[PATCH 1 of 4 V2] serve: add support for Mercurial subrepositories

[Yuya Nishihara]

On Sat, 25 Feb 2017 15:25:30 -0500, Matt Harbison wrote:
> On Tue, 21 Feb 2017 09:33:58 -0500, Yuya Nishihara <yuya at tcha.org> wrote:
> >> - At http://localhost:8000, the links on the left side were corrupt  
> >> (graph
> >> is simply 'graph/tip', tags is 'http://tags', etc.)  The pattern seemed  
> >> to
> >> be that things ending in 'tip' are relative URLs, and the rest get an
> >> http: prefix.  It looks from the sources like it is just a matter of '/'
> >> not translating for the {repo} template.
> >
> > [snip]
> >
> > I saw some of these problems in V1. However, the standalone hgweb can  
> > host
> > a repository at '/', so I don't think it's quite difficult to host the  
> > subrepo
> > parent at '/'.
> 
> That's a good point, so now I'm really puzzled over the subtle  
> differences.  It looks like when it finds a hosted repo, it just farms out  
> to hgweb, like the single serve would do.

Yes. hgwebdir is just a dispatcher except for generating an index page.

This is a dirty hack to make hgwebdir serve a repo at '/'.

diff --git a/mercurial/hgweb/hgweb_mod.py b/mercurial/hgweb/hgweb_mod.py
--- a/mercurial/hgweb/hgweb_mod.py
+++ b/mercurial/hgweb/hgweb_mod.py
@@ -335,7 +335,7 @@ class hgweb(object):
         req.url = req.env['SCRIPT_NAME']
         if not req.url.endswith('/'):
             req.url += '/'
-        if 'REPO_NAME' in req.env:
+        if req.env.get('REPO_NAME'):
             req.url += req.env['REPO_NAME'] + '/'
 
         if 'PATH_INFO' in req.env:
diff --git a/mercurial/hgweb/hgwebdir_mod.py b/mercurial/hgweb/hgwebdir_mod.py
--- a/mercurial/hgweb/hgwebdir_mod.py
+++ b/mercurial/hgweb/hgwebdir_mod.py
@@ -254,18 +254,19 @@ class hgwebdir(object):
                 return []
 
             # top-level index
-            elif not virtual:
+            elif False:
                 req.respond(HTTP_OK, ctype)
                 return self.makeindex(req, tmpl)
 
             # nested indexes and hgwebs
 
             repos = dict(self.repos)
-            virtualrepo = virtual
-            while virtualrepo:
-                real = repos.get(virtualrepo)
+            # XXX dirty hack to handle ''. IIRC, we have a helper function
+            virtualrepo = '/' + virtual
+            while True:
+                real = repos.get(virtualrepo[1:])
                 if real:
-                    req.env['REPO_NAME'] = virtualrepo
+                    req.env['REPO_NAME'] = virtualrepo[1:]
                     try:
                         # ensure caller gets private copy of ui
                         repo = hg.repository(self.ui.copy(), real)

> >> To me, the benefits of this series over the original is:
> >>
> >> 1) The repos aren't moved, and therefore don't incur these subtle bugs
> >> 2) These changes are totally isolated to subrepo serving, so there  
> >> should
> >> be no risk to mainstream hgweb usage
> >> 3) This is completely compatible with existing serves with a simple
> >> '/=$projects_dir/**' config file, because the real repos are not  
> >> anchored
> >> to '/'.
> >
> > I think (2) and (3) are up to how we build a repository map.
> 
> What I mean for (2) is that in V2, none of the code changes are hit  
> without -S.  That seems like a good thing, given the subtle breakage in V1.
> 
> What I mean by (3) is that the transition for current clones is to simply  
> drop --web-conf=.. and start using -S on the server side.  Clients won't  
> see a difference.

No code change is nice, but I don't buy it much since the redirection thing
seemed unnecessarily complicated.

> I assume that the hardcoded web.staticurl in V1 also means that it won't  
> be able to serve a subrepo named 'static'?  (That's admittedly an edge  
> case.)

Yes. Also, you can't see /rev/ page if there's a subrepo named 'rev'.

> I only see the two ways to build the repo map, so I'm not sure what you  
> are saying here.

I meant we could delete the root repo from self.repos map on plain hgwebdir
server if that's considered a behavior change. (otherwise the root index
page would be replaced by the hgweb for the root repo.)