hgbook
view en/ch13-mq-collab.xml @ 572:13513d2a128d
Add sensible names to chapters.
| author | Bryan O'Sullivan <bos@serpentine.com> |
|---|---|
| date | Mon Mar 09 23:37:29 2009 -0700 (2009-03-09) |
| parents | 8fcd44708f41 |
| children | 8366882f67f2 cfdb601a3c8b |
line source
1 <!-- vim: set filetype=docbkxml shiftwidth=2 autoindent expandtab tw=77 : -->
3 <chapter id="chap:mq-collab">
4 <?dbhtml filename="advanced-uses-of-mercurial-queues.html"?>
5 <title>Advanced uses of Mercurial Queues</title>
7 <para>While it's easy to pick up straightforward uses of Mercurial
8 Queues, use of a little discipline and some of MQ's less
9 frequently used capabilities makes it possible to work in
10 complicated development environments.</para>
12 <para>In this chapter, I will use as an example a technique I have
13 used to manage the development of an Infiniband device driver for
14 the Linux kernel. The driver in question is large (at least as
15 drivers go), with 25,000 lines of code spread across 35 source
16 files. It is maintained by a small team of developers.</para>
18 <para>While much of the material in this chapter is specific to
19 Linux, the same principles apply to any code base for which you're
20 not the primary owner, and upon which you need to do a lot of
21 development.</para>
23 <sect1>
24 <title>The problem of many targets</title>
26 <para>The Linux kernel changes rapidly, and has never been
27 internally stable; developers frequently make drastic changes
28 between releases. This means that a version of the driver that
29 works well with a particular released version of the kernel will
30 not even <emphasis>compile</emphasis> correctly against,
31 typically, any other version.</para>
33 <para>To maintain a driver, we have to keep a number of distinct
34 versions of Linux in mind.</para>
35 <itemizedlist>
36 <listitem><para>One target is the main Linux kernel development
37 tree. Maintenance of the code is in this case partly shared
38 by other developers in the kernel community, who make
39 <quote>drive-by</quote> modifications to the driver as they
40 develop and refine kernel subsystems.</para>
41 </listitem>
42 <listitem><para>We also maintain a number of
43 <quote>backports</quote> to older versions of the Linux
44 kernel, to support the needs of customers who are running
45 older Linux distributions that do not incorporate our
46 drivers. (To <emphasis>backport</emphasis> a piece of code
47 is to modify it to work in an older version of its target
48 environment than the version it was developed for.)</para>
49 </listitem>
50 <listitem><para>Finally, we make software releases on a schedule
51 that is necessarily not aligned with those used by Linux
52 distributors and kernel developers, so that we can deliver
53 new features to customers without forcing them to upgrade
54 their entire kernels or distributions.</para>
55 </listitem></itemizedlist>
57 <sect2>
58 <title>Tempting approaches that don't work well</title>
60 <para>There are two <quote>standard</quote> ways to maintain a
61 piece of software that has to target many different
62 environments.</para>
64 <para>The first is to maintain a number of branches, each
65 intended for a single target. The trouble with this approach
66 is that you must maintain iron discipline in the flow of
67 changes between repositories. A new feature or bug fix must
68 start life in a <quote>pristine</quote> repository, then
69 percolate out to every backport repository. Backport changes
70 are more limited in the branches they should propagate to; a
71 backport change that is applied to a branch where it doesn't
72 belong will probably stop the driver from compiling.</para>
74 <para>The second is to maintain a single source tree filled with
75 conditional statements that turn chunks of code on or off
76 depending on the intended target. Because these
77 <quote>ifdefs</quote> are not allowed in the Linux kernel
78 tree, a manual or automatic process must be followed to strip
79 them out and yield a clean tree. A code base maintained in
80 this fashion rapidly becomes a rat's nest of conditional
81 blocks that are difficult to understand and maintain.</para>
83 <para>Neither of these approaches is well suited to a situation
84 where you don't <quote>own</quote> the canonical copy of a
85 source tree. In the case of a Linux driver that is
86 distributed with the standard kernel, Linus's tree contains
87 the copy of the code that will be treated by the world as
88 canonical. The upstream version of <quote>my</quote> driver
89 can be modified by people I don't know, without me even
90 finding out about it until after the changes show up in
91 Linus's tree.</para>
93 <para>These approaches have the added weakness of making it
94 difficult to generate well-formed patches to submit
95 upstream.</para>
97 <para>In principle, Mercurial Queues seems like a good candidate
98 to manage a development scenario such as the above. While
99 this is indeed the case, MQ contains a few added features that
100 make the job more pleasant.</para>
102 </sect2>
103 </sect1>
104 <sect1>
105 <title>Conditionally applying patches with guards</title>
107 <para>Perhaps the best way to maintain sanity with so many targets
108 is to be able to choose specific patches to apply for a given
109 situation. MQ provides a feature called <quote>guards</quote>
110 (which originates with quilt's <literal>guards</literal>
111 command) that does just this. To start off, let's create a
112 simple repository for experimenting in.</para>
114 &interaction.mq.guards.init;
116 <para>This gives us a tiny repository that contains two patches
117 that don't have any dependencies on each other, because they
118 touch different files.</para>
120 <para>The idea behind conditional application is that you can
121 <quote>tag</quote> a patch with a <emphasis>guard</emphasis>,
122 which is simply a text string of your choosing, then tell MQ to
123 select specific guards to use when applying patches. MQ will
124 then either apply, or skip over, a guarded patch, depending on
125 the guards that you have selected.</para>
127 <para>A patch can have an arbitrary number of guards; each one is
128 <emphasis>positive</emphasis> (<quote>apply this patch if this
129 guard is selected</quote>) or <emphasis>negative</emphasis>
130 (<quote>skip this patch if this guard is selected</quote>). A
131 patch with no guards is always applied.</para>
133 </sect1>
134 <sect1>
135 <title>Controlling the guards on a patch</title>
137 <para>The <command role="hg-ext-mq">qguard</command> command lets
138 you determine which guards should apply to a patch, or display
139 the guards that are already in effect. Without any arguments, it
140 displays the guards on the current topmost patch.</para>
142 &interaction.mq.guards.qguard;
144 <para>To set a positive guard on a patch, prefix the name of the
145 guard with a <quote><literal>+</literal></quote>.</para>
147 &interaction.mq.guards.qguard.pos;
149 <para>To set a negative guard
150 on a patch, prefix the name of the guard with a
151 <quote><literal>-</literal></quote>.</para>
153 &interaction.mq.guards.qguard.neg;
155 <note>
156 <para> The <command role="hg-ext-mq">qguard</command> command
157 <emphasis>sets</emphasis> the guards on a patch; it doesn't
158 <emphasis>modify</emphasis> them. What this means is that if
159 you run <command role="hg-cmd">hg qguard +a +b</command> on a
160 patch, then <command role="hg-cmd">hg qguard +c</command> on
161 the same patch, the <emphasis>only</emphasis> guard that will
162 be set on it afterwards is <literal>+c</literal>.</para>
163 </note>
165 <para>Mercurial stores guards in the <filename
166 role="special">series</filename> file; the form in which they
167 are stored is easy both to understand and to edit by hand. (In
168 other words, you don't have to use the <command
169 role="hg-ext-mq">qguard</command> command if you don't want
170 to; it's okay to simply edit the <filename
171 role="special">series</filename> file.)</para>
173 &interaction.mq.guards.series;
175 </sect1>
176 <sect1>
177 <title>Selecting the guards to use</title>
179 <para>The <command role="hg-ext-mq">qselect</command> command
180 determines which guards are active at a given time. The effect
181 of this is to determine which patches MQ will apply the next
182 time you run <command role="hg-ext-mq">qpush</command>. It has
183 no other effect; in particular, it doesn't do anything to
184 patches that are already applied.</para>
186 <para>With no arguments, the <command
187 role="hg-ext-mq">qselect</command> command lists the guards
188 currently in effect, one per line of output. Each argument is
189 treated as the name of a guard to apply.</para>
191 &interaction.mq.guards.qselect.foo;
193 <para>In case you're interested, the currently selected guards are
194 stored in the <filename role="special">guards</filename> file.</para>
196 &interaction.mq.guards.qselect.cat;
198 <para>We can see the effect the selected guards have when we run
199 <command role="hg-ext-mq">qpush</command>.</para>
201 &interaction.mq.guards.qselect.qpush;
203 <para>A guard cannot start with a
204 <quote><literal>+</literal></quote> or
205 <quote><literal>-</literal></quote> character. The name of a
206 guard must not contain white space, but most other characters
207 are acceptable. If you try to use a guard with an invalid name,
208 MQ will complain:</para>
210 &interaction.mq.guards.qselect.error;
212 <para>Changing the selected guards changes the patches that are
213 applied.</para>
215 &interaction.mq.guards.qselect.quux;
217 <para>You can see in the example below that negative guards take
218 precedence over positive guards.</para>
220 &interaction.mq.guards.qselect.foobar;
222 </sect1>
223 <sect1>
224 <title>MQ's rules for applying patches</title>
226 <para>The rules that MQ uses when deciding whether to apply a
227 patch are as follows.</para>
228 <itemizedlist>
229 <listitem><para>A patch that has no guards is always
230 applied.</para>
231 </listitem>
232 <listitem><para>If the patch has any negative guard that matches
233 any currently selected guard, the patch is skipped.</para>
234 </listitem>
235 <listitem><para>If the patch has any positive guard that matches
236 any currently selected guard, the patch is applied.</para>
237 </listitem>
238 <listitem><para>If the patch has positive or negative guards,
239 but none matches any currently selected guard, the patch is
240 skipped.</para>
241 </listitem></itemizedlist>
243 </sect1>
244 <sect1>
245 <title>Trimming the work environment</title>
247 <para>In working on the device driver I mentioned earlier, I don't
248 apply the patches to a normal Linux kernel tree. Instead, I use
249 a repository that contains only a snapshot of the source files
250 and headers that are relevant to Infiniband development. This
251 repository is 1% the size of a kernel repository, so it's easier
252 to work with.</para>
254 <para>I then choose a <quote>base</quote> version on top of which
255 the patches are applied. This is a snapshot of the Linux kernel
256 tree as of a revision of my choosing. When I take the snapshot,
257 I record the changeset ID from the kernel repository in the
258 commit message. Since the snapshot preserves the
259 <quote>shape</quote> and content of the relevant parts of the
260 kernel tree, I can apply my patches on top of either my tiny
261 repository or a normal kernel tree.</para>
263 <para>Normally, the base tree atop which the patches apply should
264 be a snapshot of a very recent upstream tree. This best
265 facilitates the development of patches that can easily be
266 submitted upstream with few or no modifications.</para>
268 </sect1>
269 <sect1>
270 <title>Dividing up the <filename role="special">series</filename>
271 file</title>
273 <para>I categorise the patches in the <filename
274 role="special">series</filename> file into a number of logical
275 groups. Each section of like patches begins with a block of
276 comments that describes the purpose of the patches that
277 follow.</para>
279 <para>The sequence of patch groups that I maintain follows. The
280 ordering of these groups is important; I'll describe why after I
281 introduce the groups.</para>
282 <itemizedlist>
283 <listitem><para>The <quote>accepted</quote> group. Patches that
284 the development team has submitted to the maintainer of the
285 Infiniband subsystem, and which he has accepted, but which
286 are not present in the snapshot that the tiny repository is
287 based on. These are <quote>read only</quote> patches,
288 present only to transform the tree into a similar state as
289 it is in the upstream maintainer's repository.</para>
290 </listitem>
291 <listitem><para>The <quote>rework</quote> group. Patches that I
292 have submitted, but that the upstream maintainer has
293 requested modifications to before he will accept
294 them.</para>
295 </listitem>
296 <listitem><para>The <quote>pending</quote> group. Patches that
297 I have not yet submitted to the upstream maintainer, but
298 which we have finished working on. These will be <quote>read
299 only</quote> for a while. If the upstream maintainer
300 accepts them upon submission, I'll move them to the end of
301 the <quote>accepted</quote> group. If he requests that I
302 modify any, I'll move them to the beginning of the
303 <quote>rework</quote> group.</para>
304 </listitem>
305 <listitem><para>The <quote>in progress</quote> group. Patches
306 that are actively being developed, and should not be
307 submitted anywhere yet.</para>
308 </listitem>
309 <listitem><para>The <quote>backport</quote> group. Patches that
310 adapt the source tree to older versions of the kernel
311 tree.</para>
312 </listitem>
313 <listitem><para>The <quote>do not ship</quote> group. Patches
314 that for some reason should never be submitted upstream.
315 For example, one such patch might change embedded driver
316 identification strings to make it easier to distinguish, in
317 the field, between an out-of-tree version of the driver and
318 a version shipped by a distribution vendor.</para>
319 </listitem></itemizedlist>
321 <para>Now to return to the reasons for ordering groups of patches
322 in this way. We would like the lowest patches in the stack to
323 be as stable as possible, so that we will not need to rework
324 higher patches due to changes in context. Putting patches that
325 will never be changed first in the <filename
326 role="special">series</filename> file serves this
327 purpose.</para>
329 <para>We would also like the patches that we know we'll need to
330 modify to be applied on top of a source tree that resembles the
331 upstream tree as closely as possible. This is why we keep
332 accepted patches around for a while.</para>
334 <para>The <quote>backport</quote> and <quote>do not ship</quote>
335 patches float at the end of the <filename
336 role="special">series</filename> file. The backport patches
337 must be applied on top of all other patches, and the <quote>do
338 not ship</quote> patches might as well stay out of harm's
339 way.</para>
341 </sect1>
342 <sect1>
343 <title>Maintaining the patch series</title>
345 <para>In my work, I use a number of guards to control which
346 patches are to be applied.</para>
348 <itemizedlist>
349 <listitem><para><quote>Accepted</quote> patches are guarded with
350 <literal>accepted</literal>. I enable this guard most of
351 the time. When I'm applying the patches on top of a tree
352 where the patches are already present, I can turn this patch
353 off, and the patches that follow it will apply
354 cleanly.</para>
355 </listitem>
356 <listitem><para>Patches that are <quote>finished</quote>, but
357 not yet submitted, have no guards. If I'm applying the
358 patch stack to a copy of the upstream tree, I don't need to
359 enable any guards in order to get a reasonably safe source
360 tree.</para>
361 </listitem>
362 <listitem><para>Those patches that need reworking before being
363 resubmitted are guarded with
364 <literal>rework</literal>.</para>
365 </listitem>
366 <listitem><para>For those patches that are still under
367 development, I use <literal>devel</literal>.</para>
368 </listitem>
369 <listitem><para>A backport patch may have several guards, one
370 for each version of the kernel to which it applies. For
371 example, a patch that backports a piece of code to 2.6.9
372 will have a <literal>2.6.9</literal> guard.</para>
373 </listitem></itemizedlist>
374 <para>This variety of guards gives me considerable flexibility in
375 determining what kind of source tree I want to end up with. For
376 most situations, the selection of appropriate guards is
377 automated during the build process, but I can manually tune the
378 guards to use for less common circumstances.</para>
380 <sect2>
381 <title>The art of writing backport patches</title>
383 <para>Using MQ, writing a backport patch is a simple process.
384 All such a patch has to do is modify a piece of code that uses
385 a kernel feature not present in the older version of the
386 kernel, so that the driver continues to work correctly under
387 that older version.</para>
389 <para>A useful goal when writing a good backport patch is to
390 make your code look as if it was written for the older version
391 of the kernel you're targeting. The less obtrusive the patch,
392 the easier it will be to understand and maintain. If you're
393 writing a collection of backport patches to avoid the
394 <quote>rat's nest</quote> effect of lots of
395 <literal>#ifdef</literal>s (hunks of source code that are only
396 used conditionally) in your code, don't introduce
397 version-dependent <literal>#ifdef</literal>s into the patches.
398 Instead, write several patches, each of which makes
399 unconditional changes, and control their application using
400 guards.</para>
402 <para>There are two reasons to divide backport patches into a
403 distinct group, away from the <quote>regular</quote> patches
404 whose effects they modify. The first is that intermingling the
405 two makes it more difficult to use a tool like the <literal
406 role="hg-ext">patchbomb</literal> extension to automate the
407 process of submitting the patches to an upstream maintainer.
408 The second is that a backport patch could perturb the context
409 in which a subsequent regular patch is applied, making it
410 impossible to apply the regular patch cleanly
411 <emphasis>without</emphasis> the earlier backport patch
412 already being applied.</para>
414 </sect2>
415 </sect1>
416 <sect1>
417 <title>Useful tips for developing with MQ</title>
419 <sect2>
420 <title>Organising patches in directories</title>
422 <para>If you're working on a substantial project with MQ, it's
423 not difficult to accumulate a large number of patches. For
424 example, I have one patch repository that contains over 250
425 patches.</para>
427 <para>If you can group these patches into separate logical
428 categories, you can if you like store them in different
429 directories; MQ has no problems with patch names that contain
430 path separators.</para>
432 </sect2>
433 <sect2 id="mq-collab:tips:interdiff">
434 <title>Viewing the history of a patch</title>
436 <para>If you're developing a set of patches over a long time,
437 it's a good idea to maintain them in a repository, as
438 discussed in section <xref linkend="sec:mq:repo"/>. If you do
439 so, you'll quickly
440 discover that using the <command role="hg-cmd">hg
441 diff</command> command to look at the history of changes to
442 a patch is unworkable. This is in part because you're looking
443 at the second derivative of the real code (a diff of a diff),
444 but also because MQ adds noise to the process by modifying
445 time stamps and directory names when it updates a
446 patch.</para>
448 <para>However, you can use the <literal
449 role="hg-ext">extdiff</literal> extension, which is bundled
450 with Mercurial, to turn a diff of two versions of a patch into
451 something readable. To do this, you will need a third-party
452 package called <literal role="package">patchutils</literal>
453 <citation>web:patchutils</citation>. This provides a command
454 named <command>interdiff</command>, which shows the
455 differences between two diffs as a diff. Used on two versions
456 of the same diff, it generates a diff that represents the diff
457 from the first to the second version.</para>
459 <para>You can enable the <literal
460 role="hg-ext">extdiff</literal> extension in the usual way,
461 by adding a line to the <literal
462 role="rc-extensions">extensions</literal> section of your
463 <filename role="special"> /.hgrc</filename>.</para>
464 <programlisting>[extensions] extdiff =</programlisting>
465 <para>The <command>interdiff</command> command expects to be
466 passed the names of two files, but the <literal
467 role="hg-ext">extdiff</literal> extension passes the program
468 it runs a pair of directories, each of which can contain an
469 arbitrary number of files. We thus need a small program that
470 will run <command>interdiff</command> on each pair of files in
471 these two directories. This program is available as <filename
472 role="special">hg-interdiff</filename> in the <filename
473 class="directory">examples</filename> directory of the
474 source code repository that accompanies this book. <!--
475 &example.hg-interdiff; --></para>
477 <para>With the <filename role="special">hg-interdiff</filename>
478 program in your shell's search path, you can run it as
479 follows, from inside an MQ patch directory:</para>
480 <programlisting>hg extdiff -p hg-interdiff -r A:B
481 my-change.patch</programlisting>
482 <para>Since you'll probably want to use this long-winded command
483 a lot, you can get <literal role="hg-ext">hgext</literal> to
484 make it available as a normal Mercurial command, again by
485 editing your <filename role="special">
486 /.hgrc</filename>.</para>
487 <programlisting>[extdiff] cmd.interdiff =
488 hg-interdiff</programlisting>
489 <para>This directs <literal role="hg-ext">hgext</literal> to
490 make an <literal>interdiff</literal> command available, so you
491 can now shorten the previous invocation of <command
492 role="hg-ext-extdiff">extdiff</command> to something a
493 little more wieldy.</para>
494 <programlisting>hg interdiff -r A:B
495 my-change.patch</programlisting>
497 <note>
498 <para> The <command>interdiff</command> command works well
499 only if the underlying files against which versions of a
500 patch are generated remain the same. If you create a patch,
501 modify the underlying files, and then regenerate the patch,
502 <command>interdiff</command> may not produce useful
503 output.</para>
504 </note>
506 <para>The <literal role="hg-ext">extdiff</literal> extension is
507 useful for more than merely improving the presentation of MQ
508 patches. To read more about it, go to section <xref
509 linkend="sec:hgext:extdiff"/>.</para>
511 </sect2>
512 </sect1>
513 </chapter>
515 <!--
516 local variables:
517 sgml-parent-document: ("00book.xml" "book" "chapter")
518 end:
519 -->