So happy to see this finally get written up. I appreciate you doing this! This will be very helpful for new contributors.

Much better name! Thanks.

Looks good - just left a comment on a minor typo. Thanks!

Excited to see this come together!

Thanks for the quick fix from the discussion on the issue today.

Thanks for adding this for others. This is the same script I use locally as well that I got from you ages ago.

Big +1 from me. Thanks for doing this.

I can't speak to the correctness of the implementation of findModulesAndOffsets here, but I'd love to see this make progress.

Rebase and mark the paper status as completed

Rebase since the clang-format requirement patch landed

In D143975#4126246, @sunshaoce wrote:

#libc is blocking this patch to make sure it's approved by the people working on libc++.

Thanks!

CMake
    target_add_compile_flags_if_supported(${target} PRIVATE
      -Wno-user-defined-literals
      )

It seems that this code does not work correctly. I think we can modify it to suppress the warning in the building stage.

Remove unnecessary constexpr_char_traits.h include

@Mordante do you want me to wait until you land https://reviews.llvm.org/D143845? Or, you can just feel free to mark LWG 3857 as Completed in LLVM 17.0.0 if I land this first — whatever you prefer.

Drop the static_assert tests that verified traits_type mismatches as they are no longer relevant

Undo extra clang-format in from_range.pass.cpp

Seems reasonable to me. We should close https://reviews.llvm.org/D125394 when/if we land your review here.

In D143007#4104804, @Mordante wrote:

Thanks for the reviews!

In D143007#4103951, @jloser wrote:

In D143007#4103560, @ldionne wrote:

Is LLVM_HEAD_VERSION=17 going to work yet?

I don't think so based on looking at the code:

declare -A LLVM_VERSION_PATTERNS
LLVM_VERSION_PATTERNS[9]="-9"
LLVM_VERSION_PATTERNS[10]="-10"
LLVM_VERSION_PATTERNS[11]="-11"
LLVM_VERSION_PATTERNS[12]="-12"
LLVM_VERSION_PATTERNS[13]="-13"
LLVM_VERSION_PATTERNS[14]="-14"
LLVM_VERSION_PATTERNS[15]="-15"
LLVM_VERSION_PATTERNS[16]="-16"
LLVM_VERSION_PATTERNS[17]=""

if [ ! ${LLVM_VERSION_PATTERNS[$LLVM_VERSION]+_} ]; then
    echo "This script does not support LLVM version $LLVM_VERSION"
    exit 3
fi

since LLVM_VERSION_PATTERNS[17] is an empty string.

AFAIK That part of the code is to make sure lines like

deb http://apt.llvm.org/bullseye/ llvm-toolchain-bullseye-15 main
deb http://apt.llvm.org/bullseye/ llvm-toolchain-bullseye main

work properly in sources.list files in Debian/Ubuntu. Note the the ToT version has no version suffix.

I have tested the Docker file locally with several Clang versions, including 17, before I created this review.

There is one fix queued to be backported to the LLVM-16 after that has landed I'll land this and upload a new Docker image.
(Without that fix the LLVM-16 branch modular build will fail.)

In D143007#4103560, @ldionne wrote:

Is LLVM_HEAD_VERSION=17 going to work yet?

LGTM!

In D143094#4098208, @LuoYuanke wrote:

LGTM, thanks

Update commit message to properly display the #define

Thanks for writing this up for others' benefit! I remember when you first showed me this a long time ago, I was really thankful for helping track down a CI failure.

Update LLVM_DEPRECATED messages to use MutableArrayRef

I landed this as the CI failures are unrelated

In D141327#4040331, @dblaikie wrote:

I'd be OK with the testing just being the static_asserts without the rest of the testing (since that's all the deduction guide do - specifies what type to use, then the usual ctor rules kick in - and those ctors are already tested elsewhere) - but this is OK too, if you prefer it.

Rebase in quest for the green CI

Remove deduction guide from MutableArrayRef: it's not needed. Add a test from both const and non-const MutableArrayRef to show there's no issues, or ambiguous deduction guides like before.

In D141345#4038479, @philnik wrote:

LGTM, but let's see if anybody else has any comments. We should definitely get better at working through the TODO LLVM <version>s.

Thanks for the fix! LGTM FWIW.

In D141327#4038061, @dblaikie wrote:

Test coverage might be nice - some static asserts about the type that's deduced is probably sufficient (static_assert(std::is_same_v<MutableArrayRef<int, 3>, MutableArrayRef(std::declval<int[3]>())>) or something like that for each deduction guide)?

Add tests for deduction guides and runtime behavior

In D140280#4005992, @dblaikie wrote:

Is there a way to write this in a more portable fashion?

(if the uint64_t on nthe RHS versions were non-members, would that help?)

Add [llvm] to title

LGTM, but I haven't been following as closely this release, so there may be other things I didn't notice in the updated release notes.

Use decltype(auto) in operator* in case the base type is const qualified

Rebase

@philnik mind if I land this? CI isn't green yet due to unrelated issues.

Rebase and use >= 20 instead of > 17

In D141239#4035827, @philnik wrote:

I've landed this for you, since this fixes the trunk CI.

Sorry for the delay here; this patch LGTM! Thanks.

Prettify commit message formatting

I'm +1 for the direction here; I took a brief look but will look closely either later tonight or tomorrow before I approve. Thanks for doing this!

LGTM with just removing the const. Please add a comment though as we don't have an existing regression test to show this (i.e. prevent someone from adding const back IIUC).

In D131847#3977994, @sebastian-ne wrote:

I put up D139532, which includes the work done here and finishes it (also replaces any_cast with std::any_cast, compiles and passes llvm and clang tests).