[PATCH] gitweb: add pf= to limit project list to a subdirectory

This commit changes the project listings (project_list, project_index
and opml) to limit the output to only projects in a subdirectory if the
new optional parameter ?pf=directory name is used.

It uses the infrastructure already there for 'forks' (which also filters
projects but needs a project called like the filter directory to work).

This feature is disabled if strict_export is used and there is no
projects_list to avoid showing more than intended.
Without strict_export enabled this change should not show any projects
one could not get details from anyway. So if the validate_pathname
checks are not sufficient this would at most make it easier to get a
list of viewable content.

Reusing $project instead of adding a new parameter would have been
nicer from a UI point-of-view (including PATH_INFO support) but
complicate the $project validating code that is currently being
used to ensure nothing is exported that should not be viewable.

Signed-off-by: Bernhard R. Link <redacted>

---
As most parameters are not documented in documentation/gitweb.txt,
I did not add documentation for this one either.

 gitweb/gitweb.perl |   26 ++++++++++++++++++++++----
 1 files changed, 22 insertions(+), 4 deletions(-)

diff --git a/gitweb/gitweb.perl b/gitweb/gitweb.perl
index abb5a79..00dd79e 100755
--- a/gitweb/gitweb.perl
+++ b/gitweb/gitweb.perl

@@ -760,6 +760,7 @@ our @cgi_param_mapping = (
 	search_use_regexp => "sr",
 	ctag => "by_tag",
 	diff_style => "ds",
+	project_filter => "pf",
 	# this must be last entry (for manipulation from JavaScript)
 	javascript => "js"
 );

@@ -976,7 +977,7 @@ sub evaluate_path_info {
 
 our ($action, $project, $file_name, $file_parent, $hash, $hash_parent, $hash_base,
      $hash_parent_base, @extra_options, $page, $searchtype, $search_use_regexp,
-     $searchtext, $search_regexp);
+     $searchtext, $search_regexp, $project_filter);
 sub evaluate_and_validate_params {
 	our $action = $input_params{'action'};
 	if (defined $action) {

@@ -994,6 +995,16 @@ sub evaluate_and_validate_params {
 		}
 	}
 
+	our $project_filter = $input_params{'project_filter'};
+	if (defined $project_filter) {
+		if ($strict_export and -d $projects_list) {
+			die_error(404, "project_filter disabled");
+		}
+		if (!validate_pathname($project_filter)) {
+			die_error(404, "Invalid project_filter parameter");
+		}
+	}
+
 	our $file_name = $input_params{'file_name'};
 	if (defined $file_name) {
 		if (!validate_pathname($file_name)) {

@@ -3962,6 +3973,13 @@ sub git_footer_html {
 			              -class => $feed_class}, $format)."\n";
 		}
 
+	} elsif (defined $project_filter) {
+		print $cgi->a({-href => href(project=>undef, action=>"opml",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "OPML") . " ";
+		print $cgi->a({-href => href(project=>undef, action=>"project_index",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "TXT") . "\n";
 	} else {
 		print $cgi->a({-href => href(project=>undef, action=>"opml"),
 		              -class => $feed_class}, "OPML") . " ";

@@ -5979,7 +5997,7 @@ sub git_project_list {
 		die_error(400, "Unknown order parameter");
 	}
 
-	my @list = git_get_projects_list();
+	my @list = git_get_projects_list($project_filter);
 	if (!@list) {
 		die_error(404, "No projects found");
 	}

@@ -6018,7 +6036,7 @@ sub git_forks {
 }
 
 sub git_project_index {
-	my @projects = git_get_projects_list();
+	my @projects = git_get_projects_list($project_filter);
 	if (!@projects) {
 		die_error(404, "No projects found");
 	}

@@ -7855,7 +7873,7 @@ sub git_atom {
 }
 
 sub git_opml {
-	my @list = git_get_projects_list();
+	my @list = git_get_projects_list($project_filter);
 	if (!@list) {
 		die_error(404, "No projects found");
 	}

-- 
1.7.8.3

"Bernhard R. Link" [off-list ref] writes:

This commit changes the project listings (project_list, project_index
and opml) to limit the output to only projects in a subdirectory if the
new optional parameter ?pf=directory name is used.

Could you explain why you want this feature, and why for example
project search just does not cut it?
 

It uses the infrastructure already there for 'forks' (which also filters
projects but needs a project called like the filter directory to work).

It is not entirely clear for me that what you mean here is (I think)
that using

  git_get_projects_list($project_filter);

just works thanks to forks filtering infrastructure.

This feature is disabled if strict_export is used and there is no
projects_list to avoid showing more than intended.
Without strict_export enabled this change should not show any projects
one could not get details from anyway. So if the validate_pathname
checks are not sufficient this would at most make it easier to get a
list of viewable content.

I don't wuite understand this reasoning.  Why project filtering is
disabled with strict_export?  It should just filter, regardless if
project are from scanning $project_filter subdirectory, or filtering
out project names from $projects_list file that do not begin with
$project_filter prefix.
 

Reusing $project instead of adding a new parameter would have been
nicer from a UI point-of-view (including PATH_INFO support) but
complicate the $project validating code that is currently being
used to ensure nothing is exported that should not be viewable.

That is I think quite reasonable.
 

Signed-off-by: Bernhard R. Link <redacted>

---
As most parameters are not documented in documentation/gitweb.txt,
I did not add documentation for this one either.

On the other hand IIRC getting list of projects is quite well
documented in gitweb manpage (or at least it should be).
 
[...]

I wonder if perhaps a better solution wouldn't be to use -replay=>1 in
generating projects list in other formats (OPML and TXT).

Nice!

-- 
Jakub Narebski

* Jakub Narebski [off-list ref] [120127 23:33]:

quoted

This commit changes the project listings (project_list, project_index
and opml) to limit the output to only projects in a subdirectory if the
new optional parameter ?pf=directory name is used.

Could you explain why you want this feature, and why for example
project search just does not cut it?

The project list takes often a very long time and searching in that list
takes the same time (and would also show projects not starting with the
text).

I'd for example like to be able to place a link to all projects shown
at http://anonscm.debian.org/gitweb/ which are below mirrorer/ and get
a not having to wait for description information being extracted for all
the other projects.

quoted

It uses the infrastructure already there for 'forks' (which also filters
projects but needs a project called like the filter directory to work).

It is not entirely clear for me that what you mean here is (I think)
that using

  git_get_projects_list($project_filter);

just works thanks to forks filtering infrastructure.

Yes, it uses the optional git_get_projects_list argument which is
currently only used by action=forks.

quoted

This feature is disabled if strict_export is used and there is no
projects_list to avoid showing more than intended.
Without strict_export enabled this change should not show any projects
one could not get details from anyway. So if the validate_pathname
checks are not sufficient this would at most make it easier to get a
list of viewable content.

I don't wuite understand this reasoning.  Why project filtering is
disabled with strict_export?  It should just filter, regardless if
project are from scanning $project_filter subdirectory, or filtering
out project names from $projects_list file that do not begin with
$project_filter prefix.

strict_export is a security switch to make sure that no unintended
information is exported. Without a project_list all that strict_export
ensures is that there is no way to escape the project_root by giving
some path to a project outside that is not catched by the simple check
against /./ and /../. As the new code would allow a way around this
check, I think the sane thing is to simply disable this new feature in
case of this paranoid check being activated. (And reporting an error
message if it is used).

quoted

---
As most parameters are not documented in documentation/gitweb.txt,
I did not add documentation for this one either.

On the other hand IIRC getting list of projects is quite well
documented in gitweb manpage (or at least it should be).

In Documentation/gitweb.txt there is only a list of possible actions, but no
documentation of any of the other arguments. (none of the order argument
to project_list nor anything else). If actions like blob_plain do list
essential arguments like f= then giving this option special rooms does
not seem to fit very well with the rest.

 
[...]

quoted

@@ -3962,6 +3973,13 @@ sub git_footer_html {
 			              -class => $feed_class}, $format)."\n";
 		}
 
+	} elsif (defined $project_filter) {
+		print $cgi->a({-href => href(project=>undef, action=>"opml",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "OPML") . " ";
+		print $cgi->a({-href => href(project=>undef, action=>"project_index",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "TXT") . "\n";
 	} else {
 		print $cgi->a({-href => href(project=>undef, action=>"opml"),
 		              -class => $feed_class}, "OPML") . " ";

I wonder if perhaps a better solution wouldn't be to use -replay=>1 in
generating projects list in other formats (OPML and TXT).

Wouldn't that also replay options like "order"?

        Bernhard R. Link

* Jakub Narebski [off-list ref] [120128 15:53]:

On Sat, 28 Jan 2012, Bernhard R. Link wrote:

quoted

The project list takes often a very long time and searching in that list
takes the same time (and would also show projects not starting with the
text).

There are at least two ways to speed up getting projects list page.

There are other ways, but they are both limited and quite backward:
Instead of showing me the information I want (what is in that
directory), I guess to search in the larger pool of information needing
all sort of half-working tricks causing all kinds of other problems
working around the issue that much more information never used must be
collected first.

Subversion's ViewVC for example only shows what is in some directory
directory and available subdirectories. (That of course has the
disadvantage to make it harder to find a project one does not know
the subdirectory it is in, but why shouldn't gitweb offer the best of
both worlds?)

First is to limit number of projects shown, like e.g. http://repo.or.cz
which shows only search form and tag cloud, but no projects, or at least
paginate (divide into pages of e.g. 100 projects) list of projects.

But a page showing all projects (as long as it is possible) is something
I'd quite miss if it is gone. What I want is some way to have that and
to also be able to look at some subset interesting to me directly in an
effective way.

Second solution would be to finally add caching support to gitweb,

But caching also means new projects or changes will take some time to
show up. And again it would not be a solution to by original problem.
(Showing effectively what is in some directory).

quoted

I'd for example like to be able to place a link to all projects shown
at http://anonscm.debian.org/gitweb/ which are below mirrorer/ and get
a not having to wait for description information being extracted for all
the other projects.

I would prefer instead of introducing yet another arbitrary parameter
extend project searching, so that you can specify that you want to
search project names only (IIRC I have a patch for that, or beginnings
of one, in my StGit stack), and use prefix search by the way of regexp
search.

So

  pf=mirrorer/

would be

  s=^mirrorer/;sr=1;st=project_name

That might be a workaround, but still need to look at quite some amount
of unecessary data (i.e. it would need to run at least as long as a
http://anonscm.debian.org/gitweb/?a=project_index would not to load,
wouldn't it?)

        Bernhard R. Link

On Sat, 28 Jan 2012, Bernhard R. Link wrote:

* Jakub Narebski [off-list ref] [120127 23:33]:

quoted

quoted

This commit changes the project listings (project_list, project_index
and opml) to limit the output to only projects in a subdirectory if the
new optional parameter ?pf=directory name is used.

Could you explain why you want this feature, and why for example
project search just does not cut it?

The project list takes often a very long time and searching in that list
takes the same time (and would also show projects not starting with the
text).

There are at least two ways to speed up getting projects list page.


First is to limit number of projects shown, like e.g. http://repo.or.cz
which shows only search form and tag cloud, but no projects, or at least
paginate (divide into pages of e.g. 100 projects) list of projects.

I have some patches improving projects list and searching projects, 
including slight speedup to search result generation and project list
pagination in my StGit stack.  I'll clean them up and (re)send them to
git mailing list for discussion.


Second solution would be to finally add caching support to gitweb,
either on the level of caching git command output like Lea Wiemann
work from GSoC2008, or caching intermediate Perl data structures
used by gitweb like in old patches by Petr 'Pasky' Baudis (though only
for projects list page), or caching of final HTML output like in
http://git.kernel.org fork of gitweb.

There is some work in progress that I have in gitweb/cache branch and
in gitweb/cache-kernel* branches in my repository (either on
http://github.com/jnareb/git or http://repo.or.cz/w/git/jnareb-git.git).

The problem with cleaning it up to the form suitable for inclusion is
that it is quite a bit of code, and it very much requires splitting gitweb.
Splitting gitweb in turn requires reworking of current hacky "longjump"
based (non-local goto) error handling into exception based error handling.
That in turn would be best handled using non-core CPAN modules, namely
Try::Tiny and HTTP::Exception or at least Exception::Class.  For that
we need I think a way of bundling requirements with gitweb in inc/ 
a la Module::Install.

Ehhh...

I'd for example like to be able to place a link to all projects shown
at http://anonscm.debian.org/gitweb/ which are below mirrorer/ and get
a not having to wait for description information being extracted for all
the other projects.

I would prefer instead of introducing yet another arbitrary parameter
extend project searching, so that you can specify that you want to
search project names only (IIRC I have a patch for that, or beginnings
of one, in my StGit stack), and use prefix search by the way of regexp
search.

So

  pf=mirrorer/

would be

  s=^mirrorer/;sr=1;st=project_name
 

quoted

quoted

It uses the infrastructure already there for 'forks' (which also filters
projects but needs a project called like the filter directory to work).

It is not entirely clear for me that what you mean here is (I think)
that using

  git_get_projects_list($project_filter);

just works thanks to forks filtering infrastructure.

Yes, it uses the optional git_get_projects_list argument which is
currently only used by action=forks.

This needs to be stated more clearly in the commit message.
 

quoted

quoted

This feature is disabled if strict_export is used and there is no
projects_list to avoid showing more than intended.
Without strict_export enabled this change should not show any projects
one could not get details from anyway. So if the validate_pathname
checks are not sufficient this would at most make it easier to get a
list of viewable content.

I don't quite understand this reasoning.  Why project filtering is
disabled with strict_export?  It should just filter, regardless if
project are from scanning $project_filter subdirectory, or filtering
out project names from $projects_list file that do not begin with
$project_filter prefix.

strict_export is a security switch to make sure that no unintended
information is exported. Without a project_list all that strict_export
ensures is that there is no way to escape the project_root by giving
some path to a project outside that is not catched by the simple check
against /./ and /../. As the new code would allow a way around this
check, I think the sane thing is to simply disable this new feature in
case of this paranoid check being activated. (And reporting an error
message if it is used).

Hmmm... I think this new feature can be made compatible with strict export...
if it is not replaced by search extension, as I have proposed above.

quoted

quoted

---
As most parameters are not documented in documentation/gitweb.txt,
I did not add documentation for this one either.

On the other hand IIRC getting list of projects is quite well
documented in gitweb manpage (or at least it should be).

In Documentation/gitweb.txt there is only a list of possible actions, but no
documentation of any of the other arguments. (none of the order argument
to project_list nor anything else). If actions like blob_plain do list
essential arguments like f= then giving this option special rooms does
not seem to fit very well with the rest.

I was thinking about "Controlling access to git repositories" section in
Documentation/gitweb.txt, but after closer examination I think it wouldn't
fit there.

quoted

 
[...]

quoted

@@ -3962,6 +3973,13 @@ sub git_footer_html {
 			              -class => $feed_class}, $format)."\n";
 		}
 
+	} elsif (defined $project_filter) {
+		print $cgi->a({-href => href(project=>undef, action=>"opml",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "OPML") . " ";
+		print $cgi->a({-href => href(project=>undef, action=>"project_index",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "TXT") . "\n";
 	} else {
 		print $cgi->a({-href => href(project=>undef, action=>"opml"),
 		              -class => $feed_class}, "OPML") . " ";

I wonder if perhaps a better solution wouldn't be to use -replay=>1 in
generating projects list in other formats (OPML and TXT).

Wouldn't that also replay options like "order"?

Right.  My mistake.


Anyway

+		print $cgi->a({-href => href(project=>undef, action=>"opml",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "OPML") . " ";
+		print $cgi->a({-href => href(project=>undef, action=>"project_index",
+		                             project_filter => $project_filter),
+		              -class => $feed_class}, "TXT") . "\n";

would do the right thing because of the way href() works when passed
key-value pair with undefined key.

In other words there is no need for conditional; just add 

   project_filter => $project_filter

and href() would do the right thing.
-- 
Jakub Narebski
Poland

On Sat, 28 Jan 2012, Bernhard R. Link wrote:

* Jakub Narebski [off-list ref] [120128 15:53]:

quoted

On Sat, 28 Jan 2012, Bernhard R. Link wrote:

quoted

quoted

The project list takes often a very long time and searching in that list
takes the same time (and would also show projects not starting with the
text).

There are at least two ways to speed up getting projects list page.

There are other ways, but they are both limited and quite backward:
Instead of showing me the information I want (what is in that
directory), I guess to search in the larger pool of information needing
all sort of half-working tricks causing all kinds of other problems
working around the issue that much more information never used must be
collected first.

Subversion's ViewVC for example only shows what is in some directory
directory and available subdirectories. (That of course has the
disadvantage to make it harder to find a project one does not know
the subdirectory it is in, but why shouldn't gitweb offer the best of
both worlds?)

Well, ViewVC follows Subversion concepts... including broken-by-design
branching by copying ;-)


Anyway, I think I am getting convinced that supporting project filtering
by subdirectory, in a way similar to existing support for forks (and reusing
it implementation) could be a good idea.  

The one serious disadvantage of "pf=" is that it is not solution for a
faster gitweb homepage.  It works _only_ when you have pf=subdirectory/
link.

And I'd rather it worked with strict export, correctly and safely,
if "pf=" is to be included in gitweb.
 

quoted

First is to limit number of projects shown, like e.g. http://repo.or.cz
which shows only search form and tag cloud, but no projects, or at least
paginate (divide into pages of e.g. 100 projects) list of projects.

But a page showing all projects (as long as it is possible) is something
I'd quite miss if it is gone. What I want is some way to have that and
to also be able to look at some subset interesting to me directly in an
effective way.

The http://repo.or.cz homepage has 'Show all projects' link, but it is
at explicit request. 

quoted

Second solution would be to finally add caching support to gitweb,

But caching also means new projects or changes will take some time to
show up. And again it would not be a solution to by original problem.
(Showing effectively what is in some directory).

Well, it would solve "The project list takes often a very long time"
part of your problem.  Though very large list of projects can cause
performance problems rendering on client side, I guess...
 

quoted

quoted

I'd for example like to be able to place a link to all projects shown
at http://anonscm.debian.org/gitweb/ which are below mirrorer/ and get
a not having to wait for description information being extracted for all
the other projects.

I would prefer instead of introducing yet another arbitrary parameter
extend project searching, so that you can specify that you want to
search project names only (IIRC I have a patch for that, or beginnings
of one, in my StGit stack), and use prefix search by the way of regexp
search.

So

  pf=mirrorer/

would be

  s=^mirrorer/;sr=1;st=project_name

That might be a workaround, but still need to look at quite some amount
of unecessary data (i.e. it would need to run at least as long as a
http://anonscm.debian.org/gitweb/?a=project_index would not to load,
wouldn't it?)

Not with 'gitweb: Faster project search' from this series of commits

  # gitweb: Allow underscore in $searchtype ('st')
  # gitweb: Improve projects search form
  # gitweb: Option for filling only specified info in fill_project_list_info
  # gitweb: Faster project search
  # gitweb: Highlight matched part of project name when searching projects
  # gitweb: Highlight matched part of project description when searching projects

which does not search for description nor check for latest changes in
repository for repositories it doesn't show.

-- 
Jakub Narebski
Poland