patch, 3 partsSupport setting SymrefsOnly=true from scripts

Hi,

this series of patches does three things:

	- support for setting config variables
	- ditto, but from scripts
	- set SymrefsOnly=true right from the start with

		git-init-db --dont-use-symrefs

Ciao, Dscho

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

I am almost tempted to say that we should be doing --template= option to git-init-db for something like this, and make the template instantiation first thing before even reading the config file.

The rest might be useful to better support end-users, but I am not sure. Is it that much hassle to them to edit an .ini file?

Hi,

On Tue, 15 Nov 2005, Junio C Hamano wrote:

Well, problem is that if you use templates to install a .git/config, the code in init-db which checks for safe filemode no longer gets called. And we don't know yet what else we might want to write into the config from init-db, or from anywhere else for that matter.

Furthermore: if you have a nice template which says that SymrefsOnly=true, 
init-db will still create HEAD as a symlink! (Note that my patch sets 
the *global* only_use_symrefs from the command line option, not a 
local variable!)

So, if you want a repository to explicitely force symrefs, you have to have special handling in init-db. And for consistency, you should modify/write the config file to include "SymrefsOnly=true" right away.

Alternatively, or in addition to my patch, you could do a git_config() just after copy_templates() (you allude to this, but note that init-db does not read the config at any stage yet).

However! This only works if you are sure that the templates you need are installed. I *never* install git, but *always* have my private working directory in the PATH. So, I never have templates installed. Neither do I plan to.

Obviously, I'd prefer to have at least the patch adding git_config_set() applied. This is core plumbing. It actually checks if the name is valid, and it automatically escapes the value when needed. Before doing anything, it checks that the config file is valid. It does locking. It creates a simple config if none was there yet. And it's small.

As for git-config-set, I guess it would be rather handy for Porcelains (and t/blabla-config.sh), but I could be convinced otherwise.

Ciao, Dscho

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

Funny. These two are what I meant by "make the template instantiation first thing". So instead of the hardcoded only_symbolic_ref assignment and filemode check code, can't we do something like this?

 - copy the template; if the commandline specifies where the
   templates are, take them from there otherwise use the
   built-in location.

 - read from config; note if you actually read any (may not have
   config in your template suite, or you might not have
   templates at all).

 - do filemode check, and maybe git_config_set() there and write
   the result back as the new config for that repository.

 - create the rest, including the HEAD ref.

The user could even have more than one template set (one for VFAT and another for non VFAT). Once we go that route, we may not even need the special filemode check code in init-db. No?

Dear diary, on Wed, Nov 16, 2005 at 02:09:05AM CET, I got a letter where Junio C Hamano <junkio@cox.net> said that...

The disadvantage is that I will have to maintain my own template for Cogito, which is silly - I would much rather just use GIT's default templates and only add the symrefonly option on behalf of Cogito.

Petr Baudis <pasky@suse.cz> writes:

That's very true. However I suspect Cogito users are expected to run cg-admin-init not git-init-db, so admin-init can do whatever postprocessing necessary after calling init-db? That way would let you do more than what template would, I presume.