Re: [PATCH/WIP] Starting work on a man page for /etc/gitweb.conf

On Thu, 2011-05-12 at 17:01 +0200, Jakub Narebski wrote:

On Thu, 12 May 2011, Jonathan Nieder wrote:

quoted

Drew Northup wrote:

quoted

This is a work in progress. Much of what is in it has been pulled
directly from the README and INSTALL files of gitweb. No effort has yet
been made to de-duplicate any of this.

While it might be a good idea to split main part of gitweb/README into
gitweb.conf.txt (documenting configuration), and perhaps also separate
gitweb.txt (main page for gitweb, like SVN::Web manpage), I don't think
that much of gitweb/INSTALL should be moved.

That's pretty much my conclusion as well. I'm focusing solely on the
configuration portion right now. A main page for Gitweb itself will
definitely be useful.

quoted

quoted

TODO:
  * Clean up README and INSTALL files
  * Add Makefile rules to build man / HTML pages.
  * Remove or rephrase redundant portions of original documentation
  * A lot more...

I agree with this TODO list. :)  It should be possible to reuse rules from
Documentation/Makefile if you put this under Documentation/.  gitweb already
keeps its tests under t/ for convenience; I think it's okay if it
puts some documentation under Documentation/.

Note that git-gui and gitk both also keep their manpages in Documentation/
as Documentation/git-gui.txt and Documentation/gitk.txt

I've been keeping my work in the gitweb directory mostly for
convenience. I've put my "compilation" script there for testing the
asciidoc compilation just so that I don't need to mess with make for the
time being.

We can add "doc" target to gitweb/Makefile, which would delegate work to
../Documentation/Makefile, similarly to existing "test" target in
gitweb/Makefile.

That makes sense to me. Shouldn't be too much work for me to piece that
together when I'm ready.

quoted

quoted

--- /dev/null
+++ b/gitweb/gitweb.conf.txt

@@ -0,0 +1,294 @@
+gitweb.conf(5)
+==============
+
+NAME
+----
+gitweb.conf - Gitweb configuration file

It sounds like a tautology.  Maybe "configuration for git's web
interface"?  Except that there is at least one other web interface for
git (cgit).  Hm.

Let's take a look how it is done in other section 5 manpages documenting
configuration files:

  dhclient.conf(5):  dhclient.conf - DHCP client configuration file
  ldap.conf(5):      ldap.conf, .ldaprc - ldap configuration file
  yum.con(5):        yum.conf - Configuration file for yum(8).

So it is not much of tautology, I think.

I had started with resolv.conf(5) actually. Same idea--concise but
perhaps not amazingly informative.

quoted

quoted

+
+SYNOPSIS
+--------
+/etc/gitweb.conf

I'd say

    +SYNOPSIS
    +--------
    +gitweb_conf.perl
    +/etc/gitweb.conf

or

    +SYNOPSIS
    +--------
    +$GITWEBDIR/gitweb_conf.perl
    +/etc/gitweb.conf
 

quoted

gitweb will also look for gitweb_config.perl along @INC, and
the $GITWEB_CONFIG and $GITWEB_CONFIG_SYSTEM envvars can override
these paths.

I think that we don't need to describe envvars in synopsis, but we
should have per-gitweb configuration file (gitweb_conf.perl) in
"Synopsis" section.

Sounds reasonable to me.
 

quoted

quoted

+
+DESCRIPTION
+-----------
+'Gitweb' is a CGI application for viewing Git repositories over the web. The
+configuration file is used to override the default settings that were built
+into gitweb at the time Git itself was compiled.

Style nit: I'd launch into what gitweb.conf contains right away, like
so:

	The gitweb CGI script for viewing Git repositories over the
	web uses a perl script fragment as its configuration file.
	You can set variables using "our $variable = value"; text
	from a "#" character until the end of a line is ignored.
	See *perlsyn*(1) for details.

	The configuration file is used to override the default
	settings that were built into gitweb at the time it was
	installed.

I agree with Jonathan here.

Comment noted. Perhaps resolv.conf(5) wasn't the best example for
this...
 

quoted

quoted

+While one could just alter
+the configuration settings in the gitweb CGI itself, those changes would be
+lost upon upgrade. Configuration settings my also be placed into a file in
+the same directory as the CGI script with the default name
+`gitweb_config.perl` – allowing one to have multiple gitweb instances
+with different configurations by the use of symlinks.

Good point; I hadn't thought about the multiple-configurations use case.

Right.

quoted

quoted

+DISCUSSION
+----------
+
+The location of `gitweb.conf` is defined at compile time using the
+configuration value `GITWEB_CONFIG_SYSTEM` and defaults to /etc/gitweb.conf.
+The name of the per-instance configuration file is defined in gitweb by
+`GITWEB_CONFIG`.
+
+*NOTE:* Values defined in the per-instance configuration file override both
+values found in the gitweb CGI as well as values found in the sytem-wide
+gitweb.conf file.

Doesn't gitweb_config.perl suppress the effect of gitweb.conf altogether?

Yes it does.

The sequence is as following:

1. Per-gitweb configuration file is given by envvar GITWEB_CONFIG; if it
   is not set, then by default gitweb_conf.perl is used (one can override
   the latter name via build-time configuration variable GITWEB_CONFIG).
   Note: relative path means relative to installed gitweb.cgi script.

2. System-wide configuration file is given by envvar GITWEB_CONFIG_SYSTEM;
   if it is not set, then by default /etc/gitweb.conf is used (one can
   override the latter name via build-time configuration variable
   GITWEB_CONFIG_SYSTEM).  Note: sysconfdir is not taken into account
   by gitweb/Makefile, but perhaps it should.
 
Gitweb obtains configuration data from the following sources in the
following order:

  1. gitweb's installation configuration file ($GITWEBDIR/gitweb_conf.perl)
  2. system-wide configuration file (/etc/gitweb.conf)

First existing config file is used.

Let me see if I understand this correctly: If there is an instance-local
configuration file ($GITWEBDIR/gitweb_conf.perl by default) we outright
ignore all settings in the system-wide (/etc/gitweb.conf) configuration
file? That would mean that the system-wide configuration file isn't
really a system-wide set of defaults--to be built upon by a local
configuration--at all, but is something much more akin to /etc/skel;
used in total or completely overridden.

Coming from the perspective of an administrator of a fair amount of web
hosting space I'd actually prefer that it be settable that both the
system-wide configuration and the local one affect operation. This would
allow for local setting of things such as adding a stylesheet while
ensuring consistency otherwise. The order of evaluation of configuration
settings sets would be "built-in, system, local" in that case (provided
all exist). I can provide a patch to gitweb.perl for that if there is
interest.

Sidenote: we could replace GITWEB_CONFIG and GITWEB_CONFIG_SYSTEM in 
gitweb.conf.txt during building documentation.

I wasn't sure yet if I wanted to do @@thingy@@ substitution or not. (I'm
still not...)

quoted

quoted

+
+The syntax of the configuration files is that of PERL, as these files are
+indeed handled as fragments of PERL code (the language that gitweb itself is
+written in). Variables may be set using "'our $variable = value'"; text from
+"#" character until the end of a line is ignored. See the perlsyn(1) man page
+for more information.

Actually using 'our $variable = <value>' is a safety check: if newer gitweb
does no longer use given value, using 'our' wouldn't cause errors.

That was boilerplate language taken pretty directly from the INSTALL and
README files. If we change it here we should change it there as well (if
it remains once we're done).

quoted

The perl manual spells the name of that language as "Perl".  

I've seen it spelled all sorts of ugly ways, but if that's what they're
using I can too. ;-)

I think

quoted

it might make sense to mention the syntax earlier, since it makes it
less daunting to dive into the file right away.

I think it might be good idea to provide bare-bones example here.
 

quoted

quoted

+
+One good reason to take advatage of the system-wide and local gitweb
+configuration files is that not all settings may be set up directly in the CGI
+itself. Optional features – defined using the '%features' variable
+– must be set in one of the two configuration files.

I don't follow what this paragraph is saying.  Is the idea something
like this?

	The default configuration with no configuration file at all
	may work perfectly well for some installations.  Still, a
	configuration file is useful for customizing or tweaking the
	behavior of gitweb in many ways, and some optional features
	will not be present unless explicitly enabled using the
	configurable %features variable.

I think the idea is that not all of configuration knobs can be tweaked
during "compile"-time.  Some require setting from configuration file.

Yes, that would benefit from some rewording.

Note: we probably want to mention gitweb/config.mak or config.mak somewhere
as place to save build-time configuration, persistently.

Given your preference for keeping gitweb/INSTALL as intact as possible,
should that not go there?

quoted

quoted

+
+CONFIGURATION SETTINGS
+----------------------

Configuration settings as opposed to non-configuration settings? :)

Maybe something like

	VARIABLES
	---------

would be clearer.

Perhaps.

But variables to do what?

quoted

quoted

+Standard Options
+~~~~~~~~~~~~~~~~~
+The following are not typically set or overridden at build time:

[...]

quoted

quoted

+
+$GIT::
+	Core git executable to use.

Ah, what a variable to start with.  Maybe this can be snuck later in
the list somehow, so the reader can get to juicier bits first.

This is set at build time, by default to $(bindir)/git

I chose to leave that in due to the fact that we've had a few threads on
the list and other places on the Internet in recent memory where it was
asked how to override that. Order is thus far as given in my source
materials. If it is set at build time I can move it to the other
(non-standard) list.

quoted

quoted

+$version::
+	Gitweb version, set automatically when creating gitweb.cgi from
+	gitweb.perl. You might want to modify it if you are running modified
+	gitweb.

Why would I want to set this in the config file?  Wouldn't my patch to
gitweb.cgi modify $version?

Well, for running gitweb.perl (not build!) from command line I use

  our $version = "current";

But this is I think rare case.

I would hope so. In any case, as we part gitweb out I suspect it will be
useful to mention somewhere that this is settable, as distribution
maintainers are apt to fuss with things (for starters). As far as I'm
concerned the README and INSTALL files are fair game for modification if
this could be made more clear in one of them.

quoted

quoted

+$projectroot::
+	Absolute filesystem path which will be prepended to project path;
+	the path to repository is `$projectroot/$project`.  Set to
+	`$GITWEB_PROJECTROOT` during installation.  This variable has to be
+	set correctly for gitweb to find repositories.

This is an interesting one.  The description is not so clear to me ---
I guess the idea is that if $projectroot = "/srv/git" then

 http://path/to/gitweb/installation/?p=foo/bar.git

will map to /srv/git/foo/bar.git on the filesystem (and likewise for
PATH_INFO based URLs)?

Yes.

This was the one which bit me hard enough to decide to take this task
on. I really didn't appreciate having my gitweb and gitolite
installations which had been getting along so well suddenly refuse to
talk to each other.

quoted

quoted

+$projects_list::
+	Source of projects list, either directory to scan, or text file
+	with list of repositories (in the "`<URI-encoded repository path> SP
+	<URI-encoded repository owner>`" line format; actually there can be
+	any sequence of whitespace in place of space (SP)).  Set to
+	`$GITWEB_LIST` during installation.  If empty, `$projectroot` is used
+	to scan for repositories.

Maybe clearer to emphasize the kinds of values it takes first?  That
is:

	Space-separated list of paths to files listing projects or
	directories to be scanned for projects.  Project list files
	should list one project per line, with each line having the
	format "`<URI-encoded filesystem path to repository> SP
	<URI-encoded repository owner>`.  The default is determined
	by the GITWEB_LIST makefile variable at installation time.
	If this variable is empty, gitweb will fall back to scanning
	the `$projectroot` for repositories.

I think to not have long wall of text here, we should reference part
of documentation that explains how gitweb finds repositories, including
format of $projects_list file.

Perhaps this longer explanation is fodder for gitweb.txt?

quoted

[...]

quoted

+Configuration Options Often Set at Compile Time
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+These configuration variables are often specified at compile time and are
+defined by default in the gitweb CGI itself:
+
+GIT_BINDIR::
+	Points where to find the git executable.  You should set it up to
+	the place where the git binary was installed (usually /usr/bin) if you
+	don't install git from sources together with gitweb.  [Default: $(bindir)]

I don't think there is a GIT_BINDIR configuration variable, though
there is a makefile variable with that name used to determine the
default value of $GIT.

Ok, upon checking that is indeed "compiled out" during the build
process.

quoted

Likewise for the others.  I don't think they belong in the manpage.

I agree.  What is important during build time "only" should remain in
gitweb/INSTALL.

We need to define what that is then, as some of these "build time only"
listed values are indeed settable items (granted, once the case is set
properly and a dollar sign is prefixed) such as $site_name,
$site_header, and $site_footer. 

Remember, many people install from packages and not from source. If we
make them second-class citizens we only make our lives more difficult.

In any case, as I noted earlier, thus far I've been mostly just pulling
stuff in from other places. It is now time to start cleaning it up.

-- 
-Drew Northup
________________________________________________
"As opposed to vegetable or mineral error?"
-John Pescatore, SANS NewsBites Vol. 12 Num. 59