LLVM 19 will be released. As usual, I maintain lld/ELF and have addedsome notes to https://github.com/llvm/llvm-project/blob/release/19.x/lld/docs/ReleaseNotes.rst.I've meticulously reviewed nearly all the patches that are not authoredby me. I'll delve into some of the key changes.

• EI_OSABI in the output is now inferred from inputobject files. (#97144)
• --compress-sections <section-glib>={none,zlib,zstd}[:level]is added to compress matched output sections without theSHF_ALLOC flag. (#84855) (#90567)
• The default compression level for zlib is now independent of linkeroptimization level (Z_BEST_SPEED).
• zstd compression parallelism no longer requiresZSTD_MULITHREAD build.
• GNU_PROPERTY_AARCH64_FEATURE_PAUTH notes,R_AARCH64_AUTH_ABS64 andR_AARCH64_AUTH_RELATIVE relocations are now supported. (#72714)
• --no-allow-shlib-undefined now rejects non-exporteddefinitions in the def-hidden.so ref.so case. (#86777)
• --debug-names is added to create a merged.debug_names index from input .debug_namessections. Type units are not handled yet. (#86508)
• --enable-non-contiguous-regions option allowsautomatically packing input sections into memory regions byautomatically spilling to later matches if a region would overflow. Thisreduces the toil of manually packing regions (typical for embedded). Italso makes full LTO feasible in such cases, since IR merging currentlyprevents the linker script from referring to input files. (#90007)
• --default-script/-dT is implemented tospecify a default script that is processed if--script/-T is not specified. (#89327)
• --force-group-allocation is implemented to discardSHT_GROUP sections and combine relocation sections if theirrelocated section group members are placed to the same output section.(#94704)
• --build-id now defaults to generating a 20-byte digest("sha1") instead of 8-byte ("fast"). This improves compatibility withRPM packaging tools. (#93943)
• -z lrodata-after-bss is implemented to place.lrodata after .bss. (#81224)
• --export-dynamic no longer creates dynamic sections for-no-pie static linking.
• --lto-emit-asm is now added as the canonical spellingof --plugin-opt=emit-llvm.
• --lto-emit-llvm now uses the pre-codegen module. (#97480)
• When AArch64 PAuth is enabled, -z pack-relative-relocsnow encodes R_AARCH64_AUTH_RELATIVE relocations in.rela.auth.dyn. (#96496)
• -z gcs and -z gcs-report are now supportedfor AArch64 Guarded Control Stack extension.
• -r now forces -Bstatic.
• Thumb2 PLT is now supported for Cortex-M processors. (#93644)
• DW_EH_sdata4 of addresses larger than 0x80000000 is nowsupported for MIPS32. (#92438)
• Certain unknown section types are rejected. (#85173)
• PROVIDE(lhs = rhs) PROVIDE(rhs = ...), lhsis now defined only if rhs is needed. (#74771) (#87530)
• OUTPUT_FORMAT(binary) is now supported. (#98837)
• NOCROSSREFS and NOCRFOSSREFS_TO commandsnow supported to prohibit cross references between certain outputsections. (#98773)
• Orphan placement is refined to prefer the last similar section whenits rank <= orphan's rank. (#94099)Non-alloc orphan sections are now placed at the end. (#94519)
• R_X86_64_REX_GOTPCRELX of the addq form is no longer incorrectlyoptimized when the address is larger than 0x80000000.

--compress-sections

The --compress-sections option has been enhanced. Youcan choose between zlib and zstd for compression, along with specifyingthe desired compression level. Looking ahead, zlib is deprecated infavor of zstd. While zstd offers additional tuning options, we onlyprovide the compression level.

My Compressedarbitrary sections has analyzed potential use cases.

Orphan sections

Check out Understandingorphan sections.

Linker scripts

There are quite a few enhancements to the linker script support.NOCROSSREFS and--enable-non-contiguous-regions are noteworthy newfeatures. There is now an increasing demand of features for embeddedprogramming.

The world of embedded programming is a fascinating mix of open andclosed ecosystems. Developers of proprietary hardware and closed-sourcesoftware are increasingly interested in migrating their toolchains tothe LLVM Linker (LLD). The allure of faster link speeds, a cleancodebase, and seamless LTO integration is undeniable. However, as LLD'smaintainer, I must tread carefully. While accommodating these users isnice for LLD's growth, incorporating custom linker extensions riskscompromising the project's code quality and maintainability. Strikingthe right balance between flexibility and code integrity is essential toensure LLD remains a robust and efficient linker for a wide range ofusers.

GNU ld also supports extensions for embedded programming. Icategorize these extensions into two groups: mature and experimental.Many of the established extensions exhibit well-defined semantics andhave been incorporated into LLD. However, some newer extensions in GNUld appear less thoughtfully designed and inflexible.

When considering a specific extension, we should prioritize practicalneeds over arbitrary adherence to GNU ld's implementation. If compellingreasons justify a particular feature and GNU ld's approach provesrestrictive, we should feel empowered to innovate within LLD.

Conversely, when developing new extensions, it's essential to engagewith the broader community. I often submit feature requests to GNU ld toinform decisions we are going to make. I believe this collaborativeapproach fosters knowledge sharing.

CREL

I've developed CREL (compact relocations) to reduce relocatable filetremendously for LLVM 19. LLD now supports CREL with explicit addends.See Integratedassembler improvements in LLVM 19 for details.

Link: lld 18 ELFchanges