[clangd] Add clangd 18 release notes

[HighCommander4]

No description provided.

[llvmbot]

@llvm/pr-subscribers-clang-tools-extra

Author: Nathan Ridge (HighCommander4)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/84436.diff

1 Files Affected:

• (modified) clang-tools-extra/docs/ReleaseNotes.rst (+31-2)

diff --git a/clang-tools-extra/docs/ReleaseNotes.rst b/clang-tools-extra/docs/ReleaseNotes.rst
index 5758b5acbc0b56..d963642b812fbf 100644
--- a/clang-tools-extra/docs/ReleaseNotes.rst
+++ b/clang-tools-extra/docs/ReleaseNotes.rst
@@ -51,21 +51,40 @@ Improvements to clangd
 Inlay hints
 ^^^^^^^^^^^
 
+- Type hints
+    * Improved heuristics for showing sugared vs. desguared types
+    * Some hints which provide no information (e.g. ``<dependent-type>``) are now omitted
+- Parameter hints
+    * Parameter hints are now shown for calls through function pointers
+    * Parameter hints are now shown for calls to a class's ``operator()``
+    * No longer show bogus parameter hints for some builtins like ``__builtin_dump_struct``
+
 Diagnostics
 ^^^^^^^^^^^
 
-Semantic Highlighting
-^^^^^^^^^^^^^^^^^^^^^
+- Improved quality of include-cleaner diagnostics (missing and unused includes)
 
 Compile flags
 ^^^^^^^^^^^^^
 
+- System include extractor (``--query-driver``) improvements
+    * The directory containing builtin headers is now excluded from extracted system includes
+    * Various flags which can affect the system includes (``--target``, ``--stdlib``, ``-specs``) are now forwarded to the driver
+    * Fixed a bug where clangd would sometimes try to call a driver that didn't have obj-c support with ``-x objective-c++-header``
+    * The driver path is now dot-normalized before being compared to the ``--query-driver`` pattern
+    * ``--query-driver`` is now supported by ``clangd-indexer``
+- Fixed a regression in clangd 17 where response files would not be expanded
+
 Hover
 ^^^^^
 
+- Hover now shows alignment info for fields and records
+
 Code completion
 ^^^^^^^^^^^^^^^
 
+- Refined heuristics for determining whether the use of a function can be a call or not
+
 Code actions
 ^^^^^^^^^^^^
 
@@ -75,15 +94,25 @@ Code actions
 Signature help
 ^^^^^^^^^^^^^^
 
+- Improved support for calls through function pointer types
+
 Cross-references
 ^^^^^^^^^^^^^^^^
 
+- Improved support for C++20 concepts
+- Find-references now works for labels
+- Improvements to template heuristics
+
 Objective-C
 ^^^^^^^^^^^
 
 Miscellaneous
 ^^^^^^^^^^^^^
 
+- Various stability improvements, e.g. crash fixes
+- Improved error recovery on invalid code
+- Clangd now bails gracefully on assembly and IR source files
+
 Improvements to clang-doc
 -------------------------
 

[HighCommander4]

To prepare these, I went through the clangd commits in clangd 18 and made note of what seemed notable to me.

In some areas of the code that I'm less familiar with, I was somewhat vague (e.g. under "Diagnostics" I added "Improved quality of include-cleaner diagnostics (missing and unused includes)"; if someone more familiar with include-cleaner would like to expand on that with more specific improvements, that would be welcome).

[kadircet]

thanks a lot for doing this @HighCommander4 !

[kadircet]

i'd move this one into include-cleaner release notes (I am also happy to do that myself), a more detailed version could look like:

• Support for --only-headers flag to limit analysis to headers matching a regex
• Recognizes references through concepts
• Builtin headers are not analyzed
• Handling of references through friend declarations
• Fixes around handling of IWYU pragmas on stdlib headers
• Improved handling around references to/from template specializations

[HighCommander4]

Thanks, I added these in an include-cleaner section

[HighCommander4]

@kadircet do you know what is the process for getting this approved and merged? As this is targeting the release/18.x branch, I do not have permissions to merge it myself like a regular PR.

[kadircet]

i believe @tstellar should have that power 🚀

Do you mind merging this into the branch, or advise on how we should do that instead?