Historical Clang Language WG Meeting Minutes (Mar 2022-Mar 2023)

Clang Frontend
1

Meeting Agenda March 1, 2023

Agenda

  • No meeting March 15
  • Perennial/Plumhall status - haven’t got any response back from Plumhall. Have talked to Perennial, but initial feedback is that the license is not set up for opensource use.
  • Status of other floating point formats, specifically DFP. Fair amount of design work and some done internally. It’s not really ready to share yet. Ariel: would like to discuss. Tom: would be happy to share current status and possibly collaborate.
  • Undefined behaviour: have got a number of bug reports recently. One about unreachable, other about indeterminate values. May be related to changes and adoption of LLVM 15, may also be more awareness. Hubert: regarding indeterminate behaviour, it may be C specific. Changes in LLVM 15 have definite impact. Aaron: C committee working on this. Committee needs to engage with LLVM community. Shafik: does this warrant a Discourse discussion? Aaron/Hubert: yes, but want to make sure that discussion is not had in multiple venues. Discussion also needs to be documented. Shafik to kick off the discussion, likely on the side for now.

Round Table

  • Aaron
    • Largely working on code reviews
    • Working on Clang 16 release work, tagging bugs. Please make sure that potential release blockers are identified
  • Ariel
    • Locale issue, some work relating to floating point
  • Corentin
    • Waiting on review from Shafik on some changes that were reverted
    • Looking at start_lifetime_as issue, thinks it needs a new intrinsic
      • Probably need to talk to Richard
    • PR for deducing this, likely targeting Clang 17
  • Fahad
    • Implementing intrinsic for new type traits std::is_layout_compatible
    • std::is_corresponding_member, hoping to land PR later this week
    • Any discussion on extending standard testing?
      • Nothing specific, but lots of discussion
  • Hubert
    • Responded to one of the issues that Shafik raised
  • Luke
    • C++20 compiles are ~10-20% slower, likely const_eval and template stuff happening. Some libc++ changes also have impact. Profiling and further investigation necessary.
      • Have we seen any runtime improvements? None identified thus far. Should investigate PGO compilers with C++20 mode turned on.
      • Corentin: anyone working on the const_expr interpreter? Yes for quite some time from Tim, but it’s a huge project. Some help in reviewing the PRs would be helpful.
    • Tracking down a crash in modules
      • Tom: tried to reproduce with preprocessed source?
        • Luke: able to reproduce with a debug
        • Hubert: possibly a serialization problem?
        • Luke: Decl data structure is malformed
        • Reach out to Chuanqi Xu (ChuanqiXu), he’s investigated something similar.
  • Shafik
    • Bug screening, trying to catch up on code reviews.
  • Tom
    • Plan to turn to DFP
  • Tony
    • C23 auto keyword. Reverting patch, treating auto as a type specifier. Some tests to fix to reflect the standard.

Meeting Agenda February 15, 2023

Agenda

  • Feedback from Perennial - their license model doesn’t allow for us to put the tests into version control and so we’re not able to find a way for arbitrary developers to see test failures after a change to the compiler.
  • GSOC proposals - looking for mentors on each project. Can reach out to Jeff on Discourse.

  1. Improve C and C++ language conformance by triaging and fixing DRs

  2. Issue here is that advanced C++ knowledge would be necessary, and would also require a mentor who has advanced understanding of the standard.

  3. Improve C and C++ language conformance by creating new conformance test cases

  4. Issue here is that advanced C++ knowledge would be necessary, and would also require a mentor who has advanced understanding of the standard,

  5. Focus on newer features-concepts (maybe secondary), modules and coroutines. Template engine.

  6. Often find crashes on invalid code. This may be indicative that we’re missing test coverage. Potential issues with error recovery. Finishing recovery expressions would be helpful, its an a half baked state right now. Double check with Aaron on some background.

Round Table

Meeting Agenda February 1, 2023

Agenda

  • Feedback from perennial/plumhall - no response. Issue is how do we get access to failures and get reproducers? Expectation is that the test cases are run on a non-visible server, and test case source is not made available.
    • How do we ensure conformance? Usually a combination of test cases developed, existing tests and customer bugs
    • Have we validated that unit testing is insufficient?
    • One option is to set up a separate repository to accumulate test cases from elsewhere, ex. leveraging gcc test suite. Would need to be careful of GPL contamination.
    • Is there interest in other test cases other than conformance, ex performance. We have some performance tracking already https://llvm-compile-time-tracker.com/
    • Possibility of adding larger code bases for testing purposes? Could set up separate builders using clang trunk. Likely orthogonal.
    • Action: Aaron to see if he can reengage with perennial/plumhall
    • Suggestion: possibility of having perennial/plumhall to open/analyze failures
    • Once pricing information is gathered, can approach LLVM foundation
    • May want to have an action on end-to-end testing
  • Branch split - anything to note?
    • Complete last week, seems stable. Feel free to start dropping experimental items back to trunk. Although most feedback usually comes in during RC1, which is still pending.

Roundtable

  • Aaron
    • C standards meetings, catching up on reviews
    • C++ meetings next week
    • Expect code review velocity to be a bit slow
    • C++23 seems on track. C has received many comments, likely needed for next CD. C23 is a bit at risk for end of year.
  • Ariel
    • PR for a PTR32 issue, fix re OpenCL builtins rewriting prototype for parameters passed. May want OpenCL to review.
  • Fahad
    • Working on a Concept issue, having parentheses on constraints in atomic clause. Waiting on review comments. To reach out to Hubert for clarification.
    • C++ 20 demand? Aaron answers: concepts, const expr, modules. Seen use cases of ranges, need to validate exactly which applications. Some bug reports open.
    • C++ 23? Aaron answers: still early
  • Hubert
    • C++ committee meeting next week
    • Tracking some of the C committee traffic
    • Small fix for vaopt from Shafik, still need to review
  • Luke
    • Some reprioritization on Google projects
  • Tony

Meeting Agenda January 18, 2023

Agenda

  • We’re branching next week on Tue Jan 24th! Try to get things you want in the Clang 16 in before the branch date, but please do not push things that are not believed to be rock solid (we’re running out of time for speculation) as those can wait for Clang 17.
  • Big thanks to folks who’ve been helping get things finished up for the release, like Chuanqi’s help getting the coroutines and modules TS command line flags deprecated so we can remove them in Clang 17

Round Table

  • Ariel - PTR32 change going through internal review, will take another few days
    • Question for Aaron, did he follow up with Plumhall/Perennial? He did send out an email, not sure if he got a response (but probably not yet). Will follow up next time.
  • Erich
    • Lambda deferred instantiation - still outstanding
    • Working on branch prep, catching up on reviews
    • working through some of the regressions for the sugaring patches, reverting them took some time to work through
  • Fahad
    • Still working on the issues mentioned previously
    • Looking at next work items as well, feedback appreciated
  • Hubert
    • Evaluating patch that changed ABI for default/deleted functions for POD, wanted to understand why people opted out of this change for Darwin. Change was for gcc compatibility.
      • Gcc is not of major concern on Darwin, likely why they opted out. Clang is assumed the default ABI on those platforms.
      • Is there concern about risk, or was there lack of motivation? Usually Darwin doesn’t break ABI simply for compatibility with other compilers, since Clang is the assumed de facto.
  • James
  • Luke
    • Did push patch for C++20 crash, up for review
  • Shafik
  • Tom
    • Plan to push change to add diagnostic for lambdas

Meeting Agenda January 4, 2023

Agenda

  • Previous proposal for test suite checking items in the standard
    • Some work for C std DRs and features
    • Is there some work to get a license for Plumhall/Perennial? Alternatively should we be developing a test suite ourselves
    • AR: Aaron to send an email to Plumhall/Perennial to see about the possibility of supporting open source projects

Round Table

auto b = { 1, };

auto d = { 1 };

  • Hubert
    • C++23 papers and issues status, Corentin had updated (thanks!)
    • Reviewed “Lifetime extension in range-based for loops”: https://reviews.llvm.org/D139586
      • Corentin: should be easier to implement after the source location fix [can we add a comment to the review to this effect]

Meeting Agenda Nov 16, 2022

Agenda

  • Will cancel the December meetings, restart in January
  • Hubert: const eval propagation and range for. How to add those?
    • Should show up on the cxx standard pages. Not sure if those are done for papers. May need to be done manually.
    • Tom: should we add github issues
    • Do we have a C++23 tracker? Not certain if this is done automatically. C++20 tracker C++ 20 in Clang · GitHub . Hubert/Aaron to see about adding C++23.

Round Table

Meeting Agenda Nov 2, 2022

Agenda

  • Is there a value to having a standard-ese mailing list to discuss issues
    • Discourse - create a private topic?
    • WG21 chat service?

Round Table

Meeting Agenda Oct 19, 2022

Agenda

  • Safe C++ RFC
    • Aaron got some feedback from the Apple developers working on this. There’s a lack of a design, should be less ambiguous once that’s available.
    • How are we going to deal with -Weverything warnings as a result?

Round Table

  • Aaron
    • Pushed changes for nullptr to C2x, pending direction from committee
    • Storage durations for compound literals (C2x), existing model doesn’t work too well, seeing if we can rewrite the way we do it in C. A bit experimental at this point, may just do the parsing if it goes poorly.
      • There may be a gnu extension that this proposal may collide
  • Ariel
  • Erich
    • Working through deferred concept instantiation regressions, the way we model is insufficient. Been a lot of work trying to come up with a better way to model.
  • Hubert
    • Been involved in a few reviews,
      • One for thread local guard
      • Applications for DRs, whether they should be turned on by default. One was for preemptively option control, to remove
    • Working on -p option which Clang doesn’t really implement, looking for comments on RFC
  • Reid
  • Tom
    • Pushed update for MS asm blocks
    • Bug for bitfields in gnu, but not planning on looking at this one
    • Thread local guard, agree that’s reasonable. Should respond by end of day.
  • Tony
    • Working on parsing rework for auto keyword patch for C23, mostly done. An extra attribute on AST.

Meeting Agenda Oct 5, 2022

Agenda

  • Note that build bots are failing, is anyone taking a look into it?
    • Some were failing due to C++17 changeover, but otherwise green

Round Table

  • Aaron
    • Nullptr review for C23, some differences from C++, still some chance that they implementation may change as WG14 locks down the specification
    • Most differences from C++ seem to be edge cases, so likely won’t be hit too often
    • Need to get back to Automatic type deduction proposal, hope that through review will be able to identify/resolve any discrepancies between C and C++
      • Would be good to post code samples as discrepancies are found
    • typeof and typeof_unqual change landed, thanks Matheus for comments
    • Updating document on potential breaking changes ⚙ D134878 Update developer policy on potentially breaking changes, please update documentation and note in the announcement channel if you have any breaking changes, focus on DR fixes - please include Aaron and Hubert so they can report back to WG21 in case breaking changes are found during implementation
  • Chaofan
  • Corentin
  • Erich
    • deferred concepts implementation, mostly caught up on bugs, triaging the rest
    • trying to catch up on reviews, please ping if needed
    • bugs on parsing require and lookup
    • RFC on stopping allowing to dereference void *, we seem to be the only implementation that does so RFC: Can we stop the extension to allow dereferencing void* in C++?
      • To add as a warning, with intention of switching to error down the line
  • Luke
  • Matheus
    • Rebased the resugaring work on top of concepts patch
    • Fixed all known bugs
    • One more patch in review state, need to respond. 5 other enablers in queue, but will likely be a bit unavailable over the next little while
  • Tom
    • Pushed change for improving support for immediately invoked lambdas in default arguments, please report any issues found
    • Some other small fixes, will move back to DFP support, Shafik has been making good progress here
    • Discovered that there are default arguments in blocks that is never used, suggest to create a github issue to track further
    • Bitfields that appear in MS asm blocks don’t make a lot of sense, may look into this. MS compiler adds a diagnostic, we want to create a similar diagnostic. Do we want to create an option to disable for backwards compiler?
      • Seems to only be in place to support MS headers, we may want to disable bitfields in MS asm blocks and add an error. We can respond to feedback afterwards, likely state that there is no good semantic
  • Tony
    • Not a lot of progress on the auto keyword. Working on patching on the parsing side. Trying to generate an auto type by default.
  • Utkarsh

Code Review

Meeting Agenda Sept 21, 2022

Agenda

Round Table

Code Reviews

Meeting Agenda Sept 7, 2022

Agenda

  • Congratulations to the new code owners! Thanks to Aaron for driving the code ownership changes!
    • Will similar code ownership changes be made in LLVM? Likely, but will expect the LLVM folks to drive change in the project.
  • Jeff to clean up the Google calendar
  • Bug dashboard: C++ 20 in Clang · GitHub
    • This is great! Once ready, we should share the existence of the dashboard on Discourse, and incorporate in to the main page once matured a bit.
    • Priorities are established with some rules (example crash on valid code higher than crash on invalid code), some will be resolved in future standard revisions
    • Use Github projects for establishing the dashboard
  • LLVM dev conference - Matheus will be attending

Round Table

  • Aaron
    • Mostly establishing code owners
    • Will move to C23 and CRs
    • Would like to get more folks focused on documentation, have pushed in recent versions for release notes. Recent release notes have been well received. Should focus on documenting implementation defined behavior. May formulate an RFC around this.
      • Is there some way to automate from pulling github pushes during a release? Have had some experience in the past (prior to github) and it didn’t quite work. It might be useful as a secondary check. But anyway that we can automate, would be good.
  • Ariel
    • Working on some DWARF issues
  • Chaofan
    • Looking at an old issue for a clash in Clang 41302, ambiguity in struct passing
  • Erich
    • Deferred concepts, fixing bugs that have been reported (by Facebook)
    • Catching up on reviews
  • Ilya
  • Matheus
    • Responding to many reviews
    • Working on splitting the resugaring patch in to more manageable chunks
    • Duplicate bugs reported for builtin templates, should have a fix that will resolve several of these.
    • Several loose threads to resolve, documentation to work on
  • Tony
  • Utkarsh
    • Recently started working on C++20
    • Consteval bugs and crashes, ~5-10 bugs
    • Consteval crash in codegen
    • Will move over to Concepts bugs

Code Reviews

Meeting Agenda Aug 17, 2022

Agenda

  • Matheus - bugs in Phabricator
    • Erich reviewed patch for desurgaring, split in to two
    • Asking for help in reviews ⚙ D131926 [clang-tidy] Fix for bugprone-sizeof-expression PR57167
    • Bugs exposed from desurgaring, one from checker from large number of cases exposed
    • Erich: would be good to cut down the patches into smaller patches
    • Matheus is progress from cutting down the patches
    • Ariel: how much has been committed? Matheus: elaborate type has been committed, some others still in progress. May have impacted the PTR32 work.

Round Table

Aaron - const eval (thanks to Ilya and Utkarsh for help)

  • C feature reviewing, set up directory structure for new features. Please ask for clarification if needed. Leverage paper numbers
  • Tom: what about similar C and C++ features? Aaron: probably makes sense to have separate tests

Ariel - ptr32 stuff complete, now focusing on DWARF

Chaofan

Corentin - update the DR page for C++

  • Structure binding capture was merged last week

Erich - review and work on libc++ tests, follow up bugs where Concepts isn’t checked right

Guillot Tony - a bit of C23 work, auto keyword

Ilya - looked at const eval, closed several issues on Github

Reid

Shafik - dealing with fallout from enum diagnostics, affected Boost and QT, turned into a warning. Should resolve most things, but will be temporary.

  • Aaron: -w-noerror flag led to some unexpected issues. Warnings as errors often get upgraded to errors in later releases, should serve as an early warning to migrate code

Shivam

Tom H - still wrapping up lambda expressions in default arguments, getting close

  • Will look in to char8_t support up next
  • Shafik and Tom to focus on DFP support soon
    • Aaron: will it be extended to C++? Tom: eventually but not initially
    • Ariel: library support? Tom: will look into later, focus on core language for gcc C library and LLVM C library.
    • Hope to have some contributions early next year
    • Initial focus on Intel architecture, but additional support for zOS and PPC will be needed, will collaborate with IBM
    • Any LLVM IR impacts? Tom: hope to lower to existing types

Code Reviews

Meeting Agenda Aug 2, 2022

Agenda

  • GSOC project - resugaring templates
    • Proposal to add a pass when access template specialization, improving diagnostics
    • Compiler explorer experimental implementation, please try and provide feedback Compiler Explorer
    • Patch is getting pretty big, depending on Richard for reviews, need some help here for feedback
    • Working on a demo
    • Aaron: are we always going to get diagnostics? Matheus: may put behind a flag
    • Template instantiation is slow, question is how do we limit analysis to limit impact on compile time. Memory consumption is also an issue. Recommendation is that we might want to do some performance analysis. Resource for measuring compile time http://llvm-compile-time-tracker.com/
    • Todo Jeff: add the group alias so Matheus can use

Round Table

(Chaofan) I’m on vacation today so I can’t join this meeting.

Code Review

Meeting Agenda June 15, 2022

Agenda

Round Table

  • Aaron - code reviews
    • Expect to get back to const eval after getting through CR backlog
    • Working on the C standard DR page
  • Ariel
  • Bob
  • Chaofan - continue to confirm bugs
  • Erich - deferred concepts implementation, Chuanxi has done a lot of reviews (thanks!) and run on other platforms.
    • Hoping to fix reported bug this week
  • Hubert
  • Ilya - first time, goal to help with C++20
    • Mostly looking at bugs
    • Fixed assertion, sent for review
    • Recommendation: triaging bugs above
    • Question: Is looking at C++ DRs also useful? Yes
    • Question: how to indicate ownership? A: either self assigning issue, or creating RFC on Discourse
    • Should we create issues for all DRs or use the bug tracker page to indicate? A: Either can be done, but traditionally bug tracker has been the source of truth
      • For issues that have a partial tag, it would be preferable to add details on the status page about what is missing
  • James
  • Krzysztof
  • Nathan
  • Shivam -first time
  • Nick - first time, interested in helping with C related issues
    • How to help? A: either looking at C23 features, C DRs fixed, triaging bugs
    • https://clang.llvm.org/c_status.html
    • Add clang-language-wg reviewer if you want the group to be aware of any PRs
  • Jeff

Code Reviews

Meeting Agenda June 1, 2022

Agenda

Round table

  • Jeff - need to review preexisting action items
  • Aaron - currently working on a regression, losing information on AST re qualifiers
    • Started work on page to track C defect reports, similar to C++ status page. C DRs are not as automated https://clang.llvm.org/c_dr_status.html. C89 DRs are going quite quickly, but expecting the more complex ones will slow down.
    • Question: anything new from the last C standards meeting? Not a lot was added. C committee has started tracking papers for implementers that have different language modes. https://www9.open-std.org/jtc1/sc22/wg14/www/previous.html
    • [From chat] At the last meeting, we adopted: N2886 N2888 N2861 N2992 N2897
  • Ariel
  • Chaofan - took a look at Github, there are 277 issues tagged with C++20, most are migrated from bugzilla. Do we need to triage? [Aaron] signed up for new bugs, which are triaged. Old ones are neglected though. Will park for future meeting. https://github.com/llvm/llvm-project/issues?q=is%3Aopen+label%3Ac%2B%2B20+no%3Aassignee
  • Erich - making progress on deferred concepts, got some feedback from Richard that was very useful. Now working on a mangling change. Pointed out that Modules test is required as well, would be nice to have some help. Tom mentioned that he can help with constructing a test.
  • Hubert - haven’t had a chance to review the DFP RFC as of yet. Not sure about scope of the RFC, currently highlights the core language steps
  • Joe
  • Nathan
  • Bob
  • Tom - dealing with issue relating to instantiation of default arguments https://github.com/llvm/llvm-project/issues/49178 . Struggling with the assertion errors. Not familiar with the section of code, but concerned that there is some steps out of order. Please ping Tom if anyone knows this area of code.
  • James - const eval source location bug is a blocker. Also taking a look at coroutines. Thread local variable CR merging across coroutine thread points. Difficult to evaluate the change. Current intent to prototype the small change. Local static in thread local is pretty messy right now due to OMP/objc blocks, evaluating if it can be cleaned up. [Tom] does anyone know if the objc blocks are buggy? Running in to odd errors.

Code Reviews

Meeting Agenda May 4, 2022

Agenda

  • Clang release notes

  • Phabricator group

    • clang-language-wg
    • Please let me or Bruno know if you want to be added to the group

  • STDC macro behavior in MS compatibility mode

    • User report that was surprised what STDC expands to
    • Expectation that when macro is set, you specify which version you want to conform to
    • MS compatibility mode doesn’t set the standard macro, inconsistent
    • Nathan: did MSVC not used to set, and now does?
    • There are many conformance flags, may need to see which need to be set

  • Round table

    • Jeff
      • Clang WG tag in discourse
      • Test case licensing issues
    • Aaron
      • Continuing to work on C compatibility, landed some changes on implicit int handling, error in various C modes. Test suites now need to be updated.
      • Some bug triage on C bugs
    • Chaofan Qiu
    • Hubert
      • DFP RFC, newer draft posted by Tom. needs to be looked at
    • Jan
    • Krzsztof
    • Nathan
      • Demangler, patch on unnamed enumerations in header units
      • Array conversion, have taken ownership of bug #54831
    • Steven

  • Retrospective

    • Live code reviews would be good, if people had suggestions ahead of time
    • Call for agenda items ahead of time, also propose agenda

  • Code Reviews

Meeting Agenda April 6, 2022

Agenda

  • Phabricator group created

  • Reproducers from commercial test suites

    • Licensing terms associated with test suites benchmarks, issues with creating reproducers
      • Contractual considerations in addition to copyright
    • Proposal: short reproducers would likely not be problematic
    • Probably need a legal opinion
    • Can it be rewritten in a way that removes it from the context of the source material
    • Can Creduce significantly anonymize the source code
    • Proposal: Reduce code in a way that the original essence is no longer present
    • Sources typically from benchmarks and conformance suites
    • Unit tests and gcc test suites used in Clang community
      • Potentially donation or licensing of older test suites?
    • Proposal: describing the issue and having someone else to create the reproducer. A clean reproducer. If we can point to the section of the standard that is problematic, can someone else write the reproducer.

  • Round table

    • Jeff Heath
      • Creation of ‘C/C++ WG’ subcategory
    • Ariel Burton
    • Chaofan Qiu
      • Summarized defects listed as unknown, but it’s a huge list (~1500)
      • Checked gcc defect report in GCC, even larger (~2000)
    • Chuangqi Xu
      • Initial support for reachable definition is waiting for review: D113545. Our internal experiment shows that we could use C++20 modules with some other bug fix after D113545 landed.
    • Corentin
    • Erich Keane
      • Deferred concepts to allow compiling of ranges, PR is ready for review. Pretty complex patch, may need to accept at some point.
    • Hubert Tong
      • Reviewed patch from Corentin
    • Jan Korous
    • Jan Svoboda
    • Krzysztof P
    • Michael Spencer
    • Ian
    • Shraiysh Vaishay
    • Vlad
      • Closed CWG2354
      • Working on P1787 Declarations and where to find them
    • Aaron Ballman

  • Code reviews

  • Bug triaging

  • Question - is there a recommended way to be able to decipher the C++ standard?

    • Typically discussed amongst committee members, but that is a closed discussion
    • Draft is available in github
    • Can also ask the group here, and can forward to committee as requested
    • Finding a paper where wording came from could be helpful
      • git blame on https://github.com/cplusplus/draft ?
      • if it came before github, working drafts are publicly available since 2005
      • if you’re unlucky to encounter the wording in C++98, searching through papers becomes very time-consuming, but still feasible if you have a good idea what you’re looking for

  • Contact information - Jeff to aggregate and then move to a protected document

Attendees

Meeting Agenda March 2, 2022

Agenda

  • Introductions
    • Jeff Heath, Aaron Ballman, Bruno Cardoso Lopes, Chanqi Xu, Corentin, Erich Keane, Hubert Tong, Krzystof Parzyszek, Markus Böck, Tom Honerman, Steven (Yu) Wan, Vlad Serebrennikov, Chaofan Qiu, Arthur O’Dwyer, Bob Montelone