11 This document contains the LLVM Developer Policy which defines the project's
12 policy towards developers and their contributions. The intent of this policy is
13 to eliminate miscommunication, rework, and confusion that might arise from the
14 distributed nature of LLVM's development. By stating the policy in clear terms,
15 we hope each developer can know ahead of time what to expect when making LLVM
16 contributions. This policy covers all llvm.org subprojects, including Clang,
19 This policy is also designed to accomplish the following objectives:
21 #. Attract both users and developers to the LLVM project.
23 #. Make life as simple and easy for contributors as possible.
25 #. Keep the top of Subversion trees as stable as possible.
27 #. Establish awareness of the project's :ref:`copyright, license, and patent
28 policies <copyright-license-patents>` with contributors to the project.
30 This policy is aimed at frequent contributors to LLVM. People interested in
31 contributing one-off patches can do so in an informal way by sending them to the
32 `llvm-commits mailing list
33 <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>`_ and engaging another
34 developer to see it through the process.
39 This section contains policies that pertain to frequent LLVM developers. We
40 always welcome `one-off patches`_ from people who do not routinely contribute to
41 LLVM, but we expect more from frequent contributors to keep the system as
42 efficient as possible for everyone. Frequent LLVM contributors are expected to
43 meet the following requirements in order for LLVM to maintain a high standard of
49 Developers should stay informed by reading at least the "dev" mailing list for
50 the projects you are interested in, such as `llvmdev
51 <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_ for LLVM, `cfe-dev
52 <http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev>`_ for Clang, or `lldb-dev
53 <http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev>`_ for LLDB. If you are
54 doing anything more than just casual work on LLVM, it is suggested that you also
55 subscribe to the "commits" mailing list for the subproject you're interested in,
57 <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>`_, `cfe-commits
58 <http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits>`_, or `lldb-commits
59 <http://lists.cs.uiuc.edu/mailman/listinfo/lldb-commits>`_. Reading the
60 "commits" list and paying attention to changes being made by others is a good
61 way to see what other people are interested in and watching the flow of the
64 We recommend that active developers register an email account with `LLVM
65 Bugzilla <http://llvm.org/bugs/>`_ and preferably subscribe to the `llvm-bugs
66 <http://lists.cs.uiuc.edu/mailman/listinfo/llvmbugs>`_ email list to keep track
67 of bugs and enhancements occurring in LLVM. We really appreciate people who are
68 proactive at catching incoming bugs in their components and dealing with them
71 Please be aware that all public LLVM mailing lists are public and archived, and
72 that notices of confidentiality or non-disclosure cannot be respected.
77 Making and Submitting a Patch
78 -----------------------------
80 When making a patch for review, the goal is to make it as easy for the reviewer
81 to read it as possible. As such, we recommend that you:
83 #. Make your patch against the Subversion trunk, not a branch, and not an old
84 version of LLVM. This makes it easy to apply the patch. For information on
85 how to check out SVN trunk, please see the `Getting Started
86 Guide <GettingStarted.html#checkout>`_.
88 #. Similarly, patches should be submitted soon after they are generated. Old
89 patches may not apply correctly if the underlying code changes between the
90 time the patch was created and the time it is applied.
92 #. Patches should be made with ``svn diff``, or similar. If you use a
93 different tool, make sure it uses the ``diff -u`` format and that it
94 doesn't contain clutter which makes it hard to read.
96 #. If you are modifying generated files, such as the top-level ``configure``
97 script, please separate out those changes into a separate patch from the rest
100 Once your patch is ready, submit it by emailing it to the appropriate project's
101 commit mailing list (or commit it directly if applicable). Alternatively, some
102 patches get sent to the project's development list or component of the LLVM bug
103 tracker, but the commit list is the primary place for reviews and should
104 generally be preferred.
106 When sending a patch to a mailing list, it is a good idea to send it as an
107 *attachment* to the message, not embedded into the text of the message. This
108 ensures that your mailer will not mangle the patch when it sends it (e.g. by
109 making whitespace changes or by wrapping lines).
111 *For Thunderbird users:* Before submitting a patch, please open *Preferences >
112 Advanced > General > Config Editor*, find the key
113 ``mail.content_disposition_type``, and set its value to ``1``. Without this
114 setting, Thunderbird sends your attachment using ``Content-Disposition: inline``
115 rather than ``Content-Disposition: attachment``. Apple Mail gamely displays such
116 a file inline, making it difficult to work with for reviewers using that
119 When submitting patches, please do not add confidentiality or non-disclosure
120 notices to the patches themselves. These notices conflict with the `LLVM
121 License`_ and may result in your contribution being excluded.
128 LLVM has a code review policy. Code review is one way to increase the quality of
129 software. We generally follow these policies:
131 #. All developers are required to have significant changes reviewed before they
132 are committed to the repository.
134 #. Code reviews are conducted by email on the relevant project's commit mailing
135 list, or alternatively on the project's development list or bug tracker.
137 #. Code can be reviewed either before it is committed or after. We expect major
138 changes to be reviewed before being committed, but smaller changes (or
139 changes where the developer owns the component) can be reviewed after commit.
141 #. The developer responsible for a code change is also responsible for making
142 all necessary review-related changes.
144 #. Code review can be an iterative process, which continues until the patch is
145 ready to be committed. Specifically, once a patch is sent out for review, it
146 needs an explicit "looks good" before it is submitted. Do not assume silent
147 approval, or request active objections to the patch with a deadline.
149 Sometimes code reviews will take longer than you would hope for, especially for
150 larger features. Accepted ways to speed up review times for your patches are:
152 * Review other people's patches. If you help out, everybody will be more
153 willing to do the same for you; goodwill is our currency.
154 * Ping the patch. If it is urgent, provide reasons why it is important to you to
155 get this patch landed and ping it every couple of days. If it is
156 not urgent, the common courtesy ping rate is one week. Remember that you're
157 asking for valuable time from other professional developers.
158 * Ask for help on IRC. Developers on IRC will be able to either help you
159 directly, or tell you who might be a good reviewer.
160 * Split your patch into multiple smaller patches that build on each other. The
161 smaller your patch, the higher the probability that somebody will take a quick
164 Developers should participate in code reviews as both reviewers and
165 reviewees. If someone is kind enough to review your code, you should return the
166 favor for someone else. Note that anyone is welcome to review and give feedback
167 on a patch, but only people with Subversion write access can approve it.
169 There is a web based code review tool that can optionally be used
170 for code reviews. See :doc:`Phabricator`.
175 The LLVM Project relies on two features of its process to maintain rapid
176 development in addition to the high quality of its source base: the combination
177 of code review plus post-commit review for trusted maintainers. Having both is
178 a great way for the project to take advantage of the fact that most people do
179 the right thing most of the time, and only commit patches without pre-commit
180 review when they are confident they are right.
182 The trick to this is that the project has to guarantee that all patches that are
183 committed are reviewed after they go in: you don't want everyone to assume
184 someone else will review it, allowing the patch to go unreviewed. To solve this
185 problem, we have a notion of an 'owner' for a piece of the code. The sole
186 responsibility of a code owner is to ensure that a commit to their area of the
187 code is appropriately reviewed, either by themself or by someone else. The list
188 of current code owners can be found in the file
189 `CODE_OWNERS.TXT <http://llvm.org/viewvc/llvm-project/llvm/trunk/CODE_OWNERS.TXT?view=markup>`_
190 in the root of the LLVM source tree.
192 Note that code ownership is completely different than reviewers: anyone can
193 review a piece of code, and we welcome code review from anyone who is
194 interested. Code owners are the "last line of defense" to guarantee that all
195 patches that are committed are actually reviewed.
197 Being a code owner is a somewhat unglamorous position, but it is incredibly
198 important for the ongoing success of the project. Because people get busy,
199 interests change, and unexpected things happen, code ownership is purely opt-in,
200 and anyone can choose to resign their "title" at any time. For now, we do not
201 have an official policy on how one gets elected to be a code owner.
203 .. _include a testcase:
208 Developers are required to create test cases for any bugs fixed and any new
209 features added. Some tips for getting your testcase approved:
211 * All feature and regression test cases are added to the ``llvm/test``
212 directory. The appropriate sub-directory should be selected (see the
213 :doc:`Testing Guide <TestingGuide>` for details).
215 * Test cases should be written in :doc:`LLVM assembly language <LangRef>`.
217 * Test cases, especially for regressions, should be reduced as much as possible,
218 by :doc:`bugpoint <Bugpoint>` or manually. It is unacceptable to place an
219 entire failing program into ``llvm/test`` as this creates a *time-to-test*
220 burden on all developers. Please keep them short.
222 Note that llvm/test and clang/test are designed for regression and small feature
223 tests only. More extensive test cases (e.g., entire applications, benchmarks,
224 etc) should be added to the ``llvm-test`` test suite. The llvm-test suite is
225 for coverage (correctness, performance, etc) testing, not feature or regression
231 The minimum quality standards that any change must satisfy before being
232 committed to the main development branch are:
234 #. Code must adhere to the `LLVM Coding Standards <CodingStandards.html>`_.
236 #. Code must compile cleanly (no errors, no warnings) on at least one platform.
238 #. Bug fixes and new features should `include a testcase`_ so we know if the
239 fix/feature ever regresses in the future.
241 #. Code must pass the ``llvm/test`` test suite.
243 #. The code must not cause regressions on a reasonable subset of llvm-test,
244 where "reasonable" depends on the contributor's judgement and the scope of
245 the change (more invasive changes require more testing). A reasonable subset
246 might be something like "``llvm-test/MultiSource/Benchmarks``".
248 Additionally, the committer is responsible for addressing any problems found in
249 the future that the change is responsible for. For example:
251 * The code should compile cleanly on all supported platforms.
253 * The changes should not cause any correctness regressions in the ``llvm-test``
254 suite and must not cause any major performance regressions.
256 * The change set should not cause performance or correctness regressions for the
259 * The changes should not cause performance or correctness regressions in code
260 compiled by LLVM on all applicable targets.
262 * You are expected to address any `Bugzilla bugs <http://llvm.org/bugs/>`_ that
263 result from your change.
265 We prefer for this to be handled before submission but understand that it isn't
266 possible to test all of this for every submission. Our build bots and nightly
267 testing infrastructure normally finds these problems. A good rule of thumb is
268 to check the nightly testers for regressions the day after your change. Build
269 bots will directly email you if a group of commits that included yours caused a
270 failure. You are expected to check the build bot messages to see if they are
271 your fault and, if so, fix the breakage.
273 Commits that violate these quality standards (e.g. are very broken) may be
274 reverted. This is necessary when the change blocks other developers from making
275 progress. The developer is welcome to re-commit the change after the problem has
283 Although we don't enforce the format of commit messages, we prefer that
284 you follow these guidelines to help review, search in logs, email formatting
285 and so on. These guidelines are very similar to rules used by other open source
288 Most importantly, the contents of the message should be carefully written to
289 convey the rationale of the change (without delving too much in detail). It
290 also should avoid being vague or overly specific. For example, "bits were not
291 set right" will leave the reviewer wondering about which bits, and why they
292 weren't right, while "Correctly set overflow bits in TargetInfo" conveys almost
293 all there is to the change.
295 Below are some guidelines about the format of the message itself:
297 * Separate the commit message into title, body and, if you're not the original
298 author, a "Patch by" attribution line (see below).
300 * The title should be concise. Because all commits are emailed to the list with
301 the first line as the subject, long titles are frowned upon. Short titles
302 also look better in `git log`.
304 * When the changes are restricted to a specific part of the code (e.g. a
305 back-end or optimization pass), it is customary to add a tag to the
306 beginning of the line in square brackets. For example, "[SCEV] ..."
307 or "[OpenMP] ...". This helps email filters and searches for post-commit
310 * The body, if it exists, should be separated from the title by an empty line.
312 * The body should be concise, but explanatory, including a complete
313 reasoning. Unless it is required to understand the change, examples,
314 code snippets and gory details should be left to bug comments, web
315 review or the mailing list.
317 * If the patch fixes a bug in bugzilla, please include the PR# in the message.
319 * `Attribution of Changes`_ should be in a separate line, after the end of
320 the body, as simple as "Patch by John Doe.". This is how we officially
321 handle attribution, and there are automated processes that rely on this
324 * Text formatting and spelling should follow the same rules as documentation
325 and in-code comments, ex. capitalization, full stop, etc.
327 For minor violations of these recommendations, the community normally favors
328 reminding the contributor of this policy over reverting. Minor corrections and
329 omissions can be handled by sending a reply to the commits mailing list.
331 Obtaining Commit Access
332 -----------------------
334 We grant commit access to contributors with a track record of submitting high
335 quality patches. If you would like commit access, please send an email to
336 `Chris <mailto:sabre@nondot.org>`_ with the following information:
338 #. The user name you want to commit with, e.g. "hacker".
340 #. The full name and email address you want message to llvm-commits to come
341 from, e.g. "J. Random Hacker <hacker@yoyodyne.com>".
343 #. A "password hash" of the password you want to use, e.g. "``2ACR96qjUqsyM``".
344 Note that you don't ever tell us what your password is; you just give it to
345 us in an encrypted form. To get this, run "``htpasswd``" (a utility that
346 comes with apache) in crypt mode (often enabled with "``-d``"), or find a web
347 page that will do it for you.
349 Once you've been granted commit access, you should be able to check out an LLVM
350 tree with an SVN URL of "https://username@llvm.org/..." instead of the normal
351 anonymous URL of "http://llvm.org/...". The first time you commit you'll have
352 to type in your password. Note that you may get a warning from SVN about an
353 untrusted key; you can ignore this. To verify that your commit access works,
354 please do a test commit (e.g. change a comment or add a blank line). Your first
355 commit to a repository may require the autogenerated email to be approved by a
356 mailing list. This is normal and will be done when the mailing list owner has
359 If you have recently been granted commit access, these policies apply:
361 #. You are granted *commit-after-approval* to all parts of LLVM. To get
362 approval, submit a `patch`_ to `llvm-commits
363 <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>`_. When approved,
364 you may commit it yourself.
366 #. You are allowed to commit patches without approval which you think are
367 obvious. This is clearly a subjective decision --- we simply expect you to
368 use good judgement. Examples include: fixing build breakage, reverting
369 obviously broken patches, documentation/comment changes, any other minor
372 #. You are allowed to commit patches without approval to those portions of LLVM
373 that you have contributed or maintain (i.e., have been assigned
374 responsibility for), with the proviso that such commits must not break the
375 build. This is a "trust but verify" policy, and commits of this nature are
376 reviewed after they are committed.
378 #. Multiple violations of these policies or a single egregious violation may
379 cause commit access to be revoked.
381 In any case, your changes are still subject to `code review`_ (either before or
382 after they are committed, depending on the nature of the change). You are
383 encouraged to review other peoples' patches as well, but you aren't required
386 .. _discuss the change/gather consensus:
388 Making a Major Change
389 ---------------------
391 When a developer begins a major new project with the aim of contributing it back
392 to LLVM, they should inform the community with an email to the `llvmdev
393 <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_ email list, to the extent
394 possible. The reason for this is to:
396 #. keep the community informed about future changes to LLVM,
398 #. avoid duplication of effort by preventing multiple parties working on the
399 same thing and not knowing about it, and
401 #. ensure that any technical issues around the proposed work are discussed and
402 resolved before any significant work is done.
404 The design of LLVM is carefully controlled to ensure that all the pieces fit
405 together well and are as consistent as possible. If you plan to make a major
406 change to the way LLVM works or want to add a major new extension, it is a good
407 idea to get consensus with the development community before you start working on
410 Once the design of the new feature is finalized, the work itself should be done
411 as a series of `incremental changes`_, not as a long-term development branch.
413 .. _incremental changes:
415 Incremental Development
416 -----------------------
418 In the LLVM project, we do all significant changes as a series of incremental
419 patches. We have a strong dislike for huge changes or long-term development
420 branches. Long-term development branches have a number of drawbacks:
422 #. Branches must have mainline merged into them periodically. If the branch
423 development and mainline development occur in the same pieces of code,
424 resolving merge conflicts can take a lot of time.
426 #. Other people in the community tend to ignore work on branches.
428 #. Huge changes (produced when a branch is merged back onto mainline) are
429 extremely difficult to `code review`_.
431 #. Branches are not routinely tested by our nightly tester infrastructure.
433 #. Changes developed as monolithic large changes often don't work until the
434 entire set of changes is done. Breaking it down into a set of smaller
435 changes increases the odds that any of the work will be committed to the main
438 To address these problems, LLVM uses an incremental development style and we
439 require contributors to follow this practice when making a large/invasive
442 * Large/invasive changes usually have a number of secondary changes that are
443 required before the big change can be made (e.g. API cleanup, etc). These
444 sorts of changes can often be done before the major change is done,
445 independently of that work.
447 * The remaining inter-related work should be decomposed into unrelated sets of
448 changes if possible. Once this is done, define the first increment and get
449 consensus on what the end goal of the change is.
451 * Each change in the set can be stand alone (e.g. to fix a bug), or part of a
452 planned series of changes that works towards the development goal.
454 * Each change should be kept as small as possible. This simplifies your work
455 (into a logical progression), simplifies code review and reduces the chance
456 that you will get negative feedback on the change. Small increments also
457 facilitate the maintenance of a high quality code base.
459 * Often, an independent precursor to a big change is to add a new API and slowly
460 migrate clients to use the new API. Each change to use the new API is often
461 "obvious" and can be committed without review. Once the new API is in place
462 and used, it is much easier to replace the underlying implementation of the
463 API. This implementation change is logically separate from the API
466 If you are interested in making a large change, and this scares you, please make
467 sure to first `discuss the change/gather consensus`_ then ask about the best way
468 to go about making the change.
470 Attribution of Changes
471 ----------------------
473 When contributors submit a patch to an LLVM project, other developers with
474 commit access may commit it for the author once appropriate (based on the
475 progression of code review, etc.). When doing so, it is important to retain
476 correct attribution of contributions to their contributors. However, we do not
477 want the source code to be littered with random attributions "this code written
478 by J. Random Hacker" (this is noisy and distracting). In practice, the revision
479 control system keeps a perfect history of who changed what, and the CREDITS.txt
480 file describes higher-level contributions. If you commit a patch for someone
481 else, please follow the attribution of changes in the simple manner as outlined
482 by the `commit messages`_ section. Overall, please do not add contributor names
485 Also, don't commit patches authored by others unless they have submitted the
486 patch to the project or you have been authorized to submit them on their behalf
487 (you work together and your company authorized you to contribute the patches,
488 etc.). The author should first submit them to the relevant project's commit
489 list, development list, or LLVM bug tracker component. If someone sends you
490 a patch privately, encourage them to submit it to the appropriate list first.
493 IR Backwards Compatibility
494 --------------------------
496 When the IR format has to be changed, keep in mind that we try to maintain some
497 backwards compatibility. The rules are intended as a balance between convenience
498 for llvm users and not imposing a big burden on llvm developers:
500 * The textual format is not backwards compatible. We don't change it too often,
501 but there are no specific promises.
503 * The bitcode format produced by a X.Y release will be readable by all following
504 X.Z releases and the (X+1).0 release.
506 * Newer releases can ignore features from older releases, but they cannot
507 miscompile them. For example, if nsw is ever replaced with something else,
508 dropping it would be a valid way to upgrade the IR.
510 * Debug metadata is special in that it is currently dropped during upgrades.
512 * Non-debug metadata is defined to be safe to drop, so a valid way to upgrade
513 it is to drop it. That is not very user friendly and a bit more effort is
514 expected, but no promises are made.
516 .. _copyright-license-patents:
518 Copyright, License, and Patents
519 ===============================
523 This section deals with legal matters but does not provide legal advice. We
524 are not lawyers --- please seek legal counsel from an attorney.
526 This section addresses the issues of copyright, license and patents for the LLVM
527 project. The copyright for the code is held by the individual contributors of
528 the code and the terms of its license to LLVM users and developers is the
529 `University of Illinois/NCSA Open Source License
530 <http://www.opensource.org/licenses/UoI-NCSA.php>`_ (with portions dual licensed
531 under the `MIT License <http://www.opensource.org/licenses/mit-license.php>`_,
532 see below). As contributor to the LLVM project, you agree to allow any
533 contributions to the project to licensed under these terms.
538 The LLVM project does not require copyright assignments, which means that the
539 copyright for the code in the project is held by its respective contributors who
540 have each agreed to release their contributed code under the terms of the `LLVM
543 An implication of this is that the LLVM license is unlikely to ever change:
544 changing it would require tracking down all the contributors to LLVM and getting
545 them to agree that a license change is acceptable for their contribution. Since
546 there are no plans to change the license, this is not a cause for concern.
548 As a contributor to the project, this means that you (or your company) retain
549 ownership of the code you contribute, that it cannot be used in a way that
550 contradicts the license (which is a liberal BSD-style license), and that the
551 license for your contributions won't change without your approval in the
559 We intend to keep LLVM perpetually open source and to use a liberal open source
560 license. **As a contributor to the project, you agree that any contributions be
561 licensed under the terms of the corresponding subproject.** All of the code in
562 LLVM is available under the `University of Illinois/NCSA Open Source License
563 <http://www.opensource.org/licenses/UoI-NCSA.php>`_, which boils down to
566 * You can freely distribute LLVM.
567 * You must retain the copyright notice if you redistribute LLVM.
568 * Binaries derived from LLVM must reproduce the copyright notice (e.g. in an
569 included readme file).
570 * You can't use our names to promote your LLVM derived products.
571 * There's no warranty on LLVM at all.
573 We believe this fosters the widest adoption of LLVM because it **allows
574 commercial products to be derived from LLVM** with few restrictions and without
575 a requirement for making any derived works also open source (i.e. LLVM's
576 license is not a "copyleft" license like the GPL). We suggest that you read the
577 `License <http://www.opensource.org/licenses/UoI-NCSA.php>`_ if further
578 clarification is needed.
580 In addition to the UIUC license, the runtime library components of LLVM
581 (**compiler_rt, libc++, and libclc**) are also licensed under the `MIT License
582 <http://www.opensource.org/licenses/mit-license.php>`_, which does not contain
583 the binary redistribution clause. As a user of these runtime libraries, it
584 means that you can choose to use the code under either license (and thus don't
585 need the binary redistribution clause), and as a contributor to the code that
586 you agree that any contributions to these libraries be licensed under both
587 licenses. We feel that this is important for runtime libraries, because they
588 are implicitly linked into applications and therefore should not subject those
589 applications to the binary redistribution clause. This also means that it is ok
590 to move code from (e.g.) libc++ to the LLVM core without concern, but that code
591 cannot be moved from the LLVM core to libc++ without the copyright owner's
594 Note that the LLVM Project does distribute dragonegg, **which is
595 GPL.** This means that anything "linked" into dragonegg must itself be compatible
596 with the GPL, and must be releasable under the terms of the GPL. This implies
597 that **any code linked into dragonegg and distributed to others may be subject to
598 the viral aspects of the GPL** (for example, a proprietary code generator linked
599 into dragonegg must be made available under the GPL). This is not a problem for
600 code already distributed under a more liberal license (like the UIUC license),
601 and GPL-containing subprojects are kept in separate SVN repositories whose
602 LICENSE.txt files specifically indicate that they contain GPL code.
604 We have no plans to change the license of LLVM. If you have questions or
605 comments about the license, please contact the `LLVM Developer's Mailing
606 List <mailto:llvmdev@cs.uiuc.edu>`_.
611 To the best of our knowledge, LLVM does not infringe on any patents (we have
612 actually removed code from LLVM in the past that was found to infringe). Having
613 code in LLVM that infringes on patents would violate an important goal of the
614 project by making it hard or impossible to reuse the code for arbitrary purposes
615 (including commercial use).
617 When contributing code, we expect contributors to notify us of any potential for
618 patent-related trouble with their changes (including from third parties). If
619 you or your employer own the rights to a patent and would like to contribute
620 code to LLVM that relies on it, we require that the copyright owner sign an
621 agreement that allows any other user of LLVM to freely use your patent. Please
622 contact the `oversight group <mailto:llvm-oversight@cs.uiuc.edu>`_ for more