hgbook
view en/mq-collab.tex @ 106:9cbc5d0db542
Finish off advanced MQ chapter (maybe).
| author | Bryan O'Sullivan <bos@serpentine.com> |
|---|---|
| date | Mon Oct 23 15:43:04 2006 -0700 (2006-10-23) |
| parents | ecacb6b4c9fd |
| children | a0d7e11db169 |
line source
1 \chapter{Advanced uses of Mercurial Queues}
3 While it's easy to pick up straightforward uses of Mercurial Queues,
4 use of a little discipline and some of MQ's less frequently used
5 capabilities makes it possible to work in complicated development
6 environments.
8 In this chapter, I will use as an example a technique I have used to
9 manage the development of an Infiniband device driver for the Linux
10 kernel. The driver in question is large (at least as drivers go),
11 with 25,000 lines of code spread across 35 source files. It is
12 maintained by a small team of developers.
14 While much of the material in this chapter is specific to Linux, the
15 same principles apply to any code base for which you're not the
16 primary owner, and upon which you need to do a lot of development.
18 \section{The problem of many targets}
20 The Linux kernel changes rapidly, and has never been internally
21 stable; developers frequently make drastic changes between releases.
22 This means that a version of the driver that works well with a
23 particular released version of the kernel will not even \emph{compile}
24 correctly against, typically, any other version.
26 To maintain a driver, we have to keep a number of distinct versions of
27 Linux in mind.
28 \begin{itemize}
29 \item One target is the main Linux kernel development tree.
30 Maintenance of the code is in this case partly shared by other
31 developers in the kernel community, who make ``drive-by''
32 modifications to the driver as they develop and refine kernel
33 subsystems.
34 \item We also maintain a number of ``backports'' to older versions of
35 the Linux kernel, to support the needs of customers who are running
36 older Linux distributions that do not incorporate our drivers. (To
37 \emph{backport} a piece of code is to modify it to work in an older
38 version of its target environment than the version it was developed
39 for.)
40 \item Finally, we make software releases on a schedule that is
41 necessarily not aligned with those used by Linux distributors and
42 kernel developers, so that we can deliver new features to customers
43 without forcing them to upgrade their entire kernels or
44 distributions.
45 \end{itemize}
47 \subsection{Tempting approaches that don't work well}
49 There are two ``standard'' ways to maintain a piece of software that
50 has to target many different environments.
52 The first is to maintain a number of branches, each intended for a
53 single target. The trouble with this approach is that you must
54 maintain iron discipline in the flow of changes between repositories.
55 A new feature or bug fix must start life in a ``pristine'' repository,
56 then percolate out to every backport repository. Backport changes are
57 more limited in the branches they should propagate to; a backport
58 change that is applied to a branch where it doesn't belong will
59 probably stop the driver from compiling.
61 The second is to maintain a single source tree filled with conditional
62 statements that turn chunks of code on or off depending on the
63 intended target. Because these ``ifdefs'' are not allowed in the
64 Linux kernel tree, a manual or automatic process must be followed to
65 strip them out and yield a clean tree. A code base maintained in this
66 fashion rapidly becomes a rat's nest of conditional blocks that are
67 difficult to understand and maintain.
69 Neither of these approaches is well suited to a situation where you
70 don't ``own'' the canonical copy of a source tree. In the case of a
71 Linux driver that is distributed with the standard kernel, Linus's
72 tree contains the copy of the code that will be treated by the world
73 as canonical. The upstream version of ``my'' driver can be modified
74 by people I don't know, without me even finding out about it until
75 after the changes show up in Linus's tree.
77 These approaches have the added weakness of making it difficult to
78 generate well-formed patches to submit upstream.
80 In principle, Mercurial Queues seems like a good candidate to manage a
81 development scenario such as the above. While this is indeed the
82 case, MQ contains a few added features that make the job more
83 pleasant.
85 \section{Conditionally applying patches with
86 guards}
88 Perhaps the best way to maintain sanity with so many targets is to be
89 able to choose specific patches to apply for a given situation. MQ
90 provides a feature called ``guards'' (which originates with quilt's
91 \texttt{guards} command) that does just this. To start off, let's
92 create a simple repository for experimenting in.
93 \interaction{mq.guards.init}
94 This gives us a tiny repository that contains two patches that don't
95 have any dependencies on each other, because they touch different files.
97 The idea behind conditional application is that you can ``tag'' a
98 patch with a \emph{guard}, which is simply a text string of your
99 choosing, then tell MQ to select specific guards to use when applying
100 patches. MQ will then either apply, or skip over, a guarded patch,
101 depending on the guards that you have selected.
103 A patch can have an arbitrary number of guards;
104 each one is \emph{positive} (``apply this patch if this guard is
105 selected'') or \emph{negative} (``skip this patch if this guard is
106 selected''). A patch with no guards is always applied.
108 \section{Controlling the guards on a patch}
110 The \hgcmd{qguard} command lets you determine which guards should
111 apply to a patch, or display the guards that are already in effect.
112 Without any arguments, it displays the guards on the current topmost
113 patch.
114 \interaction{mq.guards.qguard}
115 To set a positive guard on a patch, prefix the name of the guard with
116 a ``\texttt{+}''.
117 \interaction{mq.guards.qguard.pos}
118 To set a negative guard on a patch, prefix the name of the guard with
119 a ``\texttt{-}''.
120 \interaction{mq.guards.qguard.neg}
122 \begin{note}
123 The \hgcmd{qguard} command \emph{sets} the guards on a patch; it
124 doesn't \emph{modify} them. What this means is that if you run
125 \hgcmdargs{qguard}{+a +b} on a patch, then \hgcmdargs{qguard}{+c} on
126 the same patch, the \emph{only} guard that will be set on it
127 afterwards is \texttt{+c}.
128 \end{note}
130 Mercurial stores guards in the \sfilename{series} file; the form in
131 which they are stored is easy both to understand and to edit by hand.
132 (In other words, you don't have to use the \hgcmd{qguard} command if
133 you don't want to; it's okay to simply edit the \sfilename{series}
134 file.)
135 \interaction{mq.guards.series}
137 \section{Selecting the guards to use}
139 The \hgcmd{qselect} command determines which guards are active at a
140 given time. The effect of this is to determine which patches MQ will
141 apply the next time you run \hgcmd{qpush}. It has no other effect; in
142 particular, it doesn't do anything to patches that are already
143 applied.
145 With no arguments, the \hgcmd{qselect} command lists the guards
146 currently in effect, one per line of output. Each argument is treated
147 as the name of a guard to apply.
148 \interaction{mq.guards.qselect.foo}
149 In case you're interested, the currently selected guards are stored in
150 the \sfilename{guards} file.
151 \interaction{mq.guards.qselect.cat}
152 We can see the effect the selected guards have when we run
153 \hgcmd{qpush}.
154 \interaction{mq.guards.qselect.qpush}
156 A guard cannot start with a ``\texttt{+}'' or ``\texttt{-}''
157 character. The name of a guard must not contain white space, but most
158 othter characters are acceptable. If you try to use a guard with an
159 invalid name, MQ will complain:
160 \interaction{mq.guards.qselect.error}
161 Changing the selected guards changes the patches that are applied.
162 \interaction{mq.guards.qselect.quux}
163 You can see in the example below that negative guards take precedence
164 over positive guards.
165 \interaction{mq.guards.qselect.foobar}
167 \section{MQ's rules for applying patches}
169 The rules that MQ uses when deciding whether to apply a patch
170 are as follows.
171 \begin{itemize}
172 \item A patch that has no guards is always applied.
173 \item If the patch has any negative guard that matches any currently
174 selected guard, the patch is skipped.
175 \item If the patch has any positive guard that matches any currently
176 selected guard, the patch is applied.
177 \item If the patch has positive or negative guards, but none matches
178 any currently selected guard, the patch is skipped.
179 \end{itemize}
181 \section{Trimming the work environment}
183 In working on the device driver I mentioned earlier, I don't apply the
184 patches to a normal Linux kernel tree. Instead, I use a repository
185 that contains only a snapshot of the source files and headers that are
186 relevant to Infiniband development. This repository is~1\% the size
187 of a kernel repository, so it's easier to work with.
189 I then choose a ``base'' version on top of which the patches are
190 applied. This is a snapshot of the Linux kernel tree as of a revision
191 of my choosing. When I take the snapshot, I record the changeset ID
192 from the kernel repository in the commit message. Since the snapshot
193 preserves the ``shape'' and content of the relevant parts of the
194 kernel tree, I can apply my patches on top of either my tiny
195 repository or a normal kernel tree.
197 Normally, the base tree atop which the patches apply should be a
198 snapshot of a very recent upstream tree. This best facilitates the
199 development of patches that can easily be submitted upstream with few
200 or no modifications.
202 \section{Dividing up the \sfilename{series} file}
204 I categorise the patches in the \sfilename{series} file into a number
205 of logical groups. Each section of like patches begins with a block
206 of comments that describes the purpose of the patches that follow.
208 The sequence of patch groups that I maintain follows. The ordering of
209 these groups is important; I'll describe why after I introduce the
210 groups.
211 \begin{itemize}
212 \item The ``accepted'' group. Patches that the development team has
213 submitted to the maintainer of the Infiniband subsystem, and which
214 he has accepted, but which are not present in the snapshot that the
215 tiny repository is based on. These are ``read only'' patches,
216 present only to transform the tree into a similar state as it is in
217 the upstream maintainer's repository.
218 \item The ``rework'' group. Patches that I have submitted, but that
219 the upstream maintainer has requested modifications to before he
220 will accept them.
221 \item The ``pending'' group. Patches that I have not yet submitted to
222 the upstream maintainer, but which we have finished working on.
223 These will be ``read only'' for a while. If the upstream maintainer
224 accepts them upon submission, I'll move them to the end of the
225 ``accepted'' group. If he requests that I modify any, I'll move
226 them to the beginning of the ``rework'' group.
227 \item The ``in progress'' group. Patches that are actively being
228 developed, and should not be submitted anywhere yet.
229 \item The ``backport'' group. Patches that adapt the source tree to
230 older versions of the kernel tree.
231 \item The ``do not ship'' group. Patches that for some reason should
232 never be submitted upstream. For example, one such patch might
233 change embedded driver identification strings to make it easier to
234 distinguish, in the field, between an out-of-tree version of the
235 driver and a version shipped by a distribution vendor.
236 \end{itemize}
238 Now to return to the reasons for ordering groups of patches in this
239 way. We would like the lowest patches in the stack to be as stable as
240 possible, so that we will not need to rework higher patches due to
241 changes in context. Putting patches that will never be changed first
242 in the \sfilename{series} file serves this purpose.
244 We would also like the patches that we know we'll need to modify to be
245 applied on top of a source tree that resembles the upstream tree as
246 closely as possible. This is why we keep accepted patches around for
247 a while.
249 The ``backport'' and ``do not ship'' patches float at the end of the
250 \sfilename{series} file. The backport patches must be applied on top
251 of all other patches, and the ``do not ship'' patches might as well
252 stay out of harm's way.
254 \section{Maintaining the patch series}
256 In my work, I use a number of guards to control which patches are to
257 be applied.
259 \begin{itemize}
260 \item ``Accepted'' patches are guarded with \texttt{accepted}. I
261 enable this guard most of the time. When I'm applying the patches
262 on top of a tree where the patches are already present, I can turn
263 this patch off, and the paptches that follow it will apply cleanly.
264 \item Patches that are ``finished'', but not yet submitted, have no
265 guards. If I'm applying the patch stack to a copy of the upstream
266 tree, I don't need to enable any guards in order to get a reasonably
267 safe source tree.
268 \item Those patches that need reworking before being resubmitted are
269 guarded with \texttt{rework}.
270 \item For those patches that are still under development, I use
271 \texttt{devel}.
272 \item A backport patch may have several guards, one for each version
273 of the kernel to which it applies. For example, a patch that
274 backports a piece of code to~2.6.9 will have a~\texttt{2.6.9} guard.
275 \end{itemize}
276 This variety of guards gives me considerable flexibility in
277 qdetermining what kind of source tree I want to end up with. For most
278 situations, the selection of appropriate guards is automated during
279 the build process, but I can manually tune the guards to use for less
280 common circumstances.
282 \subsection{The art of writing backport patches}
284 Using MQ, writing a backport patch is a simple process. All such a
285 patch has to do is modify a piece of code that uses a kernel feature
286 not present in the older version of the kernel, so that the driver
287 continues to work correctly under that older version.
289 A useful goal when writing a good backport patch is to make your code
290 look as if it was written for the older version of the kernel you're
291 targeting. The less obtrusive the patch, the easier it will be to
292 understand and maintain. If you're writing a collection of backport
293 patches to avoid the ``rat's nest'' effect of lots of
294 \texttt{\#ifdef}s (hunks of source code that are only used
295 conditionally) in your code, don't introduce version-dependent
296 \texttt{\#ifdef}s into the patches. Instead, write several patches,
297 each of which makes unconditional changes, and control their
298 application using guards.
300 There are two reasons to divide backport patches into a distinct
301 group, away from the ``regular'' patches whose effects they modify.
302 The first is that intermingling the two makes it more difficult to use
303 a tool like the \hgext{patchbomb} extension to automate the process of
304 submitting the patches to an upstream maintainer. The second is that
305 a backport patch could perturb the context in which a subsequent
306 regular patch is applied, making it impossible to apply the regular
307 patch cleanly \emph{without} the earlier backport patch already being
308 applied.
310 \section{Useful tips for developing with MQ}
312 \subsection{Organising patches in directories}
314 If you're working on a substantial project with MQ, it's not difficult
315 to accumulate a large number of patches. For example, I have one
316 patch repository that contains over 250 patches.
318 If you can group these patches into separate logical categories, you
319 can if you like store them in different directories; MQ has no
320 problems with patch names that contain path separators.
322 \subsection{Viewing the history of a patch}
323 \label{mq-collab:tips:interdiff}
325 If you're developing a set of patches over a long time, it's a good
326 idea to maintain them in a repository, as discussed in
327 section~\ref{sec:mq:repo}. If you do so, you'll quickly discover that
328 using the \hgcmd{diff} command to look at the history of changes to a
329 patch is unworkable. This is in part because you're looking at the
330 second derivative of the real code (a diff of a diff), but also
331 because MQ adds noise to the process by modifying time stamps and
332 directory names when it updates a patch.
334 However, you can use the \hgext{extdiff} extension, which is bundled
335 with Mercurial, to turn a diff of two versions of a patch into
336 something readable. To do this, you will need a third-party package
337 called \package{patchutils}~\cite{web:patchutils}. This provides a
338 command named \command{interdiff}, which shows the differences between
339 two diffs as a diff. Used on two versions of the same diff, it
340 generates a diff that represents the diff from the first to the second
341 version.
343 You can enable the \hgext{extdiff} extension in the usual way, by
344 adding a line to the \rcsection{extensions} section of your \hgrc.
345 \begin{codesample2}
346 [extensions]
347 extdiff =
348 \end{codesample2}
349 The \command{interdiff} command expects to be passed the names of two
350 files, but the \hgext{extdiff} extension passes the program it runs a
351 pair of directories, each of which can contain an arbitrary number of
352 files. We thus need a small program that will run \command{interdiff}
353 on each pair of files in these two directories. This program is
354 available as \sfilename{hg-interdiff} in the \dirname{examples}
355 directory of the source code repository that accompanies this book.
356 \excode{hg-interdiff}
358 With the \sfilename{hg-interdiff} program in your shell's search path,
359 you can run it as follows, from inside an MQ patch directory:
360 \begin{codesample2}
361 hg extdiff -p hg-interdiff -r A:B my-change.patch
362 \end{codesample2}
363 Since you'll probably want to use this long-winded command a lot, you
364 can get \hgext{hgext} to make it available as a normal Mercurial
365 command, again by editing your \hgrc.
366 \begin{codesample2}
367 [extdiff]
368 cmd.interdiff = hg-interdiff
369 \end{codesample2}
370 This directs \hgext{hgext} to make an \texttt{interdiff} command
371 available, so you can now shorten the previous invocation of
372 \hgcmd{extdiff} to something a little more wieldy.
373 \begin{codesample2}
374 hg interdiff -r A:B my-change.patch
375 \end{codesample2}
377 %%% Local Variables:
378 %%% mode: latex
379 %%% TeX-master: "00book"
380 %%% End: