blob: e6d30706216b9711163d957f33916daa34b0c77b [file] [log] [blame]
Simon Glass26132882012-01-14 15:12:45 +00001# Copyright (c) 2011 The Chromium OS Authors.
2#
Wolfgang Denkd79de1d2013-07-08 09:37:19 +02003# SPDX-License-Identifier: GPL-2.0+
Simon Glass26132882012-01-14 15:12:45 +00004#
5
6What is this?
7=============
8
9This tool is a Python script which:
10- Creates patch directly from your branch
11- Cleans them up by removing unwanted tags
12- Inserts a cover letter with change lists
13- Runs the patches through checkpatch.pl and its own checks
14- Optionally emails them out to selected people
15
16It is intended to automate patch creation and make it a less
17error-prone process. It is useful for U-Boot and Linux work so far,
18since it uses the checkpatch.pl script.
19
20It is configured almost entirely by tags it finds in your commits.
21This means that you can work on a number of different branches at
22once, and keep the settings with each branch rather than having to
23git format-patch, git send-email, etc. with the correct parameters
24each time. So for example if you put:
25
26Series-to: fred.blogs@napier.co.nz
27
28in one of your commits, the series will be sent there.
29
Doug Andersonc2c84bc2012-12-03 14:43:16 +000030In Linux this will also call get_maintainer.pl on each of your
31patches automatically.
32
Simon Glass26132882012-01-14 15:12:45 +000033
34How to use this tool
35====================
36
37This tool requires a certain way of working:
38
39- Maintain a number of branches, one for each patch series you are
40working on
41- Add tags into the commits within each branch to indicate where the
42series should be sent, cover letter, version, etc. Most of these are
43normally in the top commit so it is easy to change them with 'git
44commit --amend'
45- Each branch tracks the upstream branch, so that this script can
46automatically determine the number of commits in it (optional)
47- Check out a branch, and run this script to create and send out your
48patches. Weeks later, change the patches and repeat, knowing that you
49will get a consistent result each time.
50
51
52How to configure it
53===================
54
Simon Glass33bdd9e2013-03-26 13:09:45 +000055For most cases of using patman for U-Boot development, patman will
Doug Andersonc2c84bc2012-12-03 14:43:16 +000056locate and use the file 'doc/git-mailrc' in your U-Boot directory.
57This contains most of the aliases you will need.
58
59For Linux the 'scripts/get_maintainer.pl' handles figuring out where
60to send patches pretty well.
Simon Glass26132882012-01-14 15:12:45 +000061
Vikram Narayanan12fb29a2012-05-23 09:01:06 +000062During the first run patman creates a config file for you by taking the default
63user name and email address from the global .gitconfig file.
64
Vikram Narayananc387d36d2012-05-23 08:58:58 +000065To add your own, create a file ~/.patman like this:
Simon Glass26132882012-01-14 15:12:45 +000066
67>>>>
68# patman alias file
69
70[alias]
71me: Simon Glass <sjg@chromium.org>
72
73u-boot: U-Boot Mailing List <u-boot@lists.denx.de>
74wolfgang: Wolfgang Denk <wd@denx.de>
75others: Mike Frysinger <vapier@gentoo.org>, Fred Bloggs <f.bloggs@napier.net>
76
77<<<<
78
79Aliases are recursive.
80
81The checkpatch.pl in the U-Boot tools/ subdirectory will be located and
82used. Failing that you can put it into your path or ~/bin/checkpatch.pl
83
84
Doug Anderson3d3077c2012-12-03 14:43:17 +000085If you want to change the defaults for patman's command-line arguments,
86you can add a [settings] section to your .patman file. This can be used
87for any command line option by referring to the "dest" for the option in
88patman.py. For reference, the useful ones (at the moment) shown below
89(all with the non-default setting):
90
91>>>
92
93[settings]
94ignore_errors: True
95process_tags: False
96verbose: True
97
98<<<
99
100
Doug Anderson31ffd7f2012-12-03 14:43:18 +0000101If you want to adjust settings (or aliases) that affect just a single
102project you can add a section that looks like [project_settings] or
103[project_alias]. If you want to use tags for your linux work, you could
104do:
105
106>>>
107
108[linux_settings]
109process_tags: True
110
111<<<
112
113
Simon Glass26132882012-01-14 15:12:45 +0000114How to run it
115=============
116
117First do a dry run:
118
Vikram Narayanane95ea8c2012-04-27 06:39:31 +0000119$ ./tools/patman/patman -n
Simon Glass26132882012-01-14 15:12:45 +0000120
121If it can't detect the upstream branch, try telling it how many patches
122there are in your series:
123
Vikram Narayanane95ea8c2012-04-27 06:39:31 +0000124$ ./tools/patman/patman -n -c5
Simon Glass26132882012-01-14 15:12:45 +0000125
126This will create patch files in your current directory and tell you who
127it is thinking of sending them to. Take a look at the patch files.
128
Vikram Narayanane95ea8c2012-04-27 06:39:31 +0000129$ ./tools/patman/patman -n -c5 -s1
Simon Glass26132882012-01-14 15:12:45 +0000130
131Similar to the above, but skip the first commit and take the next 5. This
132is useful if your top commit is for setting up testing.
133
134
135How to add tags
136===============
137
138To make this script useful you must add tags like the following into any
139commit. Most can only appear once in the whole series.
140
141Series-to: email / alias
Wolfgang Denk302007e2012-04-21 18:55:26 +0200142 Email address / alias to send patch series to (you can add this
143 multiple times)
Simon Glass26132882012-01-14 15:12:45 +0000144
145Series-cc: email / alias, ...
Wolfgang Denk302007e2012-04-21 18:55:26 +0200146 Email address / alias to Cc patch series to (you can add this
147 multiple times)
Simon Glass26132882012-01-14 15:12:45 +0000148
149Series-version: n
Wolfgang Denk302007e2012-04-21 18:55:26 +0200150 Sets the version number of this patch series
Simon Glass26132882012-01-14 15:12:45 +0000151
152Series-prefix: prefix
Wolfgang Denk302007e2012-04-21 18:55:26 +0200153 Sets the subject prefix. Normally empty but it can be RFC for
154 RFC patches, or RESEND if you are being ignored.
Simon Glass26132882012-01-14 15:12:45 +0000155
Simon Glasse7ecd3f2012-09-27 15:06:02 +0000156Series-name: name
157 Sets the name of the series. You don't need to have a name, and
158 patman does not yet use it, but it is convenient to put the branch
159 name here to help you keep track of multiple upstreaming efforts.
160
Simon Glass26132882012-01-14 15:12:45 +0000161Cover-letter:
162This is the patch set title
163blah blah
164more blah blah
165END
Wolfgang Denk302007e2012-04-21 18:55:26 +0200166 Sets the cover letter contents for the series. The first line
167 will become the subject of the cover letter
Simon Glass26132882012-01-14 15:12:45 +0000168
Simon Glassc72f3da2013-03-20 16:43:00 +0000169Cover-letter-cc: email / alias
170 Additional email addresses / aliases to send cover letter to (you
171 can add this multiple times)
172
Simon Glass26132882012-01-14 15:12:45 +0000173Series-notes:
174blah blah
175blah blah
176more blah blah
177END
Wolfgang Denk302007e2012-04-21 18:55:26 +0200178 Sets some notes for the patch series, which you don't want in
179 the commit messages, but do want to send, The notes are joined
180 together and put after the cover letter. Can appear multiple
181 times.
Simon Glass26132882012-01-14 15:12:45 +0000182
183 Signed-off-by: Their Name <email>
Wolfgang Denk302007e2012-04-21 18:55:26 +0200184 A sign-off is added automatically to your patches (this is
185 probably a bug). If you put this tag in your patches, it will
186 override the default signoff that patman automatically adds.
Simon Glass26132882012-01-14 15:12:45 +0000187
188 Tested-by: Their Name <email>
Doug Anderson80113ff2013-03-15 13:24:05 +0000189 Reviewed-by: Their Name <email>
Simon Glass26132882012-01-14 15:12:45 +0000190 Acked-by: Their Name <email>
Doug Anderson80113ff2013-03-15 13:24:05 +0000191 These indicate that someone has tested/reviewed/acked your patch.
Wolfgang Denk302007e2012-04-21 18:55:26 +0200192 When you get this reply on the mailing list, you can add this
193 tag to the relevant commit and the script will include it when
194 you send out the next version. If 'Tested-by:' is set to
195 yourself, it will be removed. No one will believe you.
Simon Glass26132882012-01-14 15:12:45 +0000196
197Series-changes: n
198- Guinea pig moved into its cage
199- Other changes ending with a blank line
200<blank line>
Wolfgang Denk302007e2012-04-21 18:55:26 +0200201 This can appear in any commit. It lists the changes for a
202 particular version n of that commit. The change list is
203 created based on this information. Each commit gets its own
204 change list and also the whole thing is repeated in the cover
205 letter (where duplicate change lines are merged).
Simon Glass26132882012-01-14 15:12:45 +0000206
Wolfgang Denk302007e2012-04-21 18:55:26 +0200207 By adding your change lists into your commits it is easier to
208 keep track of what happened. When you amend a commit, remember
209 to update the log there and then, knowing that the script will
210 do the rest.
Simon Glass26132882012-01-14 15:12:45 +0000211
Simon Glassec1d0422013-03-26 13:09:44 +0000212 Cc: Their Name <email>
Wolfgang Denk302007e2012-04-21 18:55:26 +0200213 This copies a single patch to another email address.
Simon Glass26132882012-01-14 15:12:45 +0000214
Simon Glassec1d0422013-03-26 13:09:44 +0000215Series-process-log: sort, uniq
216 This tells patman to sort and/or uniq the change logs. It is
217 assumed that each change log entry is only a single line long.
218 Use 'sort' to sort the entries, and 'uniq' to include only
219 unique entries. If omitted, no change log processing is done.
220 Separate each tag with a comma.
221
Simon Glass26132882012-01-14 15:12:45 +0000222Various other tags are silently removed, like these Chrome OS and
223Gerrit tags:
224
225BUG=...
226TEST=...
227Change-Id:
228Review URL:
229Reviewed-on:
Simon Glass26132882012-01-14 15:12:45 +0000230
231
232Exercise for the reader: Try adding some tags to one of your current
233patch series and see how the patches turn out.
234
235
236Where Patches Are Sent
237======================
238
Vikram Narayanan867ad2a2012-04-25 05:45:05 +0000239Once the patches are created, patman sends them using git send-email. The
Simon Glass26132882012-01-14 15:12:45 +0000240whole series is sent to the recipients in Series-to: and Series-cc.
241You can Cc individual patches to other people with the Cc: tag. Tags in the
242subject are also picked up to Cc patches. For example, a commit like this:
243
244>>>>
245commit 10212537b85ff9b6e09c82045127522c0f0db981
246Author: Mike Frysinger <vapier@gentoo.org>
Wolfgang Denk302007e2012-04-21 18:55:26 +0200247Date: Mon Nov 7 23:18:44 2011 -0500
Simon Glass26132882012-01-14 15:12:45 +0000248
249 x86: arm: add a git mailrc file for maintainers
250
251 This should make sending out e-mails to the right people easier.
252
253 Cc: sandbox, mikef, ag
254 Cc: afleming
255<<<<
256
257will create a patch which is copied to x86, arm, sandbox, mikef, ag and
258afleming.
259
Doug Anderson05416af2012-12-03 14:40:43 +0000260If you have a cover letter it will get sent to the union of the CC lists of
Simon Glassc72f3da2013-03-20 16:43:00 +0000261all of the other patches. If you want to sent it to additional people you
262can add a tag:
263
264Cover-letter-cc: <list of addresses>
265
266These people will get the cover letter even if they are not on the To/Cc
267list for any of the patches.
Doug Anderson05416af2012-12-03 14:40:43 +0000268
Simon Glass26132882012-01-14 15:12:45 +0000269
270Example Work Flow
271=================
272
273The basic workflow is to create your commits, add some tags to the top
274commit, and type 'patman' to check and send them.
275
276Here is an example workflow for a series of 4 patches. Let's say you have
277these rather contrived patches in the following order in branch us-cmd in
278your tree where 'us' means your upstreaming activity (newest to oldest as
279output by git log --oneline):
280
281 7c7909c wip
282 89234f5 Don't include standard parser if hush is used
283 8d640a7 mmc: sparc: Stop using builtin_run_command()
284 0c859a9 Rename run_command2() to run_command()
285 a74443f sandbox: Rename run_command() to builtin_run_command()
286
287The first patch is some test things that enable your code to be compiled,
288but that you don't want to submit because there is an existing patch for it
289on the list. So you can tell patman to create and check some patches
290(skipping the first patch) with:
291
292 patman -s1 -n
293
294If you want to do all of them including the work-in-progress one, then
295(if you are tracking an upstream branch):
296
297 patman -n
298
299Let's say that patman reports an error in the second patch. Then:
300
301 git rebase -i HEAD~6
302 <change 'pick' to 'edit' in 89234f5>
303 <use editor to make code changes>
304 git add -u
305 git rebase --continue
306
307Now you have an updated patch series. To check it:
308
309 patman -s1 -n
310
311Let's say it is now clean and you want to send it. Now you need to set up
312the destination. So amend the top commit with:
313
314 git commit --amend
315
316Use your editor to add some tags, so that the whole commit message is:
317
318 The current run_command() is really only one of the options, with
319 hush providing the other. It really shouldn't be called directly
320 in case the hush parser is bring used, so rename this function to
321 better explain its purpose.
322
323 Series-to: u-boot
324 Series-cc: bfin, marex
325 Series-prefix: RFC
326 Cover-letter:
327 Unified command execution in one place
328
329 At present two parsers have similar code to execute commands. Also
330 cmd_usage() is called all over the place. This series adds a single
331 function which processes commands called cmd_process().
332 END
333
334 Change-Id: Ica71a14c1f0ecb5650f771a32fecb8d2eb9d8a17
335
336
337You want this to be an RFC and Cc the whole series to the bfin alias and
338to Marek. Two of the patches have tags (those are the bits at the front of
339the subject that say mmc: sparc: and sandbox:), so 8d640a7 will be Cc'd to
340mmc and sparc, and the last one to sandbox.
341
342Now to send the patches, take off the -n flag:
343
344 patman -s1
345
346The patches will be created, shown in your editor, and then sent along with
347the cover letter. Note that patman's tags are automatically removed so that
348people on the list don't see your secret info.
349
350Of course patches often attract comments and you need to make some updates.
351Let's say one person sent comments and you get an Acked-by: on one patch.
352Also, the patch on the list that you were waiting for has been merged,
353so you can drop your wip commit. So you resync with upstream:
354
Wolfgang Denk302007e2012-04-21 18:55:26 +0200355 git fetch origin (or whatever upstream is called)
Simon Glass26132882012-01-14 15:12:45 +0000356 git rebase origin/master
357
358and use git rebase -i to edit the commits, dropping the wip one. You add
359the ack tag to one commit:
360
361 Acked-by: Heiko Schocher <hs@denx.de>
362
363update the Series-cc: in the top commit:
364
365 Series-cc: bfin, marex, Heiko Schocher <hs@denx.de>
366
367and remove the Series-prefix: tag since it it isn't an RFC any more. The
368series is now version two, so the series info in the top commit looks like
369this:
370
371 Series-to: u-boot
372 Series-cc: bfin, marex, Heiko Schocher <hs@denx.de>
373 Series-version: 2
374 Cover-letter:
375 ...
376
377Finally, you need to add a change log to the two commits you changed. You
378add change logs to each individual commit where the changes happened, like
379this:
380
381 Series-changes: 2
382 - Updated the command decoder to reduce code size
383 - Wound the torque propounder up a little more
384
385(note the blank line at the end of the list)
386
387When you run patman it will collect all the change logs from the different
388commits and combine them into the cover letter, if you have one. So finally
389you have a new series of commits:
390
391 faeb973 Don't include standard parser if hush is used
392 1b2f2fe mmc: sparc: Stop using builtin_run_command()
393 cfbe330 Rename run_command2() to run_command()
394 0682677 sandbox: Rename run_command() to builtin_run_command()
395
396so to send them:
397
398 patman
399
400and it will create and send the version 2 series.
401
402General points:
403
4041. When you change back to the us-cmd branch days or weeks later all your
405information is still there, safely stored in the commits. You don't need
406to remember what version you are up to, who you sent the last lot of patches
407to, or anything about the change logs.
408
4092. If you put tags in the subject, patman will Cc the maintainers
410automatically in many cases.
411
4123. If you want to keep the commits from each series you sent so that you can
413compare change and see what you did, you can either create a new branch for
414each version, or just tag the branch before you start changing it:
415
416 git tag sent/us-cmd-rfc
417 ...later...
418 git tag sent/us-cmd-v2
419
4204. If you want to modify the patches a little before sending, you can do
421this in your editor, but be careful!
422
4235. If you want to run git send-email yourself, use the -n flag which will
424print out the command line patman would have used.
425
4266. It is a good idea to add the change log info as you change the commit,
427not later when you can't remember which patch you changed. You can always
428go back and change or remove logs from commits.
429
430
431Other thoughts
432==============
433
434This script has been split into sensible files but still needs work.
435Most of these are indicated by a TODO in the code.
436
437It would be nice if this could handle the In-reply-to side of things.
438
Gerhard Sittigda255af2013-07-14 11:27:45 +0200439The tests are incomplete, as is customary. Use the --test flag to run them,
440and make sure you are in the tools/patman directory first:
Simon Glass26132882012-01-14 15:12:45 +0000441
442 $ cd /path/to/u-boot
Gerhard Sittigda255af2013-07-14 11:27:45 +0200443 $ cd tools/patman
444 $ ./patman --test
Simon Glass26132882012-01-14 15:12:45 +0000445
446Error handling doesn't always produce friendly error messages - e.g.
447putting an incorrect tag in a commit may provide a confusing message.
448
449There might be a few other features not mentioned in this README. They
450might be bugs. In particular, tags are case sensitive which is probably
451a bad thing.
452
453
454Simon Glass <sjg@chromium.org>
455v1, v2, 19-Oct-11
456revised v3 24-Nov-11