share version strings between same versions (of different architectures)

to save some space and allow quick comparisions later on

apt-pkg/deb/debversion.cc

@@ -217,16 +217,12 @@ bool debVersioningSystem::CheckDep(const char *PkgVer,
       return false;
    Op &= 0x0F;
 
-   size_t const lenPkgVer = strlen(PkgVer);
-   size_t const lenDepVer = strlen(DepVer);
-
-   // take a shortcut for equals which are string-equal as well
-   if (Op == pkgCache::Dep::Equals && lenPkgVer == lenDepVer &&
-	 memcmp(PkgVer, DepVer, lenPkgVer) == 0)
-      return true;
+   // fast track for (equal) strings [by location] which are by definition equal versions
+   if (PkgVer == DepVer)
+      return Op == pkgCache::Dep::Equals || Op == pkgCache::Dep::LessEq || Op == pkgCache::Dep::GreaterEq;
 
    // Perform the actual comparision.
-   int const Res = DoCmpVersion(PkgVer, PkgVer + lenPkgVer, DepVer, DepVer + lenDepVer);
+   int const Res = CmpVersion(PkgVer, DepVer);
    switch (Op)
    {
       case pkgCache::Dep::LessEq:

apt-pkg/pkgcachegen.cc

@@ -390,7 +390,7 @@ bool pkgCacheGenerator::MergeListVersion(ListParser &List, pkgCache::PkgIterator
    }
 
    // Add a new version
-   map_ptrloc const verindex = NewVersion(Ver,Version,*LastVer);
+   map_ptrloc const verindex = NewVersion(Ver, Version, Pkg.Index(), Hash, *LastVer);
    if (verindex == 0 && _error->PendingError())
       return _error->Error(_("Error occurred while processing %s (%s%d)"),
 			   Pkg.Name(), "NewVersion", 1);
@@ -398,8 +398,6 @@ bool pkgCacheGenerator::MergeListVersion(ListParser &List, pkgCache::PkgIterator
    if (oldMap != Map.Data())
 	 LastVer += (map_ptrloc*) Map.Data() - (map_ptrloc*) oldMap;
    *LastVer = verindex;
-   Ver->ParentPkg = Pkg.Index();
-   Ver->Hash = Hash;
 
    if (unlikely(List.NewVersion(Ver) == false))
       return _error->Error(_("Error occurred while processing %s (%s%d)"),
@@ -557,7 +555,7 @@ bool pkgCacheGenerator::MergeFileProvides(ListParser &List)
       Dynamic<pkgCache::VerIterator> DynVer(Ver);
       for (; Ver.end() == false; ++Ver)
       {
-	 if (Ver->Hash == Hash && Version.c_str() == Ver.VerStr())
+	 if (Ver->Hash == Hash && Version == Ver.VerStr())
 	 {
 	    if (List.CollectFileProvides(Cache,Ver) == false)
 	       return _error->Error(_("Error occurred while processing %s (%s%d)"),
@@ -768,6 +766,8 @@ bool pkgCacheGenerator::NewFileVer(pkgCache::VerIterator &Ver,
 /* This puts a version structure in the linked list */
 unsigned long pkgCacheGenerator::NewVersion(pkgCache::VerIterator &Ver,
 					    const string &VerStr,
+					    map_ptrloc const ParentPkg,
+					    unsigned long const Hash,
 					    unsigned long Next)
 {
    // Get a structure
@@ -779,12 +779,37 @@ unsigned long pkgCacheGenerator::NewVersion(pkgCache::VerIterator &Ver,
    Ver = pkgCache::VerIterator(Cache,Cache.VerP + Version);
    //Dynamic<pkgCache::VerIterator> DynV(Ver); // caller MergeListVersion already takes care of it
    Ver->NextVer = Next;
+   Ver->ParentPkg = ParentPkg;
+   Ver->Hash = Hash;
    Ver->ID = Cache.HeaderP->VersionCount++;
+
+   // try to find the version string in the group for reuse
+   pkgCache::PkgIterator Pkg = Ver.ParentPkg();
+   pkgCache::GrpIterator Grp = Pkg.Group();
+   if (Pkg.end() == false && Grp.end() == false)
+   {
+      for (pkgCache::PkgIterator P = Grp.PackageList(); P.end() == false; P = Grp.NextPkg(P))
+      {
+	 if (Pkg == P)
+	    continue;
+	 for (pkgCache::VerIterator V = P.VersionList(); V.end() == false; ++V)
+	 {
+	    int const cmp = strcmp(V.VerStr(), VerStr.c_str());
+	    if (cmp == 0)
+	    {
+	       Ver->VerStr = V->VerStr;
+	       return Version;
+	    }
+	    else if (cmp < 0)
+	       break;
+	 }
+      }
+   }
+   // haven't found the version string, so create
    map_ptrloc const idxVerStr = WriteStringInMap(VerStr);
    if (unlikely(idxVerStr == 0))
       return 0;
    Ver->VerStr = idxVerStr;
-   
    return Version;
 }
 									/*}}}*/
@@ -881,7 +906,7 @@ bool pkgCacheGenerator::NewDepends(pkgCache::PkgIterator &Pkg,
 	 index = WriteStringInMap(Version);
 	 if (unlikely(index == 0))
 	    return false;
-	 if (oldMap != Map.Data())
+	 if (OldDepLast != 0 && oldMap != Map.Data())
 	    OldDepLast += (map_ptrloc*) Map.Data() - (map_ptrloc*) oldMap;
       }
    }

apt-pkg/pkgcachegen.h

@@ -80,7 +80,11 @@ class pkgCacheGenerator							/*{{{*/
    bool NewDepends(pkgCache::PkgIterator &Pkg, pkgCache::VerIterator &Ver,
 		   map_ptrloc const Version, unsigned int const &Op,
 		   unsigned int const &Type, map_ptrloc* &OldDepLast);
-   unsigned long NewVersion(pkgCache::VerIterator &Ver,const std::string &VerStr,unsigned long Next);
+   __deprecated unsigned long NewVersion(pkgCache::VerIterator &Ver,const std::string &VerStr,unsigned long Next)
+   { return NewVersion(Ver, VerStr, 0, 0, Next); }
+   unsigned long NewVersion(pkgCache::VerIterator &Ver,const std::string &VerStr,
+			    map_ptrloc const ParentPkg, unsigned long const Hash,
+			    unsigned long Next);
    map_ptrloc NewDescription(pkgCache::DescIterator &Desc,const std::string &Lang,const MD5SumValue &md5sum,map_ptrloc Next);
 
    public:

debian/changelog

@@ -11,6 +11,8 @@ apt (0.9.7.8~exp2+nmu1) UNRELEASED; urgency=low
     - equal comparisions are used mostly in same-source relations,
       so use this to try to reuse some version strings
     - sort group and package names in the hashtable on insert
+    - share version strings between same versions (of different architectures)
+      to save some space and allow quick comparisions later on
   * apt-pkg/pkgcache.cc:
     - assume sorted hashtable entries for groups/packages
   * apt-pkg/cacheiterators.h: