http: use xmalloc with cURL

Commit Message

Carlo Marcelo Arenas Belón Aug. 15, 2019, 3:35 p.m. UTC

f0ed8226c9 (Add custom memory allocator to MinGW and MacOS builds, 2009-05-31)
never told cURL about it.

Correct that by using the cURL initializer available since version 7.12 to
point to xmalloc and friends for consistency which then will pass the
allocation requests along when USE_NED_ALLOCATOR=YesPlease is used (most
likely in Windows)

Signed-off-by: Carlo Marcelo Arenas Belón <[e-mail]>
---
This doesn't conflict with anything and was originally based on maint (so it
applies cleanly also to master and next), but is now rebased on top of
jk/drop-release-pack-memory so the final product wouldn't have any chance to
introduce problems (thanks Peff)

it has been built and tested in Windows through Azure Pipelines (thanks Dscho)
and shouldn't introduce any build problems even with ancient versions or cURL
(thanks Daniel) and while not strictly needed is a nice thing to have for
consistency (thanks Junio)

 http.h | 6 ++++++
 1 file changed, 6 insertions(+)


base-commit: 9827d4c185e4da728f51cd77c54a38c9de62495f

Comments

Junio C Hamano Aug. 15, 2019, 5:11 p.m. UTC | #1

Carlo Marcelo Arenas Belón  <[e-mail]> writes:

> f0ed8226c9 (Add custom memory allocator to MinGW and MacOS builds, 2009-05-31)
> never told cURL about it.
>
> Correct that by using the cURL initializer available since version 7.12 to
> point to xmalloc and friends for consistency which then will pass the
> allocation requests along when USE_NED_ALLOCATOR=YesPlease is used (most
> likely in Windows)
>
> Signed-off-by: Carlo Marcelo Arenas Belón <[e-mail]>
> ---
> This doesn't conflict with anything and was originally based on maint (so it
> applies cleanly also to master and next), but is now rebased on top of
> jk/drop-release-pack-memory so the final product wouldn't have any chance to
> introduce problems (thanks Peff)
>
> it has been built and tested in Windows through Azure Pipelines (thanks Dscho)
> and shouldn't introduce any build problems even with ancient versions or cURL
> (thanks Daniel) and while not strictly needed is a nice thing to have for
> consistency (thanks Junio)

Thanks, all.  I do not think I did anything that deserves a
"thanks", but anyway.
>  http.h | 6 ++++++
>  1 file changed, 6 insertions(+)
>
> diff --git a/http.h b/http.h
> index b429f1cf04..e5f075dcbf 100644
> --- a/http.h
> +++ b/http.h
> @@ -22,9 +22,15 @@
>  #define DEFAULT_MAX_REQUESTS 5
>  #endif
>  
> +#if LIBCURL_VERSION_NUM >= 0x070c00
> +#define curl_global_init(a) curl_global_init_mem(a, xmalloc, free, \
> +						xrealloc, xstrdup, xcalloc)
> +#endif
> +
>  #if LIBCURL_VERSION_NUM < 0x070704
>  #define curl_global_cleanup() do { /* nothing */ } while (0)
>  #endif
> +
>  #if LIBCURL_VERSION_NUM < 0x070800
>  #define curl_global_init(a) do { /* nothing */ } while (0)
>  #endif

The earlier version had definitions pertaining to curl_global_init()
at a single place, but the above version allows arbitrary amount of
unrelated stuff to be placed in between two definitions in future
updates.  Is that a good idea?  I sort of expected "we always call
global_init(), but if curl is old enough to lack it, define it as
empty, and if curl is new enough to have curl_global_init_mem(),
then call it instead" #if/#else/#if/#endif/#endif cascade.

>
> base-commit: 9827d4c185e4da728f51cd77c54a38c9de62495f

Patch

diff --git a/http.h b/http.h
index b429f1cf04..e5f075dcbf 100644
--- a/http.h
+++ b/http.h
@@ -22,9 +22,15 @@ 
 #define DEFAULT_MAX_REQUESTS 5
 #endif
 
+#if LIBCURL_VERSION_NUM >= 0x070c00
+#define curl_global_init(a) curl_global_init_mem(a, xmalloc, free, \
+						xrealloc, xstrdup, xcalloc)
+#endif
+
 #if LIBCURL_VERSION_NUM < 0x070704
 #define curl_global_cleanup() do { /* nothing */ } while (0)
 #endif
+
 #if LIBCURL_VERSION_NUM < 0x070800
 #define curl_global_init(a) do { /* nothing */ } while (0)
 #endif