threads / discuss / 6127

Re: http git and curl 7.16.0

Subject: Re: http git and curl 7.16.0

## tl;dr

5 messages between Dec 27, 2006 and Jan 1, 2007.

replies: 4people: 2as markdown or json

Horst H. von Brand· Dec 27, 2006, 14:57 UTC · lore
Sven Verdoolaege <skimo@kotnet.org> wrote:
> On Sat, Nov 18, 2006 at 08:07:08AM +0400, George Sherwood wrote:
> > I seem to be having a problem doing an http checkout with git built
> > with curl 7.16.0 enabled.  If I build against curl 7.16.0 and try a
> > clone, I get:
Show 9 quoted lines
> > git clone http://dmlb2000.homelinux.org/~dmlb2000/git-repos/local/castfs.git
> > error: Unable to start request error: Could not interpret heads/master
> > as something to pull
> > 
> > If I rebuild git against curl 7.15.5 then I get:
> [..]
> > and the checkout finishes.
> > 
> > Has any one else seen this?
> FWIW, I've seen the same with curl 7.16.0 on a Solaris 9 machine.
> It worked fine with curl 7.15.0.

It works fine for me on Aurora Corona (sparc) with curl-7.15.5-1.al3, while it fails as above on Fedora rawhide (i386) with curl-7.16.0-4.fc7.

Furthermore, with new curl pulling from HTTP repos when there are updates gives double free errors and a crash.

-- 
Dr. Horst H. von Brand                   User #22616 counter.li.org
Departamento de Informatica                    Fono: +56 32 2654431
Universidad Tecnica Federico Santa Maria             +56 32 2654239
Casilla 110-V, Valparaiso, Chile               Fax:  +56 32 2797513
Junio C Hamano· Dec 27, 2006, 19:53 UTC · re: Horst H. von Brand · lore
"Horst H. von Brand" <vonbrand@inf.utfsm.cl> writes:
Show 24 quoted lines
> Sven Verdoolaege <skimo@kotnet.org> wrote:
>> On Sat, Nov 18, 2006 at 08:07:08AM +0400, George Sherwood wrote:
>> > I seem to be having a problem doing an http checkout with git built
>> > with curl 7.16.0 enabled.  If I build against curl 7.16.0 and try a
>> > clone, I get:
> ...
>> > git clone http://dmlb2000.homelinux.org/~dmlb2000/git-repos/local/castfs.git
>> > error: Unable to start request error: Could not interpret heads/master
>> > as something to pull
>> > 
>> > If I rebuild git against curl 7.15.5 then I get:
>> [..]
>> > and the checkout finishes.
>> > 
>> > Has any one else seen this?
>
>> FWIW, I've seen the same with curl 7.16.0 on a Solaris 9 machine.
>> It worked fine with curl 7.15.0.
>
> It works fine for me on Aurora Corona (sparc) with curl-7.15.5-1.al3, while
> it fails as above on Fedora rawhide (i386) with curl-7.16.0-4.fc7.
>
> Furthermore, with new curl pulling from HTTP repos when there are updates
> gives double free errors and a crash.

Hmmm. Could somebody please run http-fetch under gdb and see where it breaks? The exact command line you need to use would be obtainable by running "sh -x git-clone" once.

[jc: Nick CC'ed, although I haven't seen him on the list for some time...]

Junio C Hamano· Dec 27, 2006, 21:46 UTC · re: Junio C Hamano · lore
Junio C Hamano <junkio@cox.net> writes:
Show 15 quoted lines
>>> FWIW, I've seen the same with curl 7.16.0 on a Solaris 9 machine.
>>> It worked fine with curl 7.15.0.
>>
>> It works fine for me on Aurora Corona (sparc) with curl-7.15.5-1.al3, while
>> it fails as above on Fedora rawhide (i386) with curl-7.16.0-4.fc7.
>>
>> Furthermore, with new curl pulling from HTTP repos when there are updates
>> gives double free errors and a crash.
>
> Hmmm.  Could somebody please run http-fetch under gdb and see
> where it breaks?  The exact command line you need to use would
> be obtainable by running "sh -x git-clone" once.
>
> [jc: Nick CC'ed, although I haven't seen him on the list for
> some time...]

When we do not have NO_CURL_EASY_DUPHANDLE defined, http.c prepares a single "CURL *curl_default" with get_curl_handle(), and new request slots clone it using curl_easy_duphandle(). However, the returned "clone" somehow forgets to set the "magic" number that says it is a GOOD_EASY_HANDLE(). This missing magic number is checked by cURL library in curl_multi_add_handle(), which we call from start_active_slot(). Hence the request fails.

Now, I do not know cURL, and cannot tell if it is just a bug in easy-duphandle of curl 7.16.0, or if we are not supposed to be using the "easy" interface when dealing with multi fetch. In either case, the attached patch seems to fix it for me.

I am NOT CC'ing curl-library@cool.haxx.se mailing list since I do not want to subscribe only to be able to post, but some kind souls who are subscribers could forward the issue to them to determine if we need to fix our code (and if so how), or if we need to wait for them to fix curl_easy_duphandle().

-- >8 --
diff --git a/http.h b/http.h
index 6e12e41..70b18dd 100644
--- a/http.h
+++ b/http.h
@@ -18,7 +18,7 @@
 #define curl_global_init(a) do { /* nothing */ } while(0)
 #endif
 
-#if LIBCURL_VERSION_NUM < 0x070c04
+#if 1 || (LIBCURL_VERSION_NUM < 0x070c04)
 #define NO_CURL_EASY_DUPHANDLE
 #endif
Junio C Hamano· Dec 29, 2006, 00:40 UTC · re: Junio C Hamano · lore

[PATCH/RFT] Work around http-fetch built with cURL 7.16.0

It appears that curl_easy_duphandle() from libcurl 7.16.0 returns a curl session handle which fails GOOD_MULTI_HANDLE() check in curl_multi_add_handle(). This causes fetch_ref() to fail because start_active_slot() cannot start the request.

For now, check for 7.16.0 to work this issue around.
Signed-off-by: Junio C Hamano <junkio@cox.net>
---
 * I think people who were having trouble with cURL 7.16.0 want
   to have the issue resolved before v1.5.0-rc1.  Please test
   and report, or else ;-).
 http.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/http.h b/http.h
index 6e12e41..324fcf4 100644
--- a/http.h
+++ b/http.h
@@ -18,7 +18,7 @@
 #define curl_global_init(a) do { /* nothing */ } while(0)
 #endif
 
-#if LIBCURL_VERSION_NUM < 0x070c04
+#if (LIBCURL_VERSION_NUM < 0x070c04) || (LIBCURL_VERSION_NUM == 0x071000)
 #define NO_CURL_EASY_DUPHANDLE
 #endif
 

-- 
1.5.0.rc0.gf5c587
Junio C Hamano· Jan 1, 2007, 08:08 UTC · lore
George Sherwood <pilot@beernabeer.com> writes:
Show 7 quoted lines
> I CC'ed the mailing curl mailing list and according to the response
> this problem is fixed in the latest cvs version of curl with this
> commit:
>
> curl_easy_duphandle() sets the magic number in the new handle
>
> I have confirmed this works now git 1.4.4.3
Thanks, so it was a breakage on the .0 release on their side.
I'll leave this in place for now:
    #if (LIBCURL_VERSION_NUM < 0x070c04) || (LIBCURL_VERSION_NUM == 0x071000)
    #define NO_CURL_EASY_DUPHANDLE
    #endif
which specifically marks 7.16.0 as dud.

← back to recent threads