Hi Willy Tarreau,
Willy Tarreau wrote on 21.07.2017:
> Hi David,
> On Fri, Jul 21, 2017 at 07:36:37AM +0100, David CARLIER wrote:
>> Hi all,
>>
>> hopefully it will be revealed useful, in this patch, we re trying to use
>> the MALLOC/FREE macros since pcre* libraries are capable of overriding
>> internally memory management.
> Given that the MALLOC and FREE macros are only used to allocate/release
> pools, and that malloc/calloc/free/strdup/realloc() are used everywhere
> else, don't you think that you should simply use the regular ones ? In
> fact the initial purpose of the MALLOC macro was to be able to redefine
> the allocator but in the end there's so much code using malloc()
> (including in libraries we have no control over) that it only became a
> way to occasionally debug the pool allocators. I think we could even get
> rid of these macros by now.
>> I had a little related question, would it be better to provide function
>> pointers of this form struct my_allocator { void *(*alloc)(void *, size_t),
>> void (*free)(void *, void *)} .... } to allow using hierarical based
>> allocators like talloc for example ... just thinking aloud not sure it s
>> relevant.
> I have no opinion here, I really don't know.
I thought something like
`my_pcrealloc` use `create_pool` inside and `my_pcrefree` use
`pool_destroy2` but one of the main question is how does the pcre2
handle the custom memory struct we have in the pools.
http://git.haproxy.org/?p=haproxy.git;a=blob;f=src/memory.c;hb=HEAD#l40
http://git.haproxy.org/?p=haproxy.git;a=blob;f=src/memory.c;hb=HEAD#l198
> However I'm seeing something :
>> +#if defined(USE_PCRE) || defined(USE_PCRE_JIT)
>> +static inline void *
>> +my_pcrealloc(size_t size)
>> +{
>> + return MALLOC(size);
>> +}
>> +
>> +static inline void
>> +my_pcrefree(void *ptr)
>> +{
>> + FREE(ptr);
>> +}
>> +#elif defined(USE_PCRE2) || defined(USE_PCRE2_JIT)
>> +static inline void *
>> +my_pcrealloc(PCRE2_SIZE size, void *a)
>> +{
>> + (void)a;
>> + return MALLOC(size);
>> +}
>> +
>> +static inline void
>> +my_pcrefree(void *ptr, void *a)
>> +{
>> + (void)a;
>> + FREE(ptr);
>> +}
>> +#endif
>> +
> (...)
>> #ifdef USE_PCRE
>> + pcre_malloc = my_pcrealloc;
>> + pcre_free = my_pcrefree;
> These two functions are never used as-is and only as pointers, so the
> "inline" type modifier is useless at best and partially wrong (in that
> it is confusing), eventhough this is harmless. But I think you should
> remove it.
> Cheers,
> Willy
--
Best Regards
Aleks
