X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?a=blobdiff_plain;f=TODO.BRANCH;h=0b28a74104d6885aa2fe68f68bae9e1cc903016d;hb=e0b5aa648f47d5ec6f3368b23f4aeada81f45546;hp=7e843ae38d5f066466346d777489043ddfd9662c;hpb=2528232969c0f5a7244a6e7d2dea8147790db3a1;p=dgit.git

diff --git a/TODO.BRANCH b/TODO.BRANCH
index 7e843ae3..0b28a741 100644
--- a/TODO.BRANCH
+++ b/TODO.BRANCH
@@ -1,20 +1,81 @@
-
 Tag rename unfinished business
 (things resulting from searching for `tag' where searching for
  `debiantag' wouldn't find them)
 
- - rpush negotiation for which tag to sign
- - test suite tests/lib
+ - protocol version 4
+      implies ability to sign new tag (this is already there
+         so simply advertising version 4 will work)
+      needs also new ability to sign old-format tag in split
+         brain way
+
  - many actual tests
 
  - other tag should be [dgit Maintainer-Tag quilt=$quilt-mode]
 
+ - in dopush (?), we need to sort out synthetic fast forard
+    this should not depend entirely on quilt mode
+    but will always be needed if quiltmode_splitbrain
+     * On push, if ff fixup mode enabled, check debian/changelog,
+       and maybe make fake merge, maybe in dgit view, maybe
+       elsewhere ?
+
+ - there is some confusion between $split_brain (did we actually
+   generate any commits) and quiltmode_splitbrain (might we have done,
+   including in previous pushes); search for $split_brain and check
+   that use sites are correct
+
 
 Want tests for
- - accepting old and new version tags
+ - start repo with old tag, push with new tag, run
+   "old" dgit to push again with new tag, push again with
+   old tag
  - rejecting push of same version with one tag when then the other
 
-----------
+Test that, where appropriate, maint view tags are pushed to dgit git repo
+
+
+ * Provide --gpbpq[=:] options
+
+ * Document --quilt=unapplied|gbp|dpm
+
+ * Mention new --quilt=* in debian/changelog
+
+ * Mention in-archive copy fix in debian/changelog
+
+ * Test --quilt=*
+
+
+ * IN ARCHIVE COPY Consider and implement fast-forward fixup
+
+    - On fetch, check if specified git hash is ff from
+      existing dgit branch, the same way we do for constructed
+      dscs.  The .dsc tells us the git tree but not necessarily
+      the git hash.
+
+    ...ERR MAYBE...
+      
+         complicated there are many branches
+	       dgit-repos: refs/dgit/suite
+               archive dsc
+	       local; remotes/dgit/dgit/suite
+	                  must be ff
+			  may contain dscs
+               local: dgit/suite (not relevant to fetch)
+
+         archive dsc must always be regarded
+	       as ff from (or same as) dgit-repos
+               err depending on version number ?
+
+       this code is what needs changing
+       	} else {
+	    fail "git head (".lrref()."=$lastpush_hash) is not a ".
+		"descendant of archive's .dsc hash ($dsc_hash)";
+	}
+       it just needs to generate the fake merge and set $hash
+       the archive skew stuff is all fine with this
+
+        xxx actually, no, we need to stop from squashing our
+	lrref with + updates
+
+ * Add in-archive-move test case
 
-Provide --gpbpq[=:] options
-Document --quilt=unapplied|gbp|dpm