Moriyoshi Koizumi <mozo@xxxxxxx> writes:
> diff --git a/http.c b/http.c
> index ee58799..41e8e8c 100644
> --- a/http.c
> +++ b/http.c
> @@ -13,18 +13,24 @@ static CURL *curl_default;
> char curl_errorstr[CURL_ERROR_SIZE];
>
> static int curl_ssl_verify = -1;
> -static const char *ssl_cert = NULL;
> +static const char *ssl_cert;
> #if LIBCURL_VERSION_NUM >= 0x070902
> -static const char *ssl_key = NULL;
> +static const char *ssl_key;
> #endif
> #if LIBCURL_VERSION_NUM >= 0x070908
> -static const char *ssl_capath = NULL;
> +static const char *ssl_capath;
> #endif
> -static const char *ssl_cainfo = NULL;
> +static const char *ssl_cainfo;
> static long curl_low_speed_limit = -1;
> static long curl_low_speed_time = -1;
> static int curl_ftp_no_epsv = 0;
> -static const char *curl_http_proxy = NULL;
> +static const char *curl_http_proxy;
Applying style fixes to the existing code is very much appreciated, *but*
please make such a clean-up patch a separate one. A two-patch series
whose [1/2] is such a pure clean-up without any feature change, with [2/2]
that adds code to the cleaned-up state would be much less distracting for
people who nitpick your changes.
> @@ -153,11 +159,69 @@ static int http_options(const char *var, const char *value, void *cb)
> return git_config_string(&curl_http_proxy, var, value);
> return 0;
> }
> +#if LIBCURL_VERSION_NUM >= 0x070a06
> + if (!strcmp("http.auth", var)) {
> + if (curl_http_auth == NULL)
> + return git_config_string(&curl_http_auth, var, value);
> + return 0;
> + }
> +#endif
> +#if LIBCURL_VERSION_NUM >= 0x070a07
> + if (!strcmp("http.proxy_auth", var)) {
> + if (curl_http_proxy_auth == NULL)
> + return git_config_string(&curl_http_proxy_auth, var, value);
> + return 0;
> + }
> +#endif
If you follow config.c::git_config() you will notice that we read from
/etc/gitconfig, $HOME/.gitconfig and then finally $GIT_DIR/config. By
implementing "if we already have read curl_http_auth already, we will
ignore the later setting" like above code does, you break the general
expectation that system-wide defaults is overridable by $HOME/.gitconfig
and that is in turn overridable by per-repository $GIT_DIR/config.
The preferred order would be:
- Use the value obtained from command line parameters, if any;
- Otherwise, if an environment variable is there, use it;
- Otherwise, the value obtained from git_config(), with "later one wins"
rule.
I think you are not adding any command line option, so favoring
environment and then using configuration is fine, but the configuration
parser must follow the usual "later one wins" rule to avoid dissapointing
the users.
> +#if LIBCURL_VERSION_NUM >= 0x070a06
> +static long get_curl_auth_bitmask(const char* auth_method)
In git codebase, asterisk that means "a pointer" sticks to the variable
name not to type name; "const char *auth_method" (I see this file is
already infested with such style violations, but if you are doing a
separate clean-up patch it would be appreciated to clean them up).
> +{
> + char buf[4096];
Do you need that much space?
> + const unsigned char *p = (const unsigned char *)auth_method;
> + long mask = CURLAUTH_NONE;
> +
> + strlcpy(buf, auth_method, sizeof(buf));
A tab is 8-char wide.
What happens when auth_method is longer than 4kB?
> +
> + for (;;) {
> + char *q = buf;
> + while (*p && isspace(*p))
> + ++p;
If there is no particular reason to choose one over the other, please use
postincrement, p++, as other existing parts of the codebase.
I'll try to demonstrate what (I think) this patch should look like as a
pair of follow-up messages to this one, but I am not sure about my rewrite
of get_curl_auth_bitmask(). Please consider it as an easter-egg bughunt
;-)
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
