1ef8493511d7e2e4e13fcb1ab78bcf8b4ca3d6bd
[mesa.git] / docs / submittingpatches.rst
1 Submitting Patches
2 ==================
3
4 Basic guidelines
5 ----------------
6
7 - Patches should not mix code changes with code formatting changes
8 (except, perhaps, in very trivial cases.)
9 - Code patches should follow Mesa `coding
10 conventions <codingstyle.rst>`__.
11 - Whenever possible, patches should only affect individual Mesa/Gallium
12 components.
13 - Patches should never introduce build breaks and should be bisectable
14 (see ``git bisect``.)
15 - Patches should be properly `formatted <#formatting>`__.
16 - Patches should be sufficiently `tested <#testing>`__ before
17 submitting.
18 - Patches should be `submitted <#submit>`__ via a merge request for
19 `review <#reviewing>`__.
20
21 .. _formatting:
22
23 Patch formatting
24 ----------------
25
26 - Lines should be limited to 75 characters or less so that git logs
27 displayed in 80-column terminals avoid line wrapping. Note that git
28 log uses 4 spaces of indentation (4 + 75 < 80).
29 - The first line should be a short, concise summary of the change
30 prefixed with a module name. Examples:
31
32 ::
33
34 mesa: Add support for querying GL_VERTEX_ATTRIB_ARRAY_LONG
35
36 gallium: add PIPE_CAP_DEVICE_RESET_STATUS_QUERY
37
38 i965: Fix missing type in local variable declaration.
39
40 - Subsequent patch comments should describe the change in more detail,
41 if needed. For example:
42
43 ::
44
45 i965: Remove end-of-thread SEND alignment code.
46
47 This was present in Eric's initial implementation of the compaction code
48 for Sandybridge (commit 077d01b6). There is no documentation saying this
49 is necessary, and removing it causes no regressions in piglit on any
50 platform.
51
52 - A "Signed-off-by:" line is not required, but not discouraged either.
53 - If a patch addresses an issue in gitlab, use the Closes: tag For
54 example:
55
56 ::
57
58 Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/1
59
60 Prefer the full url to just ``Closes: #1``, since the url makes it
61 easier to get to the bug page from ``git log``
62
63 **Do not use the Fixes: tag for this!** Mesa already uses Fixes for
64 something else.
65
66 - If a patch addresses a issue introduced with earlier commit, that
67 should be noted in the patch comment. For example:
68
69 ::
70
71 Fixes: d7b3707c612 "util/disk_cache: use stat() to check if entry is a directory"
72
73 - You can produce those fixes lines by running
74
75 ::
76
77 git config --global alias.fixes "show -s --pretty='format:Fixes: %h (\"%s\")'"
78
79 once and then using
80
81 ::
82
83 git fixes <sha1>
84
85 - If there have been several revisions to a patch during the review
86 process, they should be noted such as in this example:
87
88 ::
89
90 st/mesa: add ARB_texture_stencil8 support (v4)
91
92 if we support stencil texturing, enable texture_stencil8
93 there is no requirement to support native S8 for this,
94 the texture can be converted to x24s8 fine.
95
96 v2: fold fixes from Marek in:
97 a) put S8 last in the list
98 b) fix renderable to always test for d/s renderable
99 fixup the texture case to use a stencil only format
100 for picking the format for the texture view.
101 v3: hit fallback for getteximage
102 v4: put s8 back in front, it shouldn't get picked now (Ilia)
103
104 - If someone tested your patch, document it with a line like this:
105
106 ::
107
108 Tested-by: Joe Hacker <jhacker@foo.com>
109
110 - If the patch was reviewed (usually the case) or acked by someone,
111 that should be documented with:
112
113 ::
114
115 Reviewed-by: Joe Hacker <jhacker@foo.com>
116 Acked-by: Joe Hacker <jhacker@foo.com>
117
118 - When updating a merge request add all the tags (``Acked-by:``, ``Reviewed-by:``,
119 ``Fixes:``, ``Cc: mesa-stable`` and/or other) to the commit messages.
120 This provides reviewers with quick feedback if the patch has already
121 been reviewed.
122
123 .. _testing:
124
125 Testing Patches
126 ---------------
127
128 It should go without saying that patches must be tested. In general, do
129 whatever testing is prudent.
130
131 You should always run the Mesa test suite before submitting patches. The
132 test suite can be run using the 'meson test' command. All tests must
133 pass before patches will be accepted, this may mean you have to update
134 the tests themselves.
135
136 Whenever possible and applicable, test the patch with
137 `Piglit <https://piglit.freedesktop.org>`__ and/or
138 `dEQP <https://android.googlesource.com/platform/external/deqp/>`__ to
139 check for regressions.
140
141 As mentioned at the beginning, patches should be bisectable. A good way
142 to test this is to make use of the \`git rebase\` command, to run your
143 tests on each commit. Assuming your branch is based off
144 ``origin/master``, you can run:
145
146 ::
147
148 $ git rebase --interactive --exec "meson test -C build/" origin/master
149
150 replacing ``"meson test"`` with whatever other test you want to run.
151
152 .. _submit:
153
154 Submitting Patches
155 ------------------
156
157 Patches are submitted to the Mesa project via a
158 `GitLab <https://gitlab.freedesktop.org/mesa/mesa>`__ Merge Request.
159
160 Add labels to your MR to help reviewers find it. For example:
161
162 - Mesa changes affecting all drivers: mesa
163 - Hardware vendor specific code: amd, intel, nvidia, ...
164 - Driver specific code: anvil, freedreno, i965, iris, radeonsi, radv,
165 vc4, ...
166 - Other tag examples: gallium, util
167
168 Tick the following when creating the MR. It allows developers to rebase
169 your work on top of master.
170
171 ::
172
173 Allow commits from members who can merge to the target branch
174
175 If you revise your patches based on code review and push an update to
176 your branch, you should maintain a **clean** history in your patches.
177 There should not be "fixup" patches in the history. The series should be
178 buildable and functional after every commit whenever you push the
179 branch.
180
181 It is your responsibility to keep the MR alive and making progress, as
182 there are no guarantees that a Mesa dev will independently take interest
183 in it.
184
185 Some other notes:
186
187 - Make changes and update your branch based on feedback
188 - After an update, for the feedback you handled, close the feedback
189 discussion with the "Resolve Discussion" button. This way the
190 reviewers know which feedback got handled and which didn't.
191 - Old, stale MR may be closed, but you can reopen it if you still want
192 to pursue the changes
193 - You should periodically check to see if your MR needs to be rebased
194 - Make sure your MR is closed if your patches get pushed outside of
195 GitLab
196 - Please send MRs from a personal fork rather than from the main Mesa
197 repository, as it clutters it unnecessarily.
198
199 .. _reviewing:
200
201 Reviewing Patches
202 -----------------
203
204 To participate in code review, you can monitor the GitLab Mesa `Merge
205 Requests <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests>`__
206 page, and/or register for notifications in your gitlab settings.
207
208 When you've reviewed a patch, please be unambiguous about your review.
209 That is, state either
210
211 ::
212
213 Reviewed-by: Joe Hacker <jhacker@foo.com>
214
215 or
216
217 ::
218
219 Acked-by: Joe Hacker <jhacker@foo.com>
220
221 Rather than saying just "LGTM" or "Seems OK".
222
223 If small changes are suggested, it's OK to say something like:
224
225 ::
226
227 With the above fixes, Reviewed-by: Joe Hacker <jhacker@foo.com>
228
229 which tells the patch author that the patch can be committed, as long as
230 the issues are resolved first.
231
232 These Reviewed-by, Acked-by, and Tested-by tags should also be amended
233 into commits in a MR before it is merged.
234
235 When providing a Reviewed-by, Acked-by, or Tested-by tag in a gitlab MR,
236 enclose the tag in backticks:
237
238 ::
239
240 `Reviewed-by: Joe Hacker <jhacker@example.com>`
241
242 This is the markdown format for literal, and will prevent gitlab from
243 hiding the < and > symbols.
244
245 Review by non-experts is encouraged. Understanding how someone else goes
246 about solving a problem is a great way to learn your way around the
247 project. The submitter is expected to evaluate whether they have an
248 appropriate amount of review feedback from people who also understand
249 the code before merging their patches.
250
251 Nominating a commit for a stable branch
252 ---------------------------------------
253
254 There are three ways to nominate a patch for inclusion in the stable
255 branch and release.
256
257 - By adding the ``Cc: mesa-stable`` tag as described below.
258 - By adding the fixes: tag as described below.
259 - By submitting a merge request against the "staging/year.quarter"
260 branch on gitlab.
261
262 Please **DO NOT** send patches to mesa-stable@lists.freedesktop.org, it
263 is not monitored actively and is a historical artifact.
264
265 If you are not the author of the original patch, please Cc: them in your
266 nomination request.
267
268 The current patch status can be observed in the `staging
269 branch <releasing.rst#stagingbranch>`__.
270
271 The stable tag
272 ~~~~~~~~~~~~~~
273
274 If you want a commit to be applied to a stable branch, you should add an
275 appropriate note to the commit message.
276
277 Using a "fixes tag" as described in `Patch formatting <#formatting>`__
278 is the preferred way to nominate a commit that you know ahead of time
279 should be backported. There are scripts that will figure out which
280 releases to apply the patch to automatically, so you don't need to
281 figure it out.
282
283 Alternatively, you may use a "CC:" tag. Here are some examples of such a
284 note:
285
286 ::
287
288 CC: 20.0 19.3 <mesa-stable>
289
290 Using the CC tag **should** include the stable branches you want to
291 nominate the patch to. If you do not provide any version it is nominated
292 to all active stable branches.
293
294 .. _criteria:
295
296 Criteria for accepting patches to the stable branch
297 ---------------------------------------------------
298
299 Mesa has a designated release manager for each stable branch, and the
300 release manager is the only developer that should be pushing changes to
301 these branches. Everyone else should nominate patches using the
302 mechanism described above. The following rules define which patches are
303 accepted and which are not. The stable-release manager is also given
304 broad discretion in rejecting patches that have been nominated.
305
306 - Patch must conform with the `Basic guidelines <#guidelines>`__
307 - Patch must have landed in master first. In case where the original
308 patch is too large and/or otherwise contradicts with the rules set
309 within, a backport is appropriate.
310 - It must not introduce a regression - be that build or runtime wise.
311
312 .. note::
313 If the regression is due to faulty piglit/dEQP/CTS/other test
314 the latter must be fixed first. A reference to the offending test(s)
315 and respective fix(es) should be provided in the nominated patch.
316
317 - Patch cannot be larger than 100 lines.
318 - Patches that move code around with no functional change should be
319 rejected.
320 - Patch must be a bug fix and not a new feature.
321
322 .. note::
323 An exception to this rule, are hardware-enabling "features". For
324 example, `backports <#backports>`__ of new code to support a
325 newly-developed hardware product can be accepted if they can be
326 reasonably determined not to have effects on other hardware.
327
328 - Patch must be reviewed, For example, the commit message has
329 Reviewed-by, Signed-off-by, or Tested-by tags from someone but the
330 author.
331 - Performance patches are considered only if they provide information
332 about the hardware, program in question and observed improvement. Use
333 numbers to represent your measurements.
334
335 If the patch complies with the rules it will be
336 `cherry-picked <releasing.rst#pickntest>`__. Alternatively the release
337 manager will reply to the patch in question stating why the patch has
338 been rejected or would request a backport. The stable-release manager
339 may at times need to force-push changes to the stable branches, for
340 example, to drop a previously-picked patch that was later identified as
341 causing a regression). These force-pushes may cause changes to be lost
342 from the stable branch if developers push things directly. Consider
343 yourself warned.
344
345 .. _backports:
346
347 Sending backports for the stable branch
348 ---------------------------------------
349
350 By default merge conflicts are resolved by the stable-release manager.
351 The release maintainer should resolve trivial conflicts, but for complex
352 conflicts they should ask the original author to provide a backport or
353 de-nominate the patch.
354
355 For patches that either need to be nominated after they've landed in
356 master, or that are known ahead of time to not not apply cleanly to a
357 stable branch (such as due to a rename), using a gitlab MR is most
358 appropriate. The MR should be based on and target the
359 staging/year.quarter branch, not on the year.quarter branch, per the
360 stable branch policy. Assigning the MR to release maintainer for said
361 branch or mentioning them is helpful, but not required.
362
363 Git tips
364 --------
365
366 - ``git rebase -i ...`` is your friend. Don't be afraid to use it.
367 - Apply a fixup to commit FOO.
368
369 .. code-block:: console
370
371 git add ...
372 git commit --fixup=FOO
373 git rebase -i --autosquash ...
374
375 - Test for build breakage between patches e.g last 8 commits.
376
377 .. code-block:: console
378
379 git rebase -i --exec="ninja -C build/" HEAD~8