meyering <address@hidden> wrote:
> Jim Meyering <address@hidden> wrote:
>> Emanuele Giaquinta <address@hidden> wrote:
>>> On Mon, Jun 02, 2008 at 08:14:18AM +0200, Jim Meyering wrote:
>>>
>>>> Stefan Monnier <address@hidden> wrote:
>>>> > Sounds like a good cleanup.
>>>> > Feel free to install it unless there's a strong objection.
>>>>
>>>> Thanks.  I've committed those three change sets.
>>>
>>> This patch introduced a problem on os x, free on os x is redirected to
>>> unexmacosx.c:unexec_free, which does not support a NULL argument in an
>>> undumped emacs. The attached patch changes the problematic free call to
>>> xfree, as done a few lines below for another pointer.
>>>
>>> Emanuele
>>>
>>> diff --git a/src/lread.c b/src/lread.c
>>> index e5e77bc..3e0bd1f 100644
>>> --- a/src/lread.c
>>> +++ b/src/lread.c
>>> @@ -1269,7 +1269,7 @@ Return t if the file exists and loads successfully.  
>>> */)
>>>
>>>    UNGCPRO;
>>>
>>> -  free (saved_doc_string);
>>> +  xfree (saved_doc_string);
>>>    saved_doc_string = 0;
>>>    saved_doc_string_size = 0;
>>
>> Thanks for catching and fixing that.
>> If using xfree (with its MALLOC_BLOCK_INPUT guard) is important there,
>> as it seems to be, then your patch also fixes a platform-independent
>> race condition bug.
>>
>> The following change is probably a good idea, too.
>> It makes unexec_free (NULL) a no-op, just like free (NULL) is:
>>
>> 2008-06-08  Jim Meyering  <address@hidden>
>>
>>      make unexec_free handle NULL the same way free does
>>      * unexmacosx.c (unexec_free): Ignore a NULL argument.
>
> If no one objects soon, I'll commit both our changes.

Done.