1 Here we describe the update algorithm. This is responsible for
2 refreshing patches against updated versions of their dependencies,
3 for merging different versions of the various braches created by
4 distributed development, and for implementing decisions to add and
5 remove dependencies from patches.
7 Broadly speaking the update proceeds as follows: during the Ranking
8 phase we construct the intended graph of dependencies between patches
9 (which involves select a merge order for the base branch of each
10 patch). Then during the Traversal phase we walk that graph from the
11 bottom up, constructing for each patch by a series of merges and other
12 operations first a new base branch head commit and then a new tip
13 branch head commit. These new head commits are maximums - that is,
14 each has as ancestors all of its branches' sources and indeed all
15 relevant commits in that branch.
17 We have two possible strategies for constructing new base branch
18 heads: we can either Merge (works incrementally even if there the
19 patch has multiple dependencies, but may sometimes not be possible) or
20 we can Regenerate (trivial if there is a single dependency, and is
21 always possible, but may involve the user re-resolving conflicts if
22 there are multiple dependencies).
28 \desclabelstyle{\nextlinelabel}
30 \item[ $\depsreqof{K}$ ]
31 The set of direct dependencies (in the form $\py$)
32 requested in the commit $K$ ($K \in \pn$) for the patch $\p$.
34 \item[ $\pc \hasdirdep \p$ ]
35 The Topbloke commit set $\pc$ has as a direct contributor the
36 commit set $\p$. This is an acyclic relation.
38 \item[ $\p \hasdep \pq$ ]
39 The commit set $\p$ has as direct or indirect contributor the commit
41 Acyclic; the completion of $\hasdirdep$ into a
44 \item[ $\pendsof{\set J}{\p}$ ]
45 Convenience notation for
46 the maximal elements of $\bigcup_{J \in \set J} \pendsof{J}{\p}$
47 (where $\set J$ is some set of commits).
49 \item[ $\pendsof{\set X}{\p} \le T$ ]
50 Convenience notation for
51 $\bigforall_{E \in \pendsof{\set X}{\p}} E \le T$
53 %\item[ $\set E_{\pc}$ ]
54 %$ \bigcup_i \pendsof{S_{\pc,i}}{\pc} $.
55 %All the ends of $\pc$ in the sources.
57 %\item[ $ \tipzc, \tipcc, \tipuc, \tipfc $ ]
58 %The git ref for the Topbloke commit set $\pc$: respectively,
59 %the original, current, updated, and final values.
63 \stdsection{Inputs to the update algorithm}
67 \desclabelstyle{\nextlinelabel}
70 The topmost patch which we are trying to update. This and
71 all of its dependencies will be updated.
73 \item[ $h : \pc^{+/-} \mapsto \set H_{\pc^{+/-}}$ ]
74 Function for getting the existing heads $\set H$ of the branch $\pc^{+/-}$.
75 This will include the current local and remote git refs, as desired.
77 \item[ $g : \pc, \Gamma \mapsto \Gamma'$ ]
78 Function to allow explicit adjustment of the direct dependencies
79 of $\pc$. It is provided with a putative set of direct dependencies
80 $\Gamma$ computed as an appropriate merge of the dependencies requested by the
81 sources and should return the complete actual set $\Gamma'$ of direct
82 dependencies to use. This allows the specification of any desired
83 (acyclic) relation $\hasdirdep$.
87 \section{Ranking phase}
89 We run the following algorithm:
91 \item Set $\allpatches = \{ \}$.
94 \item Clear out the graph $\hasdirdep$ so it has no edges.
95 \item Execute {\bf Rank-Recurse}($\pc_0$)
96 \item Until $\allpatches$ remains unchanged.
100 {\bf Rank-Recurse}($\pc$) is:
103 \item If we have already done {\bf Rank-Recurse}($\pc$) in this
104 ranking iteration, do nothing. Otherwise:
106 \item Add $\pc$ to $\allpatches$ if it is not there already.
112 \bigcup_{\p \in \allpatches}
113 \bigcup_{H \in h(\pn) \lor H \in h(\py)}
114 \{ \baseof{E} \; | \; E \in \pendsof{H}{\pcy} \}
119 \item While $\exists_{S \in \set S} S \ge W$,
120 update $W \assign S$ and $\set S \assign \set S \, \backslash \{ S \}$
122 (This will often remove $W$ from $\set S$. Afterwards, $\set S$
123 is a collection of heads to be merged into $W$.)
125 \item Choose an order of $\set S$, $S_i$ for $i=1 \ldots n$.
127 \item For each $S_i$ in turn, choose a corresponding $M_i$
129 M_i \le S_i \land \left[
130 M_i \le W \lor \bigexists_{S_i, j<i} M_i \le s_i
134 \item Set $\Gamma = \depsreqof{W}$.
136 If there are multiple candidates we prefer $M_i \in \pcn$
139 \item For each $i \ldots 1..n$, update our putative direct
142 \Gamma \assign \text{\bf set-merge}\left(\Gamma,
144 M_i \in \pcn : & \depsreqof{M_i} \\
145 M_i \not\in \pcn : & \{ \}
151 \item Finalise our putative direct dependencies
153 \Gamma \assign g(\pc, \Gamma)
156 \item For each direct dependency $\pd \in \Gamma$,
159 \item Add an edge $\pc \hasdirdep \pd$ to the digraph (adding nodes
161 If this results in a cycle, abort entirely (as the function $g$ is
162 inappropriate; a different $g$ could work.)
164 \item Run ${\text{\bf Rank-Recurse}}(\pd)$.
168 The results of the ranking phase are:
170 $ \allpatches, \hasdirdep $ and hence the completion of $\hasdirdep$
171 into the partial order $\hasdep$.
173 For each $\pc$, the base branch starting point commit $W_{\pcn} = W$,
174 the direct dependencies $\Gamma_{\pc}$,
175 the ordered set of base branch sources $\set S_{\pcn} = \set S,
177 and corresponding merge bases $M_{\pcn,i} = M_i$.
181 \section{Planning phase}
183 The results of the planning phase consist of:
185 \item{ The relation $\hasdirdep$ and hence the partial order $\hasdep$. }
186 \item{ For each commit set $\pc$, a confirmed set of sources $\set S_{\pc}$. }
187 \item{ For each commit set $\pc$, the order in which to merge the sources
188 $E_{\pc,j} \in \set E_{\pc}$. }
189 \item{ For each $E_{\pc,j}$ an intended merge base $M_{\pc,j}$. }
192 We use a recursive planning algorith, recursing over Topbloke commit
193 sets (ie, sets $\py$ or $\pn$). We'll call the commit set we're
194 processing at each step $\pc$.
195 At each recursive step
196 we make a plan to merge all $\set E_{\pc} = \{ E_{\pc,j \ldots} \}$
197 and all the direct contributors of $\pc$ (as determined below)
198 into $\tipzc$, to make $\tipfc$.
200 We start with $\pc = \pl$ where $\pl = \patchof{L}$.
203 \subsection{Direct contributors for $\pc = \pcn$}
205 The direct contributors of $\pcn$ are the commit sets corresponding to
206 the tip branches for the direct dependencies of the patch $\pc$. We
207 need to calculate what the direct dependencies are going to be.
209 Choose an (arbitrary, but ideally somehow optimal in
210 a way not discussed here) ordering of $\set E_{\pc}$, $E_{\pc,j}$
212 For brevity we will write $E_j$ for $E_{\pc,j}$.
213 Remove from that set (and ordering) any $E_j$ which
214 are $\le$ and $\neq$ some other $E_k$.
216 Initially let $\set D_0 = \depsreqof{\tipzc}$.
217 For each $E_j$ starting with $j=1$ choose a corresponding intended
218 merge base $M_j$ such that $M_j \le E_j \land M_j \le T_{\pc,j-1}$.
219 Calculate $\set D_j$ as the 3-way merge of the sets $\set D_{j-1}$ and
220 $\depsreqof{E_j}$ using as a base $\depsreqof{M_j}$. This will
221 generate $D_m$ as the putative direct contributors of $\pcn$.
223 However, the invocation may give instructions that certain direct
224 dependencies are definitely to be included, or excluded. As a result
225 the set of actual direct contributors is some arbitrary set of patches
226 (strictly, some arbitrary set of Topbloke tip commit sets).
228 \subsection{Direct contributors for $\pc = \pcy$}
230 The sole direct contributor of $\pcy$ is $\pcn$.
232 \subsection{Recursive step}
234 For each direct contributor $\p$, we add the edge $\pc \hasdirdep \p$
235 and augment the ordering $\hasdep$ accordingly.
237 If this would make a cycle in $\hasdep$, we abort . The operation must
238 then be retried by the user, if desired, but with different or
239 additional instructions for modifying the direct contributors of some
240 $\pqn$ involved in the cycle.
242 For each such $\p$, after updating $\hasdep$, we recursively make a plan
247 \section{Execution phase}
249 We process commit sets from the bottom up according to the relation
250 $\hasdep$. For each commit set $\pc$ we construct $\tipfc$ from
251 $\tipzc$, as planned. By construction, $\hasdep$ has $\patchof{L}$
252 as its maximum, so this operation will finish by updating
253 $\tipca{\patchof{L}}$ with $\tipfa{\patchof{L}}$.
255 After we are done with each commit set $\pc$, the
256 new tip $\tipfc$ has the following properties:
257 \[ \eqn{Tip Sources}{
258 \bigforall_{E_i \in \set E_{\pc}} \tipfc \ge E_i
260 \[ \eqn{Tip Dependencies}{
261 \bigforall_{\pc \hasdep \p} \tipfc \ge \tipfa \p
263 \[ \eqn{Perfect Contents}{
264 \tipfc \haspatch \p \equiv \pc \hasdep \py
267 For brevity we will sometimes write $\tipu$ for $\tipuc$, etc. We will start
268 out with $\tipc = \tipz$, and at each step of the way construct some
269 $\tipu$ from $\tipc$. The final $\tipu$ becomes $\tipf$.
271 \subsection{Preparation}
273 Firstly, we will check each $E_i$ for being $\ge \tipc$. If
274 it is, are we fast forward to $E_i$
275 --- formally, $\tipu = \text{max}(\tipc, E_i)$ ---
276 and drop $E_i$ from the planned ordering.
278 Then we will merge the direct contributors and the sources' ends.
279 This generates more commits $\tipuc \in \pc$, but none in any other
280 commit set. We maintain
282 \bigforall_{\p \isdep \pc}
283 \pancsof{\tipcc}{\p} \subset
284 \pancsof{\tipfa \p}{\p}
287 For $\tipcc = \tipzc$, $T$ ...WRONG WE NEED $\tipfa \p$ TO BE IN $\set E$ SOMEHOW
290 \subsection{Merge Contributors for $\pcy$}
292 Merge $\pcn$ into $\tipc$. That is, merge with
293 $L = \tipc, R = \tipfa{\pcn}, M = \baseof{\tipc}$.
294 to construct $\tipu$.
298 Ingredients satisfied by construction.
299 Tip Merge satisfied by construction. Merge Acyclic follows
300 from Perfect Contents and $\hasdep$ being acyclic.
302 Removal Merge Ends: For $\p = \pc$, $M \nothaspatch \p$; OK.
303 For $\p \neq \pc$, by Tip Contents,
304 $M \haspatch \p \equiv L \haspatch \p$, so we need only
305 worry about $X = R, Y = L$; ie $L \haspatch \p$,
306 $M = \baseof{L} \haspatch \p$.
307 By Tip Contents for $L$, $D \le L \equiv D \le M$. OK.~~$\qed$
311 Addition Merge Ends: If $\py \isdep \pcn$, we have already
312 done the execution phase for $\pcn$ and $\py$. By
313 Perfect Contents for $\pcn$, $\tipfa \pcn \haspatch \p$ i.e.
314 $R \haspatch \p$. So we only need to worry about $Y = R = \tipfa \pcn$.
315 By Tip Dependencies $\tipfa \pcn \ge \tipfa \py$.
316 And by Tip Sources $\tipfa \py \ge $
318 want to prove $E \le \tipfc$ where $E \in \pendsof{\tipcc}{\py}$
320 $\pancsof{\tipcc}{\py} = $
323 computed $\tipfa \py$, and by Perfect Contents for $\py$
326 with $M=M_j, L=T_{\pc,j-1}, R=E_j$,
327 and calculate what the resulting desired direct dependencies file
328 (ie, the set of patches $\set D_j$)
329 would be. Eventually we
331 So, formally, we select somehow an order of sources $S_i$. For each
334 Make use of the following recursive algorithm, Plan
339 recursively make a plan to merge all $E = \pends$