Re: [gentoo-dev] question regarding inherit

[Andrew Gaffney]

Jason Stubbs wrote:
> On Friday 04 June 2004 08:49, Andrew Gaffney wrote:
>
>>>Jason Stubbs wrote:
>>>
>>>>On Friday 04 June 2004 03:53, foser wrote:
>>>>
>>>>>>On Thu, 2004-06-03 at 19:41 +0200, Marius Mauch wrote:
>>>>>>
>>>>>>>On 06/03/04  Aron Griffis wrote:
>>>>>>>
>>>>>>>>I'm under the impression that this is bad.  Could somebody please
>>>>>>>>verify and explain for the sake of people on this list.  This
>>>>>>>>particular example appears in the libxslt ebuilds, but the kde ebuilds
>>>>>>>>make use of a similar syntax:
>>>>>>>>
>>>>>>>>   use python && inherit python
>>>>>>>
>>>>>>>Bad because it breaks the cache. Depending on the status of the python
>>>>>>>flag *on the rsync master* the cache settings are or are not modified
>>>>>>>by the python.eclass and could be completely different from the values
>>>>>>>the cache should have for the user. That's also the case for other
>>>>>>>cache variables like DEPEND, SRC_URI or SLOT, if you make them
>>>>>>>conditional on the environment it's likely that the cache values are
>>>>>>>wrong for some people.
>>>>>>
>>>>>>The problem is that it shouldn't be a problem to do things like this,
>>>>>>it's portage that makes it problematic. It is not obvious why this is
>>>>>>problematic, so situations like these are going to pop up time and time
>>>>>>again. In my opinion this need to be fixed in portage, not necessarily
>>>>>>in the ebuilds/eclasses. And yeah, i'm talking here without any
>>>>>>knowledge of portage cache handling, ideal world type of talk.
>>>>
>>>>Three options:
>>>>
>>>>1) Drop portage caching altogether.
>>>>2) Separate the global section into a separate non-bash file with new
>>>>syntax. 3) Make everything constant in the global section of ebuilds.
>>>>
>>>>#1 is definately not acceptable at this stage as dep calculation time
>>>>would take about 50 times as long. #2 is probably the best but offers no
>>>>short term solution. Which leaves #3.
>>>
>>>In the Perl Portage clone I was working on, I originally didn't use caching
>>>and it could generate a (mostly) correct deptree for xfree faster than
>>>Portage could. Adding support for Portage's cache only shaved off about .2
>>>seconds.
>
> That's worth investigating then. Perhaps the bulk of the time when doing deps
> without a cache is the actual creation of the cache... It doesn't sit right
> with me, though. I'll look into it further.

Instead of calling bash to parse ebuilds, I wrote a subroutine that does 
bash-style variable interpolation with the ebuilds. It doesn't take 'inherit' 
into account, but that could be fixed easily enough.

sub get_ebuild_info {
   my $pkgname = shift;
   my $pkg = parse_package_name($pkgname);
   my @fieldnames = ('DEPEND',  'RDEPEND', 'SLOT', 'SRC_URI', 'RESTRICT', 
'HOMEPAGE', 'LICENSE', 'DESCRIPTION', 'KEYWORDS', 'INHERITS', 'IUSE', 'CDEPEND', 
'PDEPEND');
   my %ebuildinfo;

   return $pkginfo{$pkgname} if(exists $pkginfo{$pkgname});
   my $cachefname = 
"/var/cache/edb/dep/$pkg->{category}/$pkg->{name}$pkg->{version}$pkg->{suffix}$pkg->{revision}";
   if(! -e $cachefname) {
     my $ebuildcontents;
     my %ebuildvars;
     my $ebuildfname;

     my $pkg = parse_package_name($pkgname);
     $ebuildfname = 
"/usr/portage/$pkg->{category}/$pkg->{name}/$pkg->{name}$pkg->{version}$pkg->{suffix}$pkg->{revision}.ebuild";
     $pkg->{version} =~ s/^-//;
     $ebuildvars{PV} = "$pkg->{version}$pkg->{suffix}";

     open EBUILD, "< $ebuildfname" or die "Couldn't open '$ebuildfname' to get 
DEPEND\n";
     while(<EBUILD>) {
       if(/^\s*if\s*/ || /^\s*#/) {
         next;
       }
       if(/^\s*\w+\(\)\s*{\s*$/) {
         last;
       }
       $ebuildcontents .= $_;
     }
     close EBUILD;

     while($ebuildcontents =~ /\b([-A-Z0-9_]+)\s*=\s*\"(.*?)\"/sgc) {
       my ($varname, $varvalue) = ($1, $2);
       $varvalue =~ s/\$\{?(\w+)\}?/$ebuildvars{$1}/gs;
       $ebuildvars{$varname} = $varvalue;
     }

     my $depend = $ebuildvars{'DEPEND'} || '';
     $depend .= " $ebuildvars{'RDEPEND'}" if(defined $ebuildvars{'RDEPEND'});
     $depend =~ s/(\s+|\n+)/ /gs;

     foreach(@fieldnames) {
       $ebuildinfo{$_} = $ebuildvars{$_};
     }
   } else {
     open DEPCACHE, "< $cachefname" or return undef; #die "Can't open 
'$cachefname'\n";
     foreach(@fieldnames) {
       my $temp = <DEPCACHE>;
       chomp $temp;
       $ebuildinfo{$_} = $temp;
     }
     close DEPCACHE;
   }

   return \%ebuildinfo;
}

-- 
Andrew Gaffney
Network Administrator
Skyline Aeronautics, LLC.
636-357-1548


--
gentoo-dev@g.o mailing list