hg merge: some getchangegroup fixups -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 hg merge: some getchangegroup fixups - - filter more duplicates - - request batching - - lookup ordering - - request counting After these fixes, grabbing 1800 changesets on top of the 28k changeset kernel history takes 23 round trips rather than a couple hundred. manifest hash: 78e893b68000267a8c7ea49fd38758f2fc96269f -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (GNU/Linux) iD8DBQFCrea0ywK+sNU5EO8RAv3cAKCQveivdApJIlslTCMLMp2VogRXfACeNRnx Ihsf5yLz60fFpqlzAmZBACk= =Euku -----END PGP SIGNATURE-----

lazyparser speed ups -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 lazyparser speed ups When we do __contains__ on a map, we might as well load the whole index. Not doing this was slowing down finding new changesets quite by a factor of 20. When we do a full load, we also attempt to replace the revlog's index and nodemap with normal Python objects to avoid the lazymap overhead. manifest hash: 9b2b20aacc508f9027d115426c63a381d28e5485 -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (GNU/Linux) iD8DBQFCreYIywK+sNU5EO8RAoNHAJ9+LmXqsTQb9Bh3mZHq0A0VfQOleQCffHmn jC/O0vnfx5FCRsX2bUFG794= =BDTz -----END PGP SIGNATURE-----

hgweb: Fix reloading of index -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 hgweb: Fix reloading of index We were reloading every request because we weren't saving the old mtime. manifest hash: 507330727957ff53c48b38f427f0b70469a49271 -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (GNU/Linux) iD8DBQFCreVpywK+sNU5EO8RAp/YAJ9cXh2II4DGAQtRc58yj4hi49C8jgCfT7bV 22w73T2FiYBjzMk8N8STUVI= =eghX -----END PGP SIGNATURE-----