Re: [PATCH] git-compat-util: Avoid strcasecmp() being inlined

From: Junio C Hamano <gitster@pobox.com>
Date: Sep 12, 2013, 15:37 UTC
Message-ID: <xmqq61u6qcez.fsf@gitster.dls.corp.google.com>
In-Reply-To: <20130912101419.GY2582@serenity.lan>

John Keeping <john@keeping.me.uk> writes:

Show 15 quoted lines

> On Thu, Sep 12, 2013 at 11:36:56AM +0200, Sebastian Schuberth wrote:
>> > Just wondering if that is the root of the problem, or if maybe there is
>> > something else subtle going on. Also, does __CRT_INLINE just turn into
>> > "inline", or is there perhaps some other pre-processor magic going on?
>> 
>> This is the function definition from string.h after preprocessing:
>> 
>> extern __inline__ int __attribute__((__cdecl__)) __attribute__ ((__nothrow__))
>> strncasecmp (const char * __sz1, const char * __sz2, size_t __sizeMaxCompare)
>>   {return _strnicmp (__sz1, __sz2, __sizeMaxCompare);}
>
> I wonder if GCC has changed it's behaviour to more closely match C99.
> Clang as a compatibility article about this sort of issue:
>
>     http://clang.llvm.org/compatibility.html#inline

Interesting.  The ways the page suggests as fixes are

 - change it to a "statis inline";
 - remove "inline" from the definition;
 - provide an external (non-inline) def somewhere else;
 - compile with gnu899 dialect.

But the first two are non-starter, and the third one to force everybody to define an equivalent implementation is nonsense, for a definition in the standard header file.

I agree with an earlier conclusion that defining our own wrapper (with an explanation why such a redundant wrapper exists) is the best course of action at this point, until the system header is fixed.

 mailmap.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/mailmap.c b/mailmap.c
index a7969c4..d36d424 100644
--- a/mailmap.c
+++ b/mailmap.c
@@ -52,6 +52,19 @@ static void free_mailmap_entry(void *p, const char *s)
 	string_list_clear_func(&me->namemap, free_mailmap_info);
 }
 
+/*
+ * On some systems, string.h has _only_ inline definition of strcasecmp
+ * without supplying a non-inline implementation anywhere, which is, eh,
+ * "unusual"; we cannot take an address of such a function to store it in
+ * namemap.cmp.  This is here as a workaround---do not assign strcasecmp
+ * directly to namemap.cmp until we know no systems that matter have such
+ * an "unusual" string.h.
+ */
+static int namemap_cmp(const char *a, const char *b)
+{
+	return strcasecmp(a, b);
+}
+
 static void add_mapping(struct string_list *map,
 			char *new_name, char *new_email,
 			char *old_name, char *old_email)
@@ -75,7 +88,7 @@ static void add_mapping(struct string_list *map,
 		item = string_list_insert_at_index(map, index, old_email);
 		me = xcalloc(1, sizeof(struct mailmap_entry));
 		me->namemap.strdup_strings = 1;
-		me->namemap.cmp = strcasecmp;
+		me->namemap.cmp = namemap_cmp;
 		item->util = me;
 	}
 
@@ -237,7 +250,7 @@ int read_mailmap(struct string_list *map, char **repo_abbrev)
 	int err = 0;
 
 	map->strdup_strings = 1;
-	map->cmp = strcasecmp;
+	map->cmp = namemap_cmp;
 
 	if (!git_mailmap_blob && is_bare_repository())
 		git_mailmap_blob = "HEAD:.mailmap";