1 <!DOCTYPE HTML PUBLIC
"-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
4 <meta http-equiv=
"content-type" content=
"text/html; charset=utf-8">
5 <title>Development Notes
</title>
6 <link rel=
"stylesheet" type=
"text/css" href=
"mesa.css">
11 <h1>The Mesa
3D Graphics Library
</h1>
14 <iframe src=
"contents.html"></iframe>
17 <h1>Development Notes
</h1>
21 <li><a href=
"#style">Coding Style
</a>
22 <li><a href=
"#submitting">Submitting Patches
</a>
23 <li><a href=
"#release">Making a New Mesa Release
</a>
24 <li><a href=
"#extensions">Adding Extensions
</a>
28 <h2 id=
"style">Coding Style
</h2>
31 Mesa is over
20 years old and the coding style has evolved over time.
32 Some old parts use a style that's a bit out of date.
33 If the guidelines below don't cover something, try following the format of
34 existing, neighboring code.
38 Basic formatting guidelines
42 <li>3-space indentation, no tabs.
43 <li>Limit lines to
78 or fewer characters. The idea is to prevent line
44 wrapping in
80-column editors and terminals. There are exceptions, such
45 as if you're defining a large, static table of information.
46 <li>Opening braces go on the same line as the if/for/while statement.
56 <li>Put a space before/after operators. For example,
<tt>a = b + c;
</tt>
57 and not
<tt>a=b+c;
</tt>
59 <li>This GNU indent command generally does the right thing for formatting:
61 indent -br -i3 -npcs --no-tabs infile.c -o outfile.c
64 <li>Use comments wherever you think it would be helpful for other developers.
65 Several specific cases and style examples follow. Note that we roughly
66 follow
<a href=
"http://www.stack.nl/~dimitri/doxygen/">Doxygen
</a> conventions.
71 /* null-out pointer to prevent dangling reference below */
76 bufferObj = NULL; /* prevent dangling reference below */
80 /* If this is a new buffer object id, or one which was generated but
81 * never used before, allocate a buffer object now.
84 We try to quote the OpenGL specification where prudent:
86 /* Page
38 of the PDF of the OpenGL ES
3.0 spec says:
88 *
"An INVALID_OPERATION error is generated for any of the following
91 * * <length> is zero."
93 * Additionally, page
94 of the PDF of the OpenGL
4.5 core spec
94 * (
30.10.2014) also says this, so it's no longer allowed for desktop GL,
98 Function comment example:
101 * Create and initialize a new buffer object. Called via the
102 * ctx-
>Driver.CreateObject() driver callback function.
103 * \param name integer name of the object
104 * \param type one of GL_FOO, GL_BAR, etc.
105 * \return pointer to new object or NULL if error
108 _mesa_create_object(GLuint name, GLenum type)
114 <li>Put the function return type and qualifiers on one line and the function
115 name and parameters on the next, as seen above. This makes it easy to use
116 <code>grep ^function_name dir/*
</code> to find function definitions. Also,
117 the opening brace goes on the next line by itself (see above.)
119 <li>Function names follow various conventions depending on the type of function:
121 glFooBar() - a public GL entry point (in glapi_dispatch.c)
122 _mesa_FooBar() - the internal immediate mode function
123 save_FooBar() - retained mode (display list) function in dlist.c
124 foo_bar() - a static (private) function
125 _mesa_foo_bar() - an internal non-static Mesa function
128 <li>Constants, macros and enumerant names are ALL_UPPERCASE, with _ between
130 <li>Mesa usually uses camel case for local variables (Ex:
"localVarname")
131 while gallium typically uses underscores (Ex:
"local_var_name").
132 <li>Global variables are almost never used because Mesa should be thread-safe.
134 <li>Booleans. Places that are not directly visible to the GL API
135 should prefer the use of
<tt>bool
</tt>,
<tt>true
</tt>, and
136 <tt>false
</tt> over
<tt>GLboolean
</tt>,
<tt>GL_TRUE
</tt>, and
137 <tt>GL_FALSE
</tt>. In C code, this may mean that
138 <tt>#include
<stdbool.h
></tt> needs to be added. The
139 <tt>try_emit_
</tt>* methods in src/mesa/program/ir_to_mesa.cpp and
140 src/mesa/state_tracker/st_glsl_to_tgsi.cpp can serve as examples.
145 <h2 id=
"submitting">Submitting patches
</h2>
148 The basic guidelines for submitting patches are:
152 <li>Patches should be sufficiently tested before submitting.
153 <li>Code patches should follow Mesa coding conventions.
154 <li>Whenever possible, patches should only effect individual Mesa/Gallium
156 <li>Patches should never introduce build breaks and should be bisectable (see
157 <code>git bisect
</code>.)
158 <li>Patches should be properly formatted (see below).
159 <li>Patches should be submitted to mesa-dev for review using
160 <code>git send-email
</code>.
161 <li>Patches should not mix code changes with code formatting changes (except,
162 perhaps, in very trivial cases.)
165 <h3>Patch formatting
</h3>
168 The basic rules for patch formatting are:
172 <li>Lines should be limited to
75 characters or less so that git logs
173 displayed in
80-column terminals avoid line wrapping. Note that git
174 log uses
4 spaces of indentation (
4 +
75 < 80).
175 <li>The first line should be a short, concise summary of the change prefixed
176 with a module name. Examples:
178 mesa: Add support for querying GL_VERTEX_ATTRIB_ARRAY_LONG
180 gallium: add PIPE_CAP_DEVICE_RESET_STATUS_QUERY
182 i965: Fix missing type in local variable declaration.
184 <li>Subsequent patch comments should describe the change in more detail,
185 if needed. For example:
187 i965: Remove end-of-thread SEND alignment code.
189 This was present in Eric's initial implementation of the compaction code
190 for Sandybridge (commit
077d01b6). There is no documentation saying this
191 is necessary, and removing it causes no regressions in piglit on any
194 <li>A
"Signed-off-by:" line is not required, but not discouraged either.
195 <li>If a patch address a bugzilla issue, that should be noted in the
196 patch comment. For example:
198 Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=
89689
200 <li>If there have been several revisions to a patch during the review
201 process, they should be noted such as in this example:
203 st/mesa: add ARB_texture_stencil8 support (v4)
205 if we support stencil texturing, enable texture_stencil8
206 there is no requirement to support native S8 for this,
207 the texture can be converted to x24s8 fine.
209 v2: fold fixes from Marek in:
210 a) put S8 last in the list
211 b) fix renderable to always test for d/s renderable
212 fixup the texture case to use a stencil only format
213 for picking the format for the texture view.
214 v3: hit fallback for getteximage
215 v4: put s8 back in front, it shouldn't get picked now (Ilia)
217 <li>If someone tested your patch, document it with a line like this:
219 Tested-by: Joe Hacker
<jhacker@foo.com
>
221 <li>If the patch was reviewed (usually the case) or acked by someone,
222 that should be documented with:
224 Reviewed-by: Joe Hacker
<jhacker@foo.com
>
225 Acked-by: Joe Hacker
<jhacker@foo.com
>
231 <h3>Testing Patches
</h3>
234 It should go without saying that patches must be tested. In general,
235 do whatever testing is prudent.
239 You should always run the Mesa test suite before submitting patches.
240 The test suite can be run using the 'make check' command. All tests
241 must pass before patches will be accepted, this may mean you have
242 to update the tests themselves.
246 Whenever possible and applicable, test the patch with
247 <a href=
"http://people.freedesktop.org/~nh/piglit/">Piglit
</a> to
248 check for regressions.
252 <h3>Mailing Patches
</h3>
255 Patches should be sent to the Mesa mailing list for review.
256 When submitting a patch make sure to use git send-email rather than attaching
257 patches to emails. Sending patches as attachments prevents people from being
258 able to provide in-line review comments.
262 When submitting follow-up patches you can use --in-reply-to to make v2, v3,
263 etc patches show up as replies to the originals. This usually works well
264 when you're sending out updates to individual patches (as opposed to
265 re-sending the whole series). Using --in-reply-to makes
266 it harder for reviewers to accidentally review old patches.
269 <h3>Marking a commit as a candidate for a stable branch
</h3>
272 If you want a commit to be applied to a stable branch,
273 you should add an appropriate note to the commit message.
277 Here are some examples of such a note:
280 <li>CC:
<mesa-stable@lists.freedesktop.org
></li>
281 <li>CC:
"9.2 10.0" <mesa-stable@lists.freedesktop.org
></li>
282 <li>CC:
"10.0" <mesa-stable@lists.freedesktop.org
></li>
285 Simply adding the CC to the mesa-stable list address is adequate to nominate
286 the commit for the most-recently-created stable branch. It is only necessary
287 to specify a specific branch name, (such as
"9.2 10.0" or
"10.0" in the
288 examples above), if you want to nominate the commit for an older stable
289 branch. And, as in these examples, you can nominate the commit for the older
290 branch in addition to the more recent branch, or nominate the commit
291 exclusively for the older branch.
293 This
"CC" syntax for patch nomination will cause patches to automatically be
294 copied to the mesa-stable@ mailing list when you use
"git send-email" to send
295 patches to the mesa-dev@ mailing list. Also, if you realize that a commit
296 should be nominated for the stable branch after it has already been committed,
297 you can send a note directly to the mesa-stable@lists.freedesktop.org where
298 the Mesa stable-branch maintainers will receive it. Be sure to mention the
299 commit ID of the commit of interest (as it appears in the mesa master branch).
301 The latest set of patches that have been nominated, accepted, or rejected for
302 the upcoming stable release can always be seen on the
303 <a href=
"http://cworth.org/~cworth/mesa-stable-queue/">Mesa Stable Queue
</a>
306 <h3>Criteria for accepting patches to the stable branch
</h3>
308 Mesa has a designated release manager for each stable branch, and the release
309 manager is the only developer that should be pushing changes to these
310 branches. Everyone else should simply nominate patches using the mechanism
313 The stable-release manager will work with the list of nominated patches, and
314 for each patch that meets the crtieria below will cherry-pick the patch with:
315 <code>git cherry-pick -x
<commit
></code>. The
<code>-x
</code> option is
316 important so that the picked patch references the comit ID of the original
319 The stable-release manager may at times need to force-push changes to the
320 stable branches, for example, to drop a previously-picked patch that was later
321 identified as causing a regression). These force-pushes may cause changes to
322 be lost from the stable branch if developers push things directly. Consider
325 The stable-release manager is also given broad discretion in rejecting patches
326 that have been nominated for the stable branch. The most basic rule is that
327 the stable branch is for bug fixes only, (no new features, no
328 regressions). Here is a non-exhaustive list of some reasons that a patch may
332 <li>Patch introduces a regression. Any reported build breakage or other
333 regression caused by a particular patch, (game no longer work, piglit test
334 changes from PASS to FAIL), is justification for rejecting a patch.
</li>
336 <li>Patch is too large, (say, larger than
100 lines)
</li>
338 <li>Patch is not a fix. For example, a commit that moves code around with no
339 functional change should be rejected.
</li>
341 <li>Patch fix is not clearly described. For example, a commit message
342 of only a single line, no description of the bug, no mention of bugzilla,
345 <li>Patch has not obviously been reviewed, For example, the commit message
346 has no Reviewed-by, Signed-off-by, nor Tested-by tags from anyone but the
349 <li>Patch has not already been merged to the master branch. As a rule, bug
350 fixes should never be applied first to a stable branch. Patches should land
351 first on the master branch and then be cherry-picked to a stable
352 branch. (This is to avoid future releases causing regressions if the patch
353 is not also applied to master.) The only things that might look like
354 exceptions would be backports of patches from master that happen to look
355 significantly different.
</li>
357 <li>Patch depends on too many other patches. Ideally, all stable-branch
358 patches should be self-contained. It sometimes occurs that a single, logical
359 bug-fix occurs as two separate patches on master, (such as an original
360 patch, then a subsequent fix-up to that patch). In such a case, these two
361 patches should be squashed into a single, self-contained patch for the
362 stable branch. (Of course, if the squashing makes the patch too large, then
363 that could be a reason to reject the patch.)
</li>
365 <li>Patch includes new feature development, not bug fixes. New OpenGL
366 features, extensions, etc. should be applied to Mesa master and included in
367 the next major release. Stable releases are intended only for bug fixes.
369 Note: As an exception to this rule, the stable-release manager may accept
370 hardware-enabling
"features". For example, backports of new code to support
371 a newly-developed hardware product can be accepted if they can be reasonably
372 determined to not have effects on other hardware.
</li>
374 <li>Patch is a performance optimization. As a rule, performance patches are
375 not candidates for the stable branch. The only exception might be a case
376 where an application's performance was recently severely impacted so as to
377 become unusable. The fix for this performance regression could then be
378 considered for a stable branch. The optimization must also be
379 non-controversial and the patches still need to meet the other criteria of
380 being simple and self-contained
</li>
382 <li>Patch introduces a new failure mode (such as an assert). While the new
383 assert might technically be correct, for example to make Mesa more
384 conformant, this is not the kind of
"bug fix" we want in a stable
385 release. The potential problem here is that an OpenGL program that was
386 previously working, (even if technically non-compliant with the
387 specification), could stop working after this patch. So that would be a
388 regression that is unaacceptable for the stable branch.
</li>
392 <h2 id=
"release">Making a New Mesa Release
</h2>
395 These are the instructions for making a new Mesa release.
398 <h3>Get latest source files
</h3>
400 Use git to get the latest Mesa files from the git repository, from whatever
401 branch is relevant. This document uses the convention X.Y.Z for the release
402 being created, which should be created from a branch named X.Y.
405 <h3>Perform basic testing
</h3>
407 The release manager should, at the very least, test the code by compiling it,
408 installing it, and running the latest piglit to ensure that no piglit tests
409 have regressed since the previous release.
413 The release manager should do this testing with at least one hardware driver,
414 (say, whatever is contained in the local development machine), as well as on
415 both Gallium and non-Gallium software drivers. The software testing can be
416 performed by running piglit with the following environment-variable set:
420 LIBGL_ALWAYS_SOFTWARE=
1
423 And Gallium vs. non-Gallium software drivers can be obtained by using the
424 following configure flags on separate builds:
427 --with-dri-drivers=swrast
428 --with-gallium-drivers=swrast
432 Note: If both options are given in one build, both swrast_dri.so drivers will
433 be compiled, but only one will be installed. The following command can be used
434 to ensure the correct driver is being tested:
438 LIBGL_ALWAYS_SOFTWARE=
1 glxinfo | grep
"renderer string"
441 If any regressions are found in this testing with piglit, stop here, and do
442 not perform a release until regressions are fixed.
444 <h3>Update version in file VERSION
</h3>
447 Increment the version contained in the file VERSION at Mesa's top-level, then
451 <h3>Create release notes for the new release
</h3>
454 Create a new file docs/relnotes/X.Y.Z.html, (follow the style of the previous
455 release notes). Note that the sha256sums section of the release notes should
456 be empty at this point.
460 Two scripts are available to help generate portions of the release notes:
463 ./bin/bugzilla_mesa.sh
464 ./bin/shortlog_mesa.sh
468 The first script identifies commits that reference bugzilla bugs and obtains
469 the descriptions of those bugs from bugzilla. The second script generates a
470 log of all commits. In both cases, HTML-formatted lists are printed to stdout
471 to be included in the release notes.
478 <h3>Make the release archives, signatures, and the release tag
</h3>
480 From inside the Mesa directory:
487 After the tarballs are created, the sha256 checksums for the files will
488 be computed and printed. These will be used in a step below.
492 It's important at this point to also verify that the constructed tar file
497 tar xjf MesaLib-X.Y.Z.tar.bz2
499 ./configure --enable-gallium-llvm
505 Some touch testing should also be performed at this point, (run glxgears or
506 more involved OpenGL programs against the installed Mesa).
510 Create detached GPG signatures for each of the archive files created above:
514 gpg --sign --detach MesaLib-X.Y.Z.tar.gz
515 gpg --sign --detach MesaLib-X.Y.Z.tar.bz2
516 gpg --sign --detach MesaLib-X.Y.Z.zip
520 Tag the commit used for the build:
524 git tag -s mesa-X.Y.X -m
"Mesa X.Y.Z release"
528 Note: It would be nice to investigate and fix the issue that causes the
529 tarballs target to fail with multiple build process, such as with
"-j4". It
530 would also be nice to incorporate all of the above commands into a single
531 makefile target. And instead of a custom
"tarballs" target, we should
532 incorporate things into the standard
"make dist" and
"make distcheck" targets.
535 <h3>Add the sha256sums to the release notes
</h3>
538 Edit docs/relnotes/X.Y.Z.html to add the sha256sums printed as part of
"make
539 tarballs" in the previous step. Commit this change.
542 <h3>Push all commits and the tag creates above
</h3>
545 This is the first step that cannot easily be undone. The release is going
546 forward from this point:
550 git push origin X.Y --tags
553 <h3>Install the release files and signatures on the distribution server
</h3>
556 The following commands can be used to copy the release archive files and
557 signatures to the freedesktop.org server:
561 scp MesaLib-X.Y.Z* people.freedesktop.org:
562 ssh people.freedesktop.org
563 cd /srv/ftp.freedesktop.org/pub/mesa
566 mv ~/MesaLib-X.Y.Z* .
569 <h3>Back on mesa master, andd the new release notes into the tree
</h3>
572 Something like the following steps will do the trick:
576 cp docs/relnotes/X.Y.Z.html /tmp
578 cp /tmp/X.Y.Z.html docs/relnotes
579 git add docs/relnotes/X.Y.Z.html
583 Also, edit docs/relnotes.html to add a link to the new release notes, and edit
584 docs/index.html to add a news entry. Then commit and push:
588 git commit -a -m
"docs: Import X.Y.Z release notes, add news item."
592 <h3>Update the mesa3d.org website
</h3>
595 NOTE: The recent release managers have not been performing this step
596 themselves, but leaving this to Brian Paul, (who has access to the
597 sourceforge.net hosting for mesa3d.org). Brian is more than willing to grant
598 the permission necessary to future release managers to do this step on their
603 Update the web site by copying the docs/ directory's files to
604 /home/users/b/br/brianp/mesa-www/htdocs/ with:
607 sftp USERNAME,mesa3d@web.sourceforge.net
612 <h3>Announce the release
</h3>
614 Make an announcement on the mailing lists:
616 <em>mesa-dev@lists.freedesktop.org
</em>,
618 <em>mesa-announce@lists.freedesktop.org
</em>
620 Follow the template of previously-sent release announcements. The following
621 command can be used to generate the log of changes to be included in the
622 release announcement:
625 git shortlog mesa-X.Y.Z-
1..mesa-X.Y.Z
630 <h2 id=
"extensions">Adding Extensions
</h2>
633 To add a new GL extension to Mesa you have to do at least the following.
637 If glext.h doesn't define the extension, edit include/GL/gl.h and add
640 #ifndef GL_EXT_the_extension_name
641 #define GL_EXT_the_extension_name
1
642 /* declare the new enum tokens */
643 /* prototype the new functions */
644 /* TYPEDEFS for the new functions */
649 In the src/mapi/glapi/gen/ directory, add the new extension functions and
650 enums to the gl_API.xml file.
651 Then, a bunch of source files must be regenerated by executing the
652 corresponding Python scripts.
655 Add a new entry to the
<code>gl_extensions
</code> struct in mtypes.h
658 Update the
<code>extensions.c
</code> file.
661 From this point, the best way to proceed is to find another extension,
662 similar to the new one, that's already implemented in Mesa and use it
666 If the new extension adds new GL state, the functions in get.c, enable.c
667 and attrib.c will most likely require new code.
670 The dispatch tests check_table.cpp and dispatch_sanity.cpp
671 should be updated with details about the new extensions functions. These
672 tests are run using 'make check'