[ Tue Dec  5 15:39:58 PST 2000 ]

We can’t ever convert the existing symbols, those are stuck because we can’t know if we have them all and we have to have them all to construct a graph. So they are always just dangling.

Graph vs old format

Both old and new do the symbolic names like so
Symbols in the graph also have a line like one of the following:
   leaf node, not a merge:
^AcS22 l
   leaf node, merged serials 10 and 20:
^AcS10 20 l
   none leaf node, non merge, someone points at me, my parent is 24:
   none leaf node, merges 101, someone points at me, my parent is 99:
^Ac99 101
   top level node, points at non-existent parent

Data structures: d→ptag /* set if I’m in the graph and have a real parent / d→mtag / set if I am a symbol merge delta / d→symLeaf / set if I am a leaf node in the sym graph / d→symGraph / set if I participate in the sym graph */

IN_SYMGRAPH(d)	(d->ptag || d->symLeaf)

Patch format: S name /* this is the name of the symbol / s g / delta is part of the symbol graph / s l / delta is a leaf in the tag graph / s key / key for the parent delta / s key / key for the merge parent delta */

Adding an old style symbol: just add the entry to the symbol table as usual

Adding a new style symbol add to symbol table, if there is a leaf, find it, clear it, point to it else point to 1 set symLeaf flag

Note: we do not handle LODs yet.

Compat issues: A) old tags stay as they are regardless of what code touches them B) if old code pulls from a new code . translate the file format back to V3 . do not transmit any new tags C) if new code does an opush then . translate file format back to V3 . do not transmit any new tags . do warn about none transmitted tags

Variables are BK version: 1.x | 2.x file version: V3 | V4 tags: original | graph cmd: 1.x pull|push | 2.x opull|opush | 2x pull|push

Tests: Make sure we can create old style tags using new code [x] Make sure we can 1.x pull from new to old - get the csets [x] - get the old style tags [x] - don’t get the new style tags [x] Make sure we can 2.x opull from new to old - get the csets [x] - get the old style tags [x] - don’t get the new style tags [x] Make sure we can 1.x push from old to new - send the csets [x] - send the old style tags [x] Same idea using opush from old to old (hence no new tags present) - send the csets [x] - send the old style tags [x] Same idea using opush from new to old tree - send the csets [x] - send the old style tags [x] - don’t send the new style tags [x] - translate file versions back to V3 [x]


  1. Allow _BK_NO_TAG_GRAPH to override adding nodes to the graph, just add them without the pointers. Needs test case [x]

  2. Always use the 1.0 delta as the ultimate graph root for the symbol tree. This means any time we add a tag with the new code, we root it there. Needs test case [ ]

  3. Need to make sure admin doesn’t whine about tags not in the graph. Needs test case [ ]

  4. May want to have a file flags mode which doesn’t allow a tag graph for transition period.

Symbol graph notes, Sun Oct 29 11:29:36 PST 2000

We have numerous problems resulting from the symbols not participating in the graph the way the regular deltas do. We can’t merge conflicts, we can’t probe the symbol space for fast pulls, etc.

Proposed fix:

add a graph structure as follows ^AcS<parent serial> ^AcS<parent serial> <merge parent serial>

The structure defines a graph similar to the regular graph.

C DATA STRUCTURES struct delta { …​ ser_t ptag; /* the parent of this tag / ser_t mtag; / the merge parent of this tag */ …​ }

When we read in the delta table, we fill these in; when we write it out,
we save these as described above.
0 in these fields means there is no parent/merge parent.

CONVERSION FROM EXISTING FORMAT Construct a straightline graph in time order. So just walk the table in table order (newest to oldest) and for each delta which has a tag, find the next delta which has a tag, and then use that as the ptag value. [ Tue Dec 5 15:40:05 PST 2000 ] This turned out to be WRONG. You can’t autoconvert without having all the data and you can never know that you have all the data. Wrong.

LOD IMPLICATIONS We need to allow multiple open tips so long as the there is exactly one open tip per lod. In other words, when we write a resolver for this stuff, we need to not merge across lod boundaries.

PULL / PUSH We have to probe the tag graph the same we probe the delta graph.

BACKWARDS COMPAT This has to be a file format rev.

TODO + Pass the tags pointers in the patch file. + Bump the version number + Make sure the tag parent gets set when you add a tag (I think there are multiple ways to add a tag). [done, I think] + admin -hhh needs to check the tag graph for multiple open tips + need an admin command to fake merge nodes for the tags; autoinvoke this when bumping the file format.

INVARIANTS Once a tag has a parent/merge parent, those links never change, just like in the regular graph.

RESOLVER We need to walk both branches and if there is the same tag on both, invoke an interactive resolver. If there is not, we can "automerge".

We need some sort of interface to do a tag merge delta.

IMPLEMENTATION NOTES I need to make public the interface which tells me there are tag conflicts so if that is the only thing which changed, I know it.

We need a multiple tip probe protocol sort of
Eventually we may want this to be smart about not probing up the trunk
each time.
For the match[es] we just respond with all the matches we found and color
up from each of the matches.


In lmsccs, symbols are used in one of two ways:

  1. to name an absolute revision, i.e., 1.1 or but not 1 or 1.2.1

  2. to name a line of development (a branch or trunk), i.e, 1, 2 or 1.2.1

The first is the same as RCS.

The second can be the same as RCS if the RCS flag is bit 2 set.

If the RCS compat mode is off, then we are in new territory with SCCS line of development (LOD) naming. The idea is pretty simple. A symbol names a new LOD and revisions start at 1 in that LOD. It’s a naming system layered on top of the more arcane a.b.c.d naming system that SCCS has. So if I tag rev 1.20 as NEW then I can get at rev 1.21 as either "1.21" or "NEW.1".

XXX - to make this work, advisory messages should use the tag naming, not the rev naming to educate people.