What's in git.git tonight

What's in git.git tonight

[Junio C Hamano]

Here are the things merged to "master" so far:

 - The first part of the http-push over DAV; git-push frontend
   is not aware of it yet; interested people please try it out
   with git-http-push and send thanks-it-works messages to Nick.

 - Overriding CFLAGS from make command line did not quite work
   if you were on a platform that needed to acquire platform
   specific defines from the Makefile.  Hopefully this has been
   fixed.

 - My attempt to split the Debian binary package into multiple
   ones.

 - Some documentation fixes.

 - A fix to format-patch.

In the proposed updates branch:

 - svnimport fixes from Yaacov; svnimport users please either
   Ack or report regression.

 - 'git binary directory' changes, started by Linus.  I am
   running with gitdir != bindir configuration on my machine and
   it seems to work, but I'd want wider audience to test this.
   Note that the Debian binary packages have not been adjusted
   for this change yet.

 - A 'git clone' change to fully populate remotes/origin; this
   is more or less an RFC item; likes?  dislikes?

 - A change to git-push to allow push-over-http-DAV.  I'd like
   to push this part out after I hear success stories from
   people on the first half.

I plan to do a 0.99.9e Sunday night with the tip of "master"
(wherever it will be by that time).

Re: What's in git.git tonight

[Randal L. Schwartz]

http-push.c no longer compiles on OSX:

gcc -o http-push.o -c -g -O2 -Wall -DSHA1_HEADER='<openssl/sha.h>' http-push.c
http-push.c:10:19: error: expat.h: No such file or directory
http-push.c:1113: warning: type defaults to 'int' in declaration of 'XML_Char'
http-push.c:1113: error: parse error before '*' token
http-push.c: In function 'activelock_cdata':
http-push.c:1115: error: 'userData' undeclared (first use in this function)
http-push.c:1115: error: (Each undeclared identifier is reported only once
http-push.c:1115: error: for each function it appears in.)
http-push.c:1116: error: 'len' undeclared (first use in this function)
http-push.c:1117: error: 's' undeclared (first use in this function)
http-push.c: In function 'lock_remote':
http-push.c:1187: error: 'XML_Parser' undeclared (first use in this function)
http-push.c:1187: error: parse error before 'parser'
http-push.c:1188: error: storage size of 'result' isn't known
http-push.c:1276: warning: implicit declaration of function 'XML_SetUserData'
http-push.c:1276: error: 'parser' undeclared (first use in this function)
http-push.c:1277: warning: implicit declaration of function 'XML_SetElementHandler'
http-push.c:1279: warning: implicit declaration of function 'XML_SetCharacterDataHandler'
http-push.c:1280: warning: implicit declaration of function 'XML_Parse'
http-push.c:1282: error: 'XML_STATUS_OK' undeclared (first use in this function)
http-push.c:1283: warning: implicit declaration of function 'XML_ErrorString'
http-push.c:1284: warning: implicit declaration of function 'XML_GetErrorCode'
http-push.c:1284: warning: format '%s' expects type 'char *', but argument 3 has type 'int'
http-push.c:1188: warning: unused variable 'result'
http-push.c: In function 'check_locking':
http-push.c:1353: error: 'XML_Parser' undeclared (first use in this function)
http-push.c:1353: error: parse error before 'parser'
http-push.c:1354: error: storage size of 'result' isn't known
http-push.c:1392: error: 'parser' undeclared (first use in this function)
http-push.c:1397: error: 'XML_STATUS_OK' undeclared (first use in this function)
http-push.c:1399: warning: format '%s' expects type 'char *', but argument 2 has type 'int'
http-push.c:1354: warning: unused variable 'result'


-- 
Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
<merlyn@stonehenge.com> <URL:http://www.stonehenge.com/merlyn/>
Perl/Unix/security consulting, Technical writing, Comedy, etc. etc.
See PerlTraining.Stonehenge.com for onsite and open-enrollment Perl training!

Re: What's in git.git tonight

[Paul Collins]

merlyn@stonehenge.com (Randal L. Schwartz) writes:

> http-push.c no longer compiles on OSX:

It seems to require libexpat now.  I got the same error here on Debian.
With libexpat-dev installed it builds.

-- 
Dag vijandelijk luchtschip de huismeester is dood

Re: What's in git.git tonight

[Randal L. Schwartz]

>>>>> "Paul" == Paul Collins <paul@briny.ondioline.org> writes:

Paul> It seems to require libexpat now.  I got the same error here on Debian.
Paul> With libexpat-dev installed it builds.

Well, there's no standard libexpat for OSX, so if you install it
after-market, it can end up in various directories.

For example, on my machine, I've installed it with fink, so I need
"-I/sw/include -L/sw/lib" added to CFLAGS.

If you install it with darwinports, it will end up under
"/opt/local/{lib,include}" instead.

I suppose you could add those four things to CFLAGS for OSX and it
won't mess too many things up, but you'll also need to add a note that
says "git requires expat, which can be obtained through fink or
darwinports.  If you install expat manually, you may have to adjust
CFLAGS in the Makefile".

As a separate problem, why make git depend on expat if the only part
of this is for DAV pushing?  Can http-push simply refuse DAV URLs if
built without expat?

-- 
Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
<merlyn@stonehenge.com> <URL:http://www.stonehenge.com/merlyn/>
Perl/Unix/security consulting, Technical writing, Comedy, etc. etc.
See PerlTraining.Stonehenge.com for onsite and open-enrollment Perl training!

Re: What's in git.git tonight

[Marco Roeland]

On Sunday November 6th 2005 at 02:54 uur Randal L. Schwartz wrote:

> 
> http-push.c no longer compiles on OSX:
> 
> gcc -o http-push.o -c -g -O2 -Wall -DSHA1_HEADER='<openssl/sha.h>' http-push.c
> http-push.c:10:19: error: expat.h: No such file or directory

You need to have the expat (XML parsing library) header files installed
from something like the "expat development kit".

If you do already have 'expat.h' installed under say /usr/include does
changing "#include "expat.h" to "#include <expat.h>" perhaps help?
Searching for include files specified with #include "..." is somewhat
platform dependant.

Alternatively, if you don't need 'git-http-push' you can define
'NO_EXPAT' and the Makefile will take care that it doesn't get built.

Incidentally 'git whatchanged -p http-push.c' shows that this dependency
on expat and its include header has been there since the first version
on November 2nd. So 'no longer compiles' seems a bit odd!
-- 
Marco Roeland

Re: What's in git.git tonight

[Randal L. Schwartz]

>>>>> "Marco" == Marco Roeland <marco.roeland@xs4all.nl> writes:

Marco> Incidentally 'git whatchanged -p http-push.c' shows that this dependency
Marco> on expat and its include header has been there since the first version
Marco> on November 2nd. So 'no longer compiles' seems a bit odd!

I haven't built since nov 2.  Sorry, bit distracted with a conference. :)

-- 
Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
<merlyn@stonehenge.com> <URL:http://www.stonehenge.com/merlyn/>
Perl/Unix/security consulting, Technical writing, Comedy, etc. etc.
See PerlTraining.Stonehenge.com for onsite and open-enrollment Perl training!

Re: What's in git.git tonight

[Randal L. Schwartz]

>>>>> "Randal" == Randal L Schwartz <merlyn@stonehenge.com> writes:

Randal> As a separate problem, why make git depend on expat if the only part
Randal> of this is for DAV pushing?  Can http-push simply refuse DAV URLs if
Randal> built without expat?

Ignore that.  Yes, defining "NO_EXPAT" is enough.  Didn't notice that there.

And, I've confirmed that this patch does the trick, and will probably
defer any other issues with "optional" packages in the future:

diff --git a/Makefile b/Makefile
index 6f9b0d1..f89e0bd 100644
--- a/Makefile
+++ b/Makefile
@@ -185,6 +185,10 @@ uname_O := $(shell sh -c 'uname -o 2>/de
 ifeq ($(uname_S),Darwin)
        NEEDS_SSL_WITH_CRYPTO = YesPlease
        NEEDS_LIBICONV = YesPlease
+       ## fink
+       ALL_CFLAGS += -I/sw/include -L/sw/lib
+       ## darwinports
+       ALL_CFLAGS += -I/opt/local/include -L/opt/local/lib
 endif
 ifeq ($(uname_S),SunOS)
        NEEDS_SOCKET = YesPlease

Note that whitespace may have been eaten... this was a cut-n-paste.

-- 
Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
<merlyn@stonehenge.com> <URL:http://www.stonehenge.com/merlyn/>
Perl/Unix/security consulting, Technical writing, Comedy, etc. etc.
See PerlTraining.Stonehenge.com for onsite and open-enrollment Perl training!

Re: What's in git.git tonight

[Johannes Schindelin]

Hi,

On Sun, 6 Nov 2005, Junio C Hamano wrote:

>  - A 'git clone' change to fully populate remotes/origin; this
>    is more or less an RFC item; likes?  dislikes?

melikes.

Re: What's in git.git tonight

[Junio C Hamano]

merlyn@stonehenge.com (Randal L. Schwartz) writes:

> And, I've confirmed that this patch does the trick, and will probably
> defer any other issues with "optional" packages in the future:

Thanks.  Next time around could you sign-off your patch?

Re: What's in git.git tonight

[Paul Collins]

Junio C Hamano <junkio@cox.net> writes:

> merlyn@stonehenge.com (Randal L. Schwartz) writes:
>
>> And, I've confirmed that this patch does the trick, and will probably
>> defer any other issues with "optional" packages in the future:
>
> Thanks.  Next time around could you sign-off your patch?

A tiny nit:

---

Consistency is the hobgoblin of small minds, and mine is tiny indeed.
Do not search the current directory when including expat.h, since it
is not supplied by git.

Signed-off-by: Paul Collins <paul@briny.ondioline.org>

diff --git a/http-push.c b/http-push.c
index c10067c..89fda42 100644
--- a/http-push.c
+++ b/http-push.c
@@ -7,7 +7,7 @@
 
 #include <curl/curl.h>
 #include <curl/easy.h>
-#include "expat.h"
+#include <expat.h>
 
 static const char http_push_usage[] =
 "git-http-push [--complete] [--force] [--verbose] <url> <ref> [<ref>...]\n";


-- 
Dag vijandelijk luchtschip de huismeester is dood

What's in git.git tonight

[Junio C Hamano]

Since 0.99.9n, there have been a couple of fixes and some
documentation updates.  One notable is that we do not allow '?',
'*' and '[' in ref names anymore --- this was done after the
list discussion with Pasky and friends.

Another notable is "We do not like HEAD branch" patch, but I've
been thinking seriously about reverting it and replace its
effect by also reverting the misguided attempt to disambiguate
branch names and tag names.  This patch with a commit message
will follow in a separate message; it currently lives in the
proposed updates branch.  One positive effect this change brings
is that it fixes a corner case bug/confusion Johannes mentioned
the other day while diagnosing the trouble Len had with his
repository (I do not think the problem has anything to do with
Len's trouble, though).  If you have a branch called "dead", and
you also have a tag called "dead", and if your repository has
only one object whose name begins with "dead"
(e.g. "deadbeef1234..."), "git rev-parse --verify dead" would
pick up the "deadbeef1234..."  object, not "dead" tag nor "dead"
head.  When no such object exists, "git rev-parse --verify dead"
fails, saying "dead" is ambiguous between heads and tags.  This
is just confused, and "reverting the misguided disambiguation"
change will make it pick up the "dead" tag in either case.  Most
likely I'll have it in "master" after further testing over the
weekend.

One important patch waiting in the proposed updates is to give
an option to git-fetch-pack to keep the downloaded pack without
unpacking it; this was primarily done to help Cogito, but I
haven't heard back about it from Pasky yet.  The primary
difference between this and the clone-pack change that is
already in the master branch, from the user's point of view, is
that git-fetch-pack can be efficiently used in an already
populated repository [*1*].  Imagine cloning "master" branch
from linux-2.6 repository of Linus, and then fetching ALL branch
from libata-dev repository of Jeff --- the difference is large
enough that you would want to keep the downloaded stuff packed,
while you do want to take advantage of the fact that you already
have objects from Linus.  Currently it saves exploding about a
~800K pack with 1100 objects in it.  Although I mentioned we are
supposed to be in deep feature freeze, I feel it is worth to
have this one in 1.0.


[Footnote]

*1* Theoretically we could deprecate git-clone-pack and use
git-fetch-pack exclusively, if we are willing to create refs
matching what the remote has in the wrapper script git-clone.