Re: [PATCH 5/3] revert most of the http_options() change
- From
Kyle J. McKay <mackyle@gmail.com>
- Date
- Jul 30, 2013, 19:41 UTC
- Message-ID
- <3C93D9D6-8FC3-4D94-AE6E-37150202314A@gmail.com>
- In-Reply-To
- <CC3A0EB0-79FE-40C2-B1AC-E3B9683A3ED6@gmail.com>
On Jul 30, 2013, at 12:14, Kyle J. McKay wrote:
Show 70 quoted lines
> On Jul 29, 2013, at 19:13, Junio C Hamano wrote:
>
>> With the previous preparation step, the earlier 1bb00006 (config:
>> add support for http.<url>.* settings, 2013-07-21) that introduced
>> many repeated changes:
>>
>> -if (!strcmp("http.key", var)) {
>> +if (!strcmp("key", key)) {
>> + if (match_is_shorter(..., OPT_KEY_NAME))
>> + return 0;
>> ... original processing for KEY_NAME ...
>> }
>>
>> can be reverted.
>>
>> This allows us to also get rid of the "repeating yourself to the
>> death" enum http_option_type, and new code (like db/http-savecookies
>> patch) that wants to add a new configuration to the http.* subsystem
>> does not have to conflict with http.<urlpattern>.variable series at
>> all.
>>
>> This is more-or-less how I want the endgame to look like. Not even
>> compile tested, but it is getting late so I'll show it to the list
>> in case people may want to play with it and polish it while I am
>> away for the night ;-).
>>
>> If people can agree that this is going in the right direction,
>> perhaps we should rebuild Kyle's series without detouring of adding
>> and then yanking what 1bb00006 (config: add support for http.<url>.*
>> settings, 2013-07-21) did in the next cycle.
>>
>> http.c | 181 ++++++++
>> +------------------------------------------
>> test-url-normalize.c | 9 ++-
>> 2 files changed, 39 insertions(+), 151 deletions(-)
>>
>> diff --git a/http.c b/http.c
>> index a91a00b..c7f513b 100644
>> --- a/http.c
>> +++ b/http.c
> [...]
>> @@ -469,15 +342,23 @@ void http_init(struct remote *remote, const
>> char *url, int proactive_auth)
>> {
>> char *low_speed_limit;
>> char *low_speed_time;
>> - struct url_info info;
>> + char *normalized_url;
>> + struct urlmatch_config config = { STRING_LIST_INIT_DUP };
>> +
>> + config.fn = http_options;
>> + config.cascade_fn = git_default_config;
>> + config.item_alloc = NULL;
>> + config.item_clear = NULL;
>> + config.cb = NULL;
>
> Missing this:
> + config.section = "http";
>
>
> [...]
>> diff --git a/test-url-normalize.c b/test-url-normalize.c
>> index 0b809eb..fab94e5 100644
>> --- a/test-url-normalize.c
>> +++ b/test-url-normalize.c
>> @@ -15,8 +15,15 @@ static int run_http_options(const char *file,
>> {
>> struct strbuf opt_lc;
>> size_t i, len;
>> + struct urlmatch_config config = { STRING_LIST_INIT_DUP };Also needs this:
+ memcpy(&config.url, info, sizeof(struct url_info));
Show 19 quoted lines
>> - if (git_config_with_options(http_options, (void *)info, file, 0)) >> + config.fn = http_options; >> + config.cascade_fn = git_default_config; >> + config.item_alloc = NULL; >> + config.item_clear = NULL; >> + config.cb = NULL; > > > Missing this: > + config.section = "http"; > > Without these it segfaults running the tests. > > However, now the tests that actually check the results using config > files are failing: > >> # failed 1 among 11 test(s) > > I'm looking at it now to see if I can fix the problem.
And now all the t5200-url-normalize tests pass again.
FYI, I couldn't get the patches to apply against next or pu without some minor tweaks that were just conflict resolutions having to do with git_config_with_options changing its signature.